From: Andreas Tille Date: Mon, 16 Jan 2023 12:49:37 +0000 (+0000) Subject: Import r-cran-s2_1.1.2.orig.tar.gz X-Git-Tag: archive/raspbian/1.1.2-1+rpi1^2~4 X-Git-Url: https://dgit.raspbian.org/?a=commitdiff_plain;h=84a71639929a148bdc48933b8a5a26b45429a7e3;p=r-cran-s2.git Import r-cran-s2_1.1.2.orig.tar.gz [dgit import orig r-cran-s2_1.1.2.orig.tar.gz] --- 84a71639929a148bdc48933b8a5a26b45429a7e3 diff --git a/DESCRIPTION b/DESCRIPTION new file mode 100644 index 0000000..6ae7384 --- /dev/null +++ b/DESCRIPTION @@ -0,0 +1,46 @@ +Package: s2 +Title: Spherical Geometry Operators Using the S2 Geometry Library +Version: 1.1.2 +Authors@R: c( + person(given = "Dewey", + family = "Dunnington", + role = c("aut"), + email = "dewey@fishandwhistle.net", + comment = c(ORCID = "0000-0002-9415-4582")), + person(given = "Edzer", + family = "Pebesma", + role = c("aut", "cre"), + email = "edzer.pebesma@uni-muenster.de", + comment = c(ORCID = "0000-0001-8049-7069")), + person("Ege", "Rubak", email="rubak@math.aau.dk", role = c("aut")), + person("Jeroen", "Ooms", , "jeroen.ooms@stat.ucla.edu", role = "ctb", comment = "configure script"), + person(family = "Google, Inc.", role = "cph", comment = "Original s2geometry.io source code") + ) +Description: Provides R bindings for Google's s2 library for geometric calculations on + the sphere. High-performance constructors and exporters provide high compatibility + with existing spatial packages, transformers construct new geometries from existing + geometries, predicates provide a means to select geometries based on spatial + relationships, and accessors extract information about geometries. +License: Apache License (== 2.0) +Encoding: UTF-8 +LazyData: true +RoxygenNote: 7.2.3 +SystemRequirements: OpenSSL >= 1.0.1 +LinkingTo: Rcpp, wk +Imports: Rcpp, wk (>= 0.6.0) +Suggests: testthat (>= 3.0.0), vctrs +URL: https://r-spatial.github.io/s2/, https://github.com/r-spatial/s2, + https://s2geometry.io/ +BugReports: https://github.com/r-spatial/s2/issues +Depends: R (>= 3.0.0) +Config/testthat/edition: 3 +NeedsCompilation: yes +Packaged: 2023-01-11 14:23:27 UTC; edzer +Author: Dewey Dunnington [aut] (), + Edzer Pebesma [aut, cre] (), + Ege Rubak [aut], + Jeroen Ooms [ctb] (configure script), + Google, Inc. [cph] (Original s2geometry.io source code) +Maintainer: Edzer Pebesma +Repository: CRAN +Date/Publication: 2023-01-12 13:00:02 UTC diff --git a/MD5 b/MD5 new file mode 100644 index 0000000..4e57871 --- /dev/null +++ b/MD5 @@ -0,0 +1,682 @@ +559d2ae520e00ba566d56fe6ed456fbb *DESCRIPTION +8f49b889e6270ac988fc307b40230b44 *NAMESPACE +9c83f76b446cd82935fefcc104f1a7b6 *NEWS.md +6e30cf9b0b3a9e4629b7b684856b4b07 *R/RcppExports.R +26fbc9df4408ccb165ddbd0cd8577b7e *R/data.R +4ca766e1349ea7baf27d4c8638fcb62b *R/plot.R +b783a368a460bcae5d174c575d863e59 *R/s2-accessors.R +4ba8ec7d16efeadbc21c3ab5ac3dc6a6 *R/s2-bounds.R +9b93004c163961946245e2ce97d2f27f *R/s2-cell-union.R +aaa52035ceb2b49346994d2b1cde1493 *R/s2-cell.R +ed0cadf256cc791b85b392eef085c6f6 *R/s2-constructors-formatters.R +0968515e318a2c7a7372bddad1be6c23 *R/s2-earth.R +9cda6c070b015341cf92f57b9307a770 *R/s2-geography.R +cabc1e60f8ac15a3b463f4cd11143df9 *R/s2-lnglat.R +007cb436f26e501361a48b7782062eb6 *R/s2-matrix.R +e960477e2e4f889c85831e3d438bfe0f *R/s2-options.R +1bab90d590a0e66114c86739ba0d6ce4 *R/s2-package.R +994a3c25167362ce7175a580d75cc1f6 *R/s2-point.R +1618630f02d0c71a6884c2ef5359db7a *R/s2-predicates.R +e7dff1519bc26e311ae3e3f1bb894a69 *R/s2-transformers.R +a2bcb714226ffff00165255cd261dcc5 *R/utils.R +1b0b65e06a8b04d12df322094d6a16c2 *R/vctrs.R +542c90500e67cf705cba921a201f5a57 *R/wk-utils.R +fa791081760c033818970bdda18960bc *R/zzz.R +0d29d0b0030a40d9fb9632cf82c261c5 *README.md +ea124a494ad7922985532ad3b308e4d5 *build/partial.rdb +cc6eebf14a3f516ec74168f3ea416fb2 *cleanup +cbac38784589bac27de4468ecf03fb13 *configure +d41d8cd98f00b204e9800998ecf8427e *configure.win +885346a0f57b92fd80b41b8afac11bad *data/s2_data_example_wkt.rda +e59a2d1d70e5a35ca7a5d9bc17ff9554 *data/s2_data_tbl_cities.rda +044840e4280d77984dc42feb2c3f387c *data/s2_data_tbl_countries.rda +94d2fee6187d451e94b028677e579b00 *data/s2_data_tbl_timezones.rda +d41d8cd98f00b204e9800998ecf8427e *inst/extdata/emptyfile +743a5630af573f46d1916e4810419a77 *man/as_s2_geography.Rd +ed01c59b91c1a9c3382d1f2ab67e60cc *man/figures/rc300.png +4a651a96987fab04091987a9343eda03 *man/s2-package.Rd +52eef06fb0b570954600e8e619fe9756 *man/s2_boundary.Rd +a906d7f9956940ceebe0bd879c14e5f2 *man/s2_bounds_cap.Rd +ebcba3b3e746c4e15de4b309d0674c37 *man/s2_cell.Rd +74e8baaf2610d2ca7d578c95daf23937 *man/s2_cell_is_valid.Rd +79be4874bfb6397f7e4ae40b18df426f *man/s2_cell_union.Rd +71b26a6c7cc913b2dd369afcd744a8f1 *man/s2_cell_union_normalize.Rd +d2590d0319581da0f2b40f438374e5ea *man/s2_closest_feature.Rd +673de08c721dbf5fe295811ceb1a6bd6 *man/s2_contains.Rd +7c5dbf9f18b14b1fa00b61ecd5efeccf *man/s2_data_example_wkt.Rd +addd614791ada510339317f68202d348 *man/s2_data_tbl_countries.Rd +d7a1ee72a69458a577921e02c2f4142f *man/s2_earth_radius_meters.Rd +c56c83fe4daade0fd891802084cc54fc *man/s2_geog_point.Rd +9676c31cb07674f1d6935a5b0462f3ca *man/s2_interpolate.Rd +32c6a0bb1280c25f7da7dc9781315d92 *man/s2_is_collection.Rd +85431ef0005937d9132048784c2ac887 *man/s2_lnglat.Rd +43d061044e95b05b97b7cdb01d392708 *man/s2_options.Rd +f9c29bc19098f5fc595e260341adeacb *man/s2_plot.Rd +6c76792603f4ce1efc64879d62b59517 *man/s2_point.Rd +de7b3d7e917213319c2adcbda8c5bf98 *man/wk_handle.s2_geography.Rd +6c732afe5e32be7f5de7c7f7e344c8e5 *src/Makevars.in +2cec33791efd347cbcb051402d7cf9c4 *src/Makevars.ucrt +7f0e39e59982689e148a8c55669548e5 *src/Makevars.win +08d872fb5bb0af47a972af7330e486ef *src/RcppExports.cpp +642c7be2ce45690aea290338fec0675b *src/absl/algorithm/algorithm.h +fd93de75d5eb59a37dc897d538c9e428 *src/absl/algorithm/container.h +8dda102d5b63125edd164fe345e55f9b *src/absl/base/attributes.h +daad8899b1d374f22c26521512c3a8bb *src/absl/base/call_once.h +26a0a864fbee53bce7754262cf239314 *src/absl/base/casts.h +ed4fa7b00556a1a2bfca66015a39b85a *src/absl/base/config.h +f787189f34e273f4314344b53f5ccae9 *src/absl/base/const_init.h +033cd6a3c2de9bf80923f5054e2701e2 *src/absl/base/dynamic_annotations.h +376abfbcc035fb5ddd76fb513589a1e9 *src/absl/base/internal/atomic_hook.h +429261c728bd91fe9a974d2fcf8b8db6 *src/absl/base/internal/cycleclock.cc +b921282429acd16526077295c5ac6b6e *src/absl/base/internal/cycleclock.h +f5cec946747365781d14e7f395f3e908 *src/absl/base/internal/direct_mmap.h +3c6051b3b1895802b42fd070e0739173 *src/absl/base/internal/dynamic_annotations.h +f2a2f6524ea5494e7a484254ec99ce78 *src/absl/base/internal/endian.h +1ad6b95fe99ebc367d31ff7a9a8dd1fa *src/absl/base/internal/errno_saver.h +7cb685577c52ecbefba2c3f86f4247eb *src/absl/base/internal/fast_type_id.h +bf927fee390f1b62989522868b3104ec *src/absl/base/internal/hide_ptr.h +83cc5f6d80a853694ccd21d49215b38a *src/absl/base/internal/identity.h +ca95d2f7ac2c07926c2935004bab2077 *src/absl/base/internal/inline_variable.h +4139c861a8431f96daf9068fcddcb64f *src/absl/base/internal/invoke.h +f75b6817d994658a3311ae68b3f695f4 *src/absl/base/internal/low_level_alloc.cc +d6ce2c3d0d4abc8a9187e1c0d3917ef6 *src/absl/base/internal/low_level_alloc.h +99efd333efc4cc6a44a9abf4de89b2de *src/absl/base/internal/low_level_scheduling.h +cd26ae8fafb346728a1759160d46495e *src/absl/base/internal/per_thread_tls.h +c3e7ad34eacc65b541d0aa760ba1599d *src/absl/base/internal/prefetch.h +cff9e9f5e90b5e4455317dddbda47753 *src/absl/base/internal/pretty_function.h +cd8235040ec57fdc0cf6f85f6d8c752d *src/absl/base/internal/raw_logging.cc +2a32a3734e59aa4953be732a6cec3949 *src/absl/base/internal/raw_logging.h +bb74d44f7e47edcd1d865adb924db224 *src/absl/base/internal/scheduling_mode.h +fb9f075189d06957e1ae00ceefa34ee5 *src/absl/base/internal/scoped_set_env.cc +80dbd6581eea81b45ba532470673e777 *src/absl/base/internal/scoped_set_env.h +4e0f135a07c004bdecd6b414a7c6c546 *src/absl/base/internal/spinlock.cc +ac92c2ee2b026885a0fa47e95aeb0988 *src/absl/base/internal/spinlock.h +5e1d7bbdbd633f5f9e26bc6000d7bdc5 *src/absl/base/internal/spinlock_akaros.inc +184a057ae2d2273a6d02f7012585b854 *src/absl/base/internal/spinlock_linux.inc +63f56ff5eab205a2ba5a852dbd6276f2 *src/absl/base/internal/spinlock_posix.inc +3917a541890f75bcb371b26c11491bec *src/absl/base/internal/spinlock_wait.cc +3bc3b1f81150afa98a5b0abc10905f3e *src/absl/base/internal/spinlock_wait.h +d3c5fb1ed8db28e9fc573aa2939af254 *src/absl/base/internal/spinlock_win32.inc +f04df5e54b2977788e030fb2c44735c3 *src/absl/base/internal/strerror.cc +8c75ee873aa3e514edd1390f3056f89a *src/absl/base/internal/strerror.h +b3c6e089ed89e018ffdb93638014292c *src/absl/base/internal/sysinfo.cc +89920866f6792bd96338cb6f9587f3a9 *src/absl/base/internal/sysinfo.h +af1f8d88b67b54f7302728488ddb98a0 *src/absl/base/internal/thread_annotations.h +a720634e6b2aad1f7627838a08311902 *src/absl/base/internal/thread_identity.cc +675c5b31b96e0ebaa84e739ed8216b32 *src/absl/base/internal/thread_identity.h +94e874964762b519d4e59b796466116b *src/absl/base/internal/throw_delegate.cc +c1f9f8e9ee944f3b8a2640096b37f61f *src/absl/base/internal/throw_delegate.h +8c803550a22ccece7660ca6ad9a89560 *src/absl/base/internal/tsan_mutex_interface.h +c2045a4b16b8b7efce39c96a88adc6da *src/absl/base/internal/unaligned_access.h +51e2fc783ce386b7754b15fc92a92c81 *src/absl/base/internal/unscaledcycleclock.cc +17285df781bac882a78c1c4baa87d4a8 *src/absl/base/internal/unscaledcycleclock.h +6e85910a100a6c90893d108d2b80a7a1 *src/absl/base/log_severity.cc +f0fc3d187005a7cf23c54a12e04ab06c *src/absl/base/log_severity.h +9a7add8647ed73b494f794f18c11270d *src/absl/base/macros.h +7cb6ea6377947be6384172e8e4292fcf *src/absl/base/optimization.h +7553dbe117696c89a6214587a7664f9c *src/absl/base/options.h +30a55e8729c05984884801d5f5fa1422 *src/absl/base/policy_checks.h +c942a5d8341ce35b666c8a2e3625aed9 *src/absl/base/port.h +990163d6b22b150b08d9bec43b61bf8b *src/absl/base/thread_annotations.h +4903f7daa7a9b816419e6389bdffbf4a *src/absl/container/btree_map.h +d248b7efb25539492e4da9760d0f71ee *src/absl/container/btree_set.h +db720952112a075bb8ed68171b681ce1 *src/absl/container/fixed_array.h +1029770c28381ec983f929a950525aed *src/absl/container/flat_hash_map.h +31b4c898d148515c41d65eff48710843 *src/absl/container/flat_hash_set.h +72d3d49a360d5d05518b245d80da2e0b *src/absl/container/inlined_vector.h +0c9fc5d35bb8ce4a2d4b2aef0a065b40 *src/absl/container/internal/btree.h +471f4b576592fa97439abe4c1f4f5af6 *src/absl/container/internal/btree_container.h +04112e121678319b6f74e2c7c4bb2877 *src/absl/container/internal/common.h +013341274845cbce0b85a94bc2984b38 *src/absl/container/internal/compressed_tuple.h +38c85e29bfe2d83b08fb2ee63e84f428 *src/absl/container/internal/container_memory.h +a47304f7692364f874a92cce7526a13d *src/absl/container/internal/counting_allocator.h +619244bd2568402acdf114998c58fc52 *src/absl/container/internal/hash_function_defaults.h +2091e3185f7126b826ce099f50f5fdc9 *src/absl/container/internal/hash_policy_traits.h +ed69e77d87d7e4c2b42f5248904cd6d6 *src/absl/container/internal/hashtable_debug.h +85a576cfb12917cd7126d9e76e13309b *src/absl/container/internal/hashtable_debug_hooks.h +a9026aae4987bd08dd8def14e33574de *src/absl/container/internal/hashtablez_sampler.cc +e6a1d3c5965aea1bb0a84e5ce8d3ad08 *src/absl/container/internal/hashtablez_sampler.h +fcc5f2abdf1fd0edb18d122a846e84f0 *src/absl/container/internal/hashtablez_sampler_force_weak_definition.cc +0f420d68cbca4e9d9ca1274179d15e8e *src/absl/container/internal/inlined_vector.h +dde28d12969cb6df933b96f90b287506 *src/absl/container/internal/layout.h +a8db9c2e305ee7ddc082d6d1bcf74103 *src/absl/container/internal/node_slot_policy.h +7256093f3d9e6d880cf9b25295e0c61e *src/absl/container/internal/raw_hash_map.h +80426b9c7f99e24919cdfce7af00df15 *src/absl/container/internal/raw_hash_set.cc +7627e3cd5ecb10ed4ab6ba09a1ad7fee *src/absl/container/internal/raw_hash_set.h +b2c91b1fd5a05ea12fd76abd5e7401bc *src/absl/container/internal/tracked.h +dede177b23ded734f450d112907a7981 *src/absl/container/node_hash_map.h +fa098d67166a0c79cfb38f955d62aa7d *src/absl/container/node_hash_set.h +83dca33b15e6052405cf7737516cc6f1 *src/absl/debugging/failure_signal_handler.cc +33d4e12e60b432bafa7920d2b5fa67c0 *src/absl/debugging/failure_signal_handler.h +88511382784375519a3ef65a5ce247b9 *src/absl/debugging/internal/address_is_readable.cc +91ad1bd868606c791022ba6aa6ef4860 *src/absl/debugging/internal/address_is_readable.h +c8e0eb65e6f8010624c5e8b584780240 *src/absl/debugging/internal/demangle.cc +de5e466a575d551be06c658446b26ca9 *src/absl/debugging/internal/demangle.h +e54662f502407a088f04a6d51708f1e5 *src/absl/debugging/internal/elf_mem_image.cc +352cfa6b3dd2465cc293e610a0e38769 *src/absl/debugging/internal/elf_mem_image.h +dc8caa62e93f72fb1c5e4cc1fd6f0c34 *src/absl/debugging/internal/examine_stack.cc +19d1e4c50158c08334756c0fcd65728a *src/absl/debugging/internal/examine_stack.h +76431729cc82622bc53935063b049ef6 *src/absl/debugging/internal/stack_consumption.cc +817879abccfc61d31c97aa15e5b5a8df *src/absl/debugging/internal/stack_consumption.h +dc3e0ee0e29049fcd25242c4b14ca920 *src/absl/debugging/internal/stacktrace_aarch64-inl.inc +921fde95b6a08aa52d9cb628bb79c15d *src/absl/debugging/internal/stacktrace_arm-inl.inc +67da7c45a5fe837feb76fcfeff37f28b *src/absl/debugging/internal/stacktrace_config.h +d412bbe566473c31ffb571b5df159589 *src/absl/debugging/internal/stacktrace_emscripten-inl.inc +31136485c4ae4719146fe310d09794b3 *src/absl/debugging/internal/stacktrace_generic-inl.inc +4dddddc30c71242db2eff066287537fa *src/absl/debugging/internal/stacktrace_powerpc-inl.inc +8827fe270ce53a56049f0ea9d7d91429 *src/absl/debugging/internal/stacktrace_riscv-inl.inc +e7ced4e67444a66063312dc27d7bb17d *src/absl/debugging/internal/stacktrace_unimplemented-inl.inc +36cbcb6faf6816b2fd3db94ed61e077d *src/absl/debugging/internal/stacktrace_win32-inl.inc +4a9fddacc38ff421181282981ad11321 *src/absl/debugging/internal/stacktrace_x86-inl.inc +7528d1aa40e6a6645a8173fcbf8150cd *src/absl/debugging/internal/symbolize.h +66aa580ffd00da806719b6e4fee4fd67 *src/absl/debugging/internal/vdso_support.cc +32da4040ebc4e635623be36e861380d4 *src/absl/debugging/internal/vdso_support.h +944240b259867fbd920a992cb1b5610a *src/absl/debugging/leak_check.cc +803e646bd74bcc61527c082b05a41573 *src/absl/debugging/leak_check.h +fdb200b5ff89b0388299a36738915d29 *src/absl/debugging/stacktrace.cc +54920d74c95e9623f986ca3bf9a49c14 *src/absl/debugging/stacktrace.h +82338e576ae2cc4fef71e50f421164cc *src/absl/debugging/symbolize.cc +2a4f1e90fdc9a56d8812afd0f0291876 *src/absl/debugging/symbolize.h +e1ff8c65c67e769e4d929d70b4c315a4 *src/absl/debugging/symbolize_darwin.inc +f91083e80a5e31074987a0b78f5de558 *src/absl/debugging/symbolize_elf.inc +505df4d563b199a1accb368724eb8dd2 *src/absl/debugging/symbolize_emscripten.inc +37d1355403219402fa4d888e9be56fca *src/absl/debugging/symbolize_unimplemented.inc +59dde77656bded8e7a9c0141349f6230 *src/absl/debugging/symbolize_win32.inc +bde9e8a109a26a709ac7f6cd32d26335 *src/absl/functional/any_invocable.h +086310a4f0676fb86df40fcdac96d2a6 *src/absl/functional/bind_front.h +f3ea4b26c77278c38b2e7c8d88f0ba2f *src/absl/functional/function_ref.h +e62331c1f6ffc3458ada2d1d3dbff3bb *src/absl/functional/internal/any_invocable.h +ca707a4761ddef188eefa1c47a49305d *src/absl/functional/internal/front_binder.h +acc1da8f94634fb69d04d64af189f7d9 *src/absl/functional/internal/function_ref.h +c88a984158d3e1eaf700a703a747d9c0 *src/absl/memory/memory.h +a2462b42675608566733fc11ab1dbd40 *src/absl/meta/type_traits.h +7caad1d859f008e728d1a783e2007745 *src/absl/numeric/bits.h +fd2f0faf81574b2c86dd29356e120fac *src/absl/numeric/int128.cc +79cc004d6d6bf6bebf584022f33bec9e *src/absl/numeric/int128.h +2128cf980b91df6c0fc6c218d0497761 *src/absl/numeric/int128_have_intrinsic.inc +437ad46e0474b28cd231540fba4a8968 *src/absl/numeric/int128_no_intrinsic.inc +cda33272dc55c7d4e85c050048ac3b68 *src/absl/numeric/internal/bits.h +b71517976c0230c0f0be9e102124708b *src/absl/numeric/internal/representation.h +97c5bf0389ead2a8d9584b5516e03f30 *src/absl/profiling/internal/exponential_biased.cc +8dd473273bce3e970afd8a2592a00780 *src/absl/profiling/internal/exponential_biased.h +3c781cfebcbab764fa4e56e941e07a74 *src/absl/profiling/internal/periodic_sampler.cc +d815a430f4c9ff6f941a741742112bff *src/absl/profiling/internal/periodic_sampler.h +93e67b064bc20858ed6fba23748b7ddf *src/absl/profiling/internal/sample_recorder.h +d8fbe682d7f971ad41cf3585695e5b8c *src/absl/strings/ascii.cc +b1212449ec5b585102abd5fef8394f88 *src/absl/strings/ascii.h +d52529dfcbd429d31cf34475d9345029 *src/absl/strings/charconv.cc +e022adbe4ace51d7a2d3a065e30fc511 *src/absl/strings/charconv.h +8eebdbf778e886f261a8990232281e37 *src/absl/strings/cord.cc +1f72516104e63dcb7557757bfd39a4d8 *src/absl/strings/cord.h +38bf276145e6768ce7d55bc72c51f4a6 *src/absl/strings/cord_analysis.cc +9eb388a8f95c0a8b6c629e139f5b19a7 *src/absl/strings/cord_analysis.h +fc2b8427acd714ecfa2c87c4d653df20 *src/absl/strings/cord_buffer.cc +d87e5ed66957627950977d971f80dcba *src/absl/strings/cord_buffer.h +3d3ee945b740992e7977948f64d37abf *src/absl/strings/escaping.cc +ca2211cce011d818358126d737a1b960 *src/absl/strings/escaping.h +9d688a5ba34606869bbb97318c6e3583 *src/absl/strings/internal/char_map.h +9ba44e5ca0ab3eaf2eea1c1d1135cbd2 *src/absl/strings/internal/charconv_bigint.cc +effc6ae35cf71076cb6d243984ef02db *src/absl/strings/internal/charconv_bigint.h +cda5aec99cde64d542c231aec6eecaeb *src/absl/strings/internal/charconv_parse.cc +bbb89ffd31b13599454ac66c7fca1021 *src/absl/strings/internal/charconv_parse.h +265c664ad73036c39cd42631e3d56ec2 *src/absl/strings/internal/cord_data_edge.h +b2aeb1b3985e3dd0e321c60975686d7c *src/absl/strings/internal/cord_internal.cc +89fc71fb8fc9cb06b7001af62a550023 *src/absl/strings/internal/cord_internal.h +d2efb216cd07c01bfa472b8adc633706 *src/absl/strings/internal/cord_rep_btree.cc +9a883463d6431cdc5a0689383884474a *src/absl/strings/internal/cord_rep_btree.h +2143a8c51257e8b4e4b165f155b38a1a *src/absl/strings/internal/cord_rep_btree_navigator.cc +6f6e44ad036456711251cd952f475a00 *src/absl/strings/internal/cord_rep_btree_navigator.h +e7ad5632b2aed15ce6f88246a8e40f6a *src/absl/strings/internal/cord_rep_btree_reader.cc +d2b0c3364d0d8e4a11288617889bfd60 *src/absl/strings/internal/cord_rep_btree_reader.h +62b2b3eaa89a7fa5bf2f790b61c80437 *src/absl/strings/internal/cord_rep_consume.cc +f500a141d4fbada9fe73c9bde8e9b997 *src/absl/strings/internal/cord_rep_consume.h +3c8f9400090931ecc6aed37d19b4214b *src/absl/strings/internal/cord_rep_crc.cc +8a1a2a0db3e9d4ab31dbb86f624277a3 *src/absl/strings/internal/cord_rep_crc.h +d701cf5c641280417b0290e8ff177dcd *src/absl/strings/internal/cord_rep_flat.h +160f554c0b0acb5c287ff2fa457f8b5a *src/absl/strings/internal/cord_rep_ring.cc +70ff93c8943bb54f4ec4b7bc03d8e298 *src/absl/strings/internal/cord_rep_ring.h +1b95eb768dc0a65c2aca391ce917168b *src/absl/strings/internal/cord_rep_ring_reader.h +2bb5021298c3f89948deb15c1cc14e67 *src/absl/strings/internal/cordz_functions.cc +7797f5c82bed0f2dd95d9a491164d0cd *src/absl/strings/internal/cordz_functions.h +d897bb95f3792816dae12cc99600f0ee *src/absl/strings/internal/cordz_handle.cc +2f4d12d83ff4611f3e8278a08a1f1dbb *src/absl/strings/internal/cordz_handle.h +46b23e9b9e81676d6633ab1edcbef807 *src/absl/strings/internal/cordz_info.cc +cceee7615f63027b2da6eead19bbab18 *src/absl/strings/internal/cordz_info.h +f20b57c507b149391110f0c4ec58e0eb *src/absl/strings/internal/cordz_sample_token.cc +8804ca3803da6893811636e9fe775b5a *src/absl/strings/internal/cordz_sample_token.h +7e6d750252301ed305689311fc447a6d *src/absl/strings/internal/cordz_statistics.h +406e4a93e3c6af34fe85e6e4ca4ef693 *src/absl/strings/internal/cordz_update_scope.h +4a710c85201ef57bdc282b2003186b29 *src/absl/strings/internal/cordz_update_tracker.h +3b55edcf191a2ece5a8f5071f3b186d2 *src/absl/strings/internal/escaping.cc +dd1f4cd55816cd3df569601375e19dd6 *src/absl/strings/internal/escaping.h +bbcc2e8ef020b2ce6fd2e17e7e019dd7 *src/absl/strings/internal/memutil.cc +73102cf47e4f94aee5dd43faa699eed5 *src/absl/strings/internal/memutil.h +2fd122662d4524b7e60268ed88de2699 *src/absl/strings/internal/ostringstream.cc +f4dd448efcf57c35b09e4ecf12ba595a *src/absl/strings/internal/ostringstream.h +0f79e5539614bead9e8b2899109414f2 *src/absl/strings/internal/pow10_helper.cc +a2710593664b501447140371f8f2910b *src/absl/strings/internal/pow10_helper.h +d41ed4ac8b5c0439686eee2a437ba24a *src/absl/strings/internal/resize_uninitialized.h +0b127a362be3ed6ce297a9bd6704b6f5 *src/absl/strings/internal/stl_type_traits.h +7e645ceaf6302cb953b278e7bd170b11 *src/absl/strings/internal/str_format/arg.cc +bbf7a17d739c9e9388ed9bbb30852ca2 *src/absl/strings/internal/str_format/arg.h +541807e23166b8458e4fe668064abff2 *src/absl/strings/internal/str_format/bind.cc +642c25c2a95404dbfbad403297c4f6e4 *src/absl/strings/internal/str_format/bind.h +25e4d30547e25840e7ade5d5361e4a6c *src/absl/strings/internal/str_format/checker.h +f1a8157f63bb446070dac767129701f9 *src/absl/strings/internal/str_format/extension.cc +7b166dea6c1b8683f113e9bfa498896c *src/absl/strings/internal/str_format/extension.h +602d5b81943933d209d9347e2138e031 *src/absl/strings/internal/str_format/float_conversion.cc +1775ec23d1a0b61c9c3cbc2786e20b9d *src/absl/strings/internal/str_format/float_conversion.h +38f00d24ace7d912f9410c77298f9450 *src/absl/strings/internal/str_format/output.cc +b9bb4e0ffdab9b7dd39fd2448df38651 *src/absl/strings/internal/str_format/output.h +bc2c67e0646b6442d7e594737371a059 *src/absl/strings/internal/str_format/parser.cc +da983b71caca5afea2879a1877269726 *src/absl/strings/internal/str_format/parser.h +3982f532a08d74ed1a024162c0752374 *src/absl/strings/internal/str_join_internal.h +31be88550fefe52ab3ede699f1e0ffa7 *src/absl/strings/internal/str_split_internal.h +e011f29d010463642a59a736caabdbd1 *src/absl/strings/internal/string_constant.h +58e100deab3a1a5d102f6767a106b023 *src/absl/strings/internal/utf8.cc +da238e6d26546310d03ce815a181f9c3 *src/absl/strings/internal/utf8.h +c26d8dc3c6998d1e4adb73b7b3e7307d *src/absl/strings/match.cc +d78b85f24f13f203df715033ebbedac4 *src/absl/strings/match.h +1ffbf9839fb7e742c2b9b5fc5d2ba8bd *src/absl/strings/numbers.cc +4ce4d257979a3fb096d3eee6c8285449 *src/absl/strings/numbers.h +4ebd95dbd026cacebb04407f77a40998 *src/absl/strings/str_cat.cc +499e4504c480d651cbbb45878ddd07d0 *src/absl/strings/str_cat.h +a0e1b0af9234c7d668a9542a31df2bc0 *src/absl/strings/str_format.h +305b931d297cecef7a7a5b1cbc92d183 *src/absl/strings/str_join.h +2b56176c8350695b37d0d810fe705486 *src/absl/strings/str_replace.cc +4cfe52299b13b659b0b79e926310859c *src/absl/strings/str_replace.h +0e0684d40f7ebcfa5e3ee516d602bc1b *src/absl/strings/str_split.cc +85eb91c7dbf8f799c2f6fdc862a9de57 *src/absl/strings/str_split.h +9513d8a3c60cffbd288b9f571af3599e *src/absl/strings/string_view.cc +bc6088f45dcc4f91be39442fe9b9c223 *src/absl/strings/string_view.h +202bdeadb019318d35a2141b0bf90ed8 *src/absl/strings/strip.h +317fe715d78ae47d1660f1a600d34fd8 *src/absl/strings/substitute.cc +35662cdc3431d806eafd35cf21ab276c *src/absl/strings/substitute.h +e8a0e5ed8343c44e8d414a72596dbde9 *src/absl/synchronization/barrier.cc +104730dc624a422f9e0610e357d1d319 *src/absl/synchronization/barrier.h +7adb87ccae08cfb4f5163c38043307b0 *src/absl/synchronization/blocking_counter.cc +76ca084891876db3ba604bb86c36e714 *src/absl/synchronization/blocking_counter.h +bce6fe21f3819cc963d73dcc7b7ce45c *src/absl/synchronization/internal/create_thread_identity.cc +51ffeadf7c798c4c32cd7ad26786b7c3 *src/absl/synchronization/internal/create_thread_identity.h +edbc4b6b4c303dd6a1c8856356bbdccf *src/absl/synchronization/internal/futex.h +39446c3969b401efba5b1b875a97ba17 *src/absl/synchronization/internal/graphcycles.cc +98204bc87786a909620b798d82b44258 *src/absl/synchronization/internal/graphcycles.h +0c51cca58de692a4dbe0f95750a20512 *src/absl/synchronization/internal/kernel_timeout.h +4fbf2a922ba2d618cc960c516c6858aa *src/absl/synchronization/internal/per_thread_sem.cc +07ff0b00a159d1e49e78f4953a375749 *src/absl/synchronization/internal/per_thread_sem.h +dcd3d301b476895c2de035ae9f4a9c2f *src/absl/synchronization/internal/thread_pool.h +cbf5c2ab41f486164b71e940d4df35c3 *src/absl/synchronization/internal/waiter.cc +6b034e180a573148b7512a25340e53fd *src/absl/synchronization/internal/waiter.h +4c313ee4f352a843fecccbd6a4047927 *src/absl/synchronization/mutex.cc +76e58005875696e6a2b00f38b22a9cad *src/absl/synchronization/mutex.h +18f40d74c189e4cb3bade5bbfac62eac *src/absl/synchronization/notification.cc +9e989de0e62bd8e3b804141f78ded6e9 *src/absl/synchronization/notification.h +43151664d8fcbd888e2b320ea6c59d4f *src/absl/time/civil_time.cc +21f77d9a8b12daee08647a79fa2f1c24 *src/absl/time/civil_time.h +5bec957f3a6069a1bbf1747096d51829 *src/absl/time/clock.cc +7669a9d4fb695b4bdf63f2199778b7da *src/absl/time/clock.h +f770f6ea46e1f6d5f99d40b058f76a23 *src/absl/time/duration.cc +1029367259d746995cfcc86300cb550e *src/absl/time/format.cc +153fe848b8d395acdc4d7188b6b1f7bb *src/absl/time/internal/cctz/include/cctz/civil_time.h +6061b7a2fb993133e2d77ab90882e4b3 *src/absl/time/internal/cctz/include/cctz/civil_time_detail.h +58a6845478b0a635b7474602a86154e9 *src/absl/time/internal/cctz/include/cctz/time_zone.h +6f945a703ababb99477bff55062287ec *src/absl/time/internal/cctz/include/cctz/zone_info_source.h +65ca6d96e925b084e0d56180aa5fe84d *src/absl/time/internal/cctz/src/civil_time_detail.cc +7d6876b726bdc87d68adb5a3be7523a6 *src/absl/time/internal/cctz/src/time_zone_fixed.cc +3afdc610576df58fccc41a7029f1591b *src/absl/time/internal/cctz/src/time_zone_fixed.h +37a70a5e5c9f72abb36cda2b9dfef5dc *src/absl/time/internal/cctz/src/time_zone_format.cc +977f9a0fc17a72093bc4c8a87784f632 *src/absl/time/internal/cctz/src/time_zone_if.cc +5aaf522801582f1c4e511bd5f47143cc *src/absl/time/internal/cctz/src/time_zone_if.h +b9933f8f7ea6ef25b5ca3ff1bf3e0ac9 *src/absl/time/internal/cctz/src/time_zone_impl.cc +5b006ce94d5395cbd204c8eb42aa673a *src/absl/time/internal/cctz/src/time_zone_impl.h +b920d1e53b3556ba81f65140c60c5517 *src/absl/time/internal/cctz/src/time_zone_info.cc +39446561973c7f453d351f8dd0bfc947 *src/absl/time/internal/cctz/src/time_zone_info.h +9be74d3b67a28a954fc5f9cb9721e173 *src/absl/time/internal/cctz/src/time_zone_libc.cc +6205d25a90cd16fe6fef686eac98a13b *src/absl/time/internal/cctz/src/time_zone_libc.h +baf6b2ca44a23c2ffc57a5185eb28c42 *src/absl/time/internal/cctz/src/time_zone_lookup.cc +e54d65914d1f72e412abe32cb94b12df *src/absl/time/internal/cctz/src/time_zone_posix.cc +21a06775679a372f33ac8918b0b03dfe *src/absl/time/internal/cctz/src/time_zone_posix.h +de6d3a441d96d5b0632441a939602bd3 *src/absl/time/internal/cctz/src/tzfile.h +09c17fe44b3b7d5eb61c51d642d227c9 *src/absl/time/internal/cctz/src/zone_info_source.cc +f510c27e928680c4df82fcd584b13620 *src/absl/time/internal/get_current_time_chrono.inc +921b6cfae31d6e15730d211d301f5bed *src/absl/time/internal/get_current_time_posix.inc +012254523fe4c7e1842f31b277e9bc60 *src/absl/time/internal/zoneinfo.inc +8893c1ef798bfa8c6a3421eafd82d0aa *src/absl/time/time.cc +f2405de29587b4932366a36e0ce8c607 *src/absl/time/time.h +c5e6e75a58fb6134a7bdad5ff6ef94b4 *src/absl/types/any.h +31416a5f0241128632dde299c016fe61 *src/absl/types/bad_any_cast.cc +e57304b355027637c47fa5b4138b5ef0 *src/absl/types/bad_any_cast.h +76faa97137f7f2f4eb70152889e11227 *src/absl/types/bad_optional_access.cc +ffd0f2debe86db173721761f28b64480 *src/absl/types/bad_optional_access.h +339cb16d4f7c125ef6f9975e9c70e2fd *src/absl/types/bad_variant_access.cc +63b5f63942914ccdaf91a52ae46b5b8f *src/absl/types/bad_variant_access.h +216ae502b816cc541789db0915e4dd76 *src/absl/types/compare.h +2466c511230ee6497dc26dd3633f7702 *src/absl/types/internal/conformance_aliases.h +aa319c2077664d2c39a6bd0813c79c18 *src/absl/types/internal/conformance_archetype.h +f5863e36a75e055aee57a18f967c5489 *src/absl/types/internal/conformance_profile.h +a87cad8cf13eaebb402020f8d7fe4119 *src/absl/types/internal/conformance_testing_helpers.h +0ceabff37c1e8943c78fc2a10db32e94 *src/absl/types/internal/optional.h +9d178054161b76df535a760be34d6213 *src/absl/types/internal/parentheses.h +27f0c5818e53bdc944c73938edf49a15 *src/absl/types/internal/span.h +599feda2b9b004573e161a9550054e99 *src/absl/types/internal/transform_args.h +637f1331d00d757439fcba667b80a4f5 *src/absl/types/internal/variant.h +29994a10008e74ff8963787848912fbc *src/absl/types/optional.h +6e8045dbd0797fcac5d8b9dbe5c6a322 *src/absl/types/span.h +1d24f87c9fb98d810977a981043d8201 *src/absl/types/variant.h +9b15fdb150cb72ea63d34d08fffd39c2 *src/absl/utility/utility.h +5f46d29bc94bcdfc28e2a4df442310de *src/cpp-compat.cpp +afabc65ce7a01abdb14efb1abccd7247 *src/cpp-compat.h +c20b3dc7b63f3bbf6af6827ac6557ca9 *src/geography-operator.h +023f7dbc8c3dd00c481507be51a9ca91 *src/geography.h +0365ec85c22784d83cce5b84d71d794a *src/init.cpp +62ed6c893b0bc0b75b67e9b02706a0b5 *src/s2-accessors.cpp +4cfaeb17ebda5876ac7110d5e15b6e87 *src/s2-bounds.cpp +10bfc12436ce2945f667824653c829d1 *src/s2-cell-union.cpp +16dc89505d02a555613082143e1d79a7 *src/s2-cell.cpp +cf59fd9682672c99319ba906455e5b32 *src/s2-constructors-formatters.cpp +19657182677a20a4e14bcd89e03f6f06 *src/s2-geography.cpp +e9f1b666c7a410cadaa4dfa34feba4b2 *src/s2-lnglat.cpp +db77ba68fab15ce99c37d6d50471cbbf *src/s2-matrix.cpp +96980fe0bed80e6566fbbd84bc8dcdf8 *src/s2-options.h +70c6d7bd00bb98b98a35f9ca8d77677d *src/s2-predicates.cpp +9c30c6461c446d323ae75d0b9f2395e5 *src/s2-transformers.cpp +8428235fc1165d0c32375e2e21a686fe *src/s2/_fp_contract_off.h +28e49579436bf834cd114aa6f01f396e *src/s2/base/casts.h +4c10c37f35c3ea32da84e18e427e0a8f *src/s2/base/commandlineflags.h +95d68a4bda8ac1ed1238c06999c7fe8f *src/s2/base/integral_types.h +188fa699db99f7b32bf12c8f5d49afab *src/s2/base/log_severity.h +e6796dc6fac7834af2337a6298f6b4f2 *src/s2/base/logging.h +322d31aa00d2046a0f35c7e373748985 *src/s2/base/mutex.h +bc87e8d47e0131c7f3b2e943823c79f8 *src/s2/base/port.h +7217ae2988c45fe3c657a6fbb5a01334 *src/s2/base/spinlock.h +4c6c57a3c08c66211d519f387ea8e972 *src/s2/base/stringprintf.cc +df6c98233d53cb881136f5f52d2c8012 *src/s2/base/stringprintf.h +c10c3c7ca58239c0915bacffc70e37e1 *src/s2/base/strtoint.cc +ac9edc4a5c5139604384d611a9d5686f *src/s2/base/strtoint.h +8a23b88128feaefecac82381870b7e1b *src/s2/base/timer.h +a85315a801220cea6aedb7545bed0675 *src/s2/encoded_s2cell_id_vector.cc +ae93520447e17b7a5ee570e973b14058 *src/s2/encoded_s2cell_id_vector.h +d7cc642275acd754dd38bc1db012beb8 *src/s2/encoded_s2point_vector.cc +d7870427a21d7a4288a82ef3d9002404 *src/s2/encoded_s2point_vector.h +1c9eda503a7610170bb7b20b221a8123 *src/s2/encoded_s2shape_index.cc +8c71134bbac741ab14ad58645c587efd *src/s2/encoded_s2shape_index.h +7ded14eeab5eba2a2cca20c71b837d1c *src/s2/encoded_string_vector.cc +2b5b150cb6a9356b9360ab4bed8fb9e0 *src/s2/encoded_string_vector.h +9e444256aa6a7b5e5c3eb73ba1708065 *src/s2/encoded_uint_vector.h +788f48e5bbfca22ed1e396d4bd675ad3 *src/s2/id_set_lexicon.cc +e7cbe5a6c88b9a286b17adb5e690286f *src/s2/id_set_lexicon.h +e77194370c1950bc155ee61708af57b0 *src/s2/mutable_s2shape_index.cc +f2106aebeef507adcbc699ee5bb2ae5b *src/s2/mutable_s2shape_index.h +a5fd0e84e01c56c79c50f208ae1b08b9 *src/s2/r1interval.h +e363bde392175774fc26dcde1c3b5076 *src/s2/r2.h +da84b9bfc2aeabd4f8ae0f84343623e7 *src/s2/r2rect.cc +4b12844b6a0327fbc6f0488b4fbea0b3 *src/s2/r2rect.h +e7ca7f23a7e8fe6f707e4169f1aadb77 *src/s2/s1angle.cc +0e9a83acf9613dc68174c6575afa373d *src/s2/s1angle.h +d9387a9af7477f6b82ec73d206cfd8d9 *src/s2/s1chord_angle.cc +7b2a01d9f67a03390a0f34b03a46fef8 *src/s2/s1chord_angle.h +34874873bee5edbf9af38cd7692ab3ae *src/s2/s1interval.cc +2a46bfae65f4a24dbbc0c8a7df478c4c *src/s2/s1interval.h +ad8719b6e9e1c836e9f4c59b216739c9 *src/s2/s2boolean_operation.cc +99423d6009cb68d7879793f9311e1bb7 *src/s2/s2boolean_operation.h +aead3947dd83dd0ed2d4becb7d14d426 *src/s2/s2builder.cc +5cf1544718ecb4046621e2b0b89a0222 *src/s2/s2builder.h +bbdb9da0943cf55e60e846d2b0e7187b *src/s2/s2builder_graph.cc +cc61ae8fff8370c728bf6de7513e06d9 *src/s2/s2builder_graph.h +0558b2bb40025b583974ac1a5d2b5857 *src/s2/s2builder_layer.h +485fed02d2c102ef03502c959735152c *src/s2/s2builderutil_closed_set_normalizer.cc +4b17df092dd21d0d5b22499a7a0d8d63 *src/s2/s2builderutil_closed_set_normalizer.h +128b04413654742d9655bc2b12dfeda7 *src/s2/s2builderutil_find_polygon_degeneracies.cc +dc965c407adeb3070db3510de3b1f0bd *src/s2/s2builderutil_find_polygon_degeneracies.h +59390c82ae51ffd8dc8a2154c7f5cc6b *src/s2/s2builderutil_graph_shape.h +85e926aef77aeccd2d0384001b3e7c09 *src/s2/s2builderutil_lax_polygon_layer.cc +8d156a2b07ddd94dd53ed45a086ddabd *src/s2/s2builderutil_lax_polygon_layer.h +6c0c3e9f73a2d693d25392a6f053cca9 *src/s2/s2builderutil_s2point_vector_layer.cc +5bcdb1842aff634eb86151c41aa92942 *src/s2/s2builderutil_s2point_vector_layer.h +338a73e56e2a0406bc308dcb7ae2a1db *src/s2/s2builderutil_s2polygon_layer.cc +d0f7d3975bf12ad12e5a453b1c664879 *src/s2/s2builderutil_s2polygon_layer.h +bc24648fc1ba139430b08df68cc45004 *src/s2/s2builderutil_s2polyline_layer.cc +0c78a6937a60f24a9896a4d411b68e90 *src/s2/s2builderutil_s2polyline_layer.h +9ed299b94f9fd05d3b228cbbb77a29e9 *src/s2/s2builderutil_s2polyline_vector_layer.cc +8177eb92bdd7082985e5e2cef2cbb57e *src/s2/s2builderutil_s2polyline_vector_layer.h +41c634fe2532fd5946349947743d9b0b *src/s2/s2builderutil_snap_functions.cc +14fae38b064b536a50fd99090bf2bb80 *src/s2/s2builderutil_snap_functions.h +0e7eb31b407e02950751a990c528833d *src/s2/s2builderutil_testing.cc +e91fca8f2ddf6c1826122f601803cf6f *src/s2/s2builderutil_testing.h +15661d854ecdbd9b5adb37f4fa38e7fd *src/s2/s2cap.cc +1aaf20b1794973131d85bbd6d16fc296 *src/s2/s2cap.h +d89508276ed169153d1e80d85989bfd5 *src/s2/s2cell.cc +cc050d60fc6b31479ad2cfc444aebc2b *src/s2/s2cell.h +df582edaa44ad77d7a5d8e9fb7c34262 *src/s2/s2cell_id.cc +2b7fcd907f3904fa8e8fedf12fd79303 *src/s2/s2cell_id.h +b77c66885dd9ce9decb058448a53bf2b *src/s2/s2cell_index.cc +c967db283c31cf3b04720da4aaf951be *src/s2/s2cell_index.h +2df8fa84fabd7e103ecbcf3dddec4ab7 *src/s2/s2cell_union.cc +b1877f966f026dafb27bd156d26d8194 *src/s2/s2cell_union.h +cf1a07e72197832f2b1302f7b3630f98 *src/s2/s2centroids.cc +b0e46b1f9b0302bf28993c83b1c29952 *src/s2/s2centroids.h +d23a0e7ec3560f1191df801a3ded25e4 *src/s2/s2closest_cell_query.cc +1022d523b50d2186b166ca68dc0deb11 *src/s2/s2closest_cell_query.h +1b857efbfb27e180b3cb9a554e64048d *src/s2/s2closest_cell_query_base.h +16334d3128469be0f498a5e1a6427659 *src/s2/s2closest_edge_query.cc +a239c16e1797acb32e845a6da24ca0b0 *src/s2/s2closest_edge_query.h +a4484a9f30e74392ad90ac037b3404d3 *src/s2/s2closest_edge_query_base.h +0974a6dabc5a7541eebf0e81656e522c *src/s2/s2closest_edge_query_testing.h +09e2f693ad6e491a4d50eb8b5d966840 *src/s2/s2closest_point_query.cc +e982b798e56156c65fb5ddf919a0520c *src/s2/s2closest_point_query.h +ffdeee92985536a16b5a6212547872ba *src/s2/s2closest_point_query_base.h +c5ac9b7327a1f2eb8c9b8683973bc48c *src/s2/s2contains_point_query.h +8cea937d5d1881cdb71d1287f5474cff *src/s2/s2contains_vertex_query.cc +0a9a1857a2e79884b600de88b69cefbe *src/s2/s2contains_vertex_query.h +06d65d3cca94ccb99beb7e33f5267e10 *src/s2/s2convex_hull_query.cc +7c2da76e9c9917194d48939ee919f088 *src/s2/s2convex_hull_query.h +effdbbf64bb7f235381201657e021ea6 *src/s2/s2coords.cc +e2a5abdcf611242bdf5cf0edf0eea5ad *src/s2/s2coords.h +28c4af3631469e232aab685be3e33122 *src/s2/s2coords_internal.h +48655b45fa96f0da0b234ca68fe4b54d *src/s2/s2crossing_edge_query.cc +28b59ab0651bf933d5271393e3313deb *src/s2/s2crossing_edge_query.h +cbeeb65a7261dc8430dcf6edc24deb94 *src/s2/s2debug.cc +12e9446c9a552088c70b7b3062c1ba5b *src/s2/s2debug.h +7298228fc7e23751181ad62e211a81b8 *src/s2/s2distance_target.h +a863fc668a334819762480cc0646c898 *src/s2/s2earth.cc +b90e02812e7d1afa32d15de3a0245dd8 *src/s2/s2earth.h +d58b020152efbe99f672b8952460d2ce *src/s2/s2edge_clipping.cc +cb59ea7179d978f41ecac9ff2ab101c5 *src/s2/s2edge_clipping.h +c5fd68f5fab9c30285a9d1995be3ddde *src/s2/s2edge_crosser.cc +91b536b7351c85048ac5f4fc8f9c36f3 *src/s2/s2edge_crosser.h +69bb6ff8072fe88be08b2e693071ea65 *src/s2/s2edge_crossings.cc +6de8f9979a94b2001e3134b4b10391a5 *src/s2/s2edge_crossings.h +482bff39a41ea67b1073b9cfdc11400b *src/s2/s2edge_crossings_internal.h +3bdbadd260d873672d7821d8a0f557ed *src/s2/s2edge_distances.cc +db89cee1d7d37be1b7f889cc92df8e94 *src/s2/s2edge_distances.h +75acbcdde0e718a3a990fa63c78c3423 *src/s2/s2edge_tessellator.cc +0f223cfa5b4b1153617bd64d47f68212 *src/s2/s2edge_tessellator.h +99195e00ed62fc8127a6de6305730930 *src/s2/s2edge_vector_shape.h +483c797be9c9eb4b9a9059577e548e7b *src/s2/s2error.cc +30c2e0bb485124026064b8a6e680c60a *src/s2/s2error.h +fe0c39101c8c58f45b75fc4c36261a82 *src/s2/s2furthest_edge_query.cc +9ab743cac7f27a70e2f49f6a8d659c49 *src/s2/s2furthest_edge_query.h +fe5f7037943d797d93f4d2be3d1119b0 *src/s2/s2latlng.cc +f86006f14f82fb5fb2dc61b217d79be0 *src/s2/s2latlng.h +e42ec23a4fdf8b6aea983db2ca3cc4d9 *src/s2/s2latlng_rect.cc +bd642eda609967162d94ac500632e454 *src/s2/s2latlng_rect.h +f8689b6a3eef9053601d310ca6146f5c *src/s2/s2latlng_rect_bounder.cc +486a11f20722a978644e99ec990516ad *src/s2/s2latlng_rect_bounder.h +6ea59dae602c6e87765068c7c4357baf *src/s2/s2lax_loop_shape.cc +6aaf1e13f42ac06794d6179a45992a12 *src/s2/s2lax_loop_shape.h +fc760f9c0ff8545929b4a1e0da1e6ff5 *src/s2/s2lax_polygon_shape.cc +35b22571a5ddbe1e7b91e6497b39e111 *src/s2/s2lax_polygon_shape.h +56c7b3c6c942d0cffe353b03ee605a8f *src/s2/s2lax_polyline_shape.cc +4d4f60c49166c8657258de45884c4087 *src/s2/s2lax_polyline_shape.h +818253e9f331b59caf71a9e18f7dfd12 *src/s2/s2loop.cc +da57f0c9e0f4c2cdf2599e0de097dbd9 *src/s2/s2loop.h +2f78e2f1e0911445a0ffe5e5b0fe014a *src/s2/s2loop_measures.cc +eef153a4e7bb190f620c250f78ac7c1c *src/s2/s2loop_measures.h +2e54cb10538dee0beadcfcf774875b64 *src/s2/s2max_distance_targets.cc +ad2641cab77de9fe057639d5afdf4edd *src/s2/s2max_distance_targets.h +66d7e4a5a3700d4b5a8c7e013fcfc370 *src/s2/s2measures.cc +b230afb1b144cfe2dfa5241d0d935cc1 *src/s2/s2measures.h +b70ef8989674552ca3d0e0c9fcb754a8 *src/s2/s2metrics.cc +c974a90f9d2ed8199af139ad33147aa4 *src/s2/s2metrics.h +582857ac635543b496cf93ee1bbe133b *src/s2/s2min_distance_targets.cc +49d8a47ed9a0f58dd97b2258b1764e77 *src/s2/s2min_distance_targets.h +94bd23bc709062647f0bdc993dfc7933 *src/s2/s2padded_cell.cc +b2841070cce458abff94905776f74eae *src/s2/s2padded_cell.h +af4bfede404056bb6ecc81a6ec4ed141 *src/s2/s2point.h +caf1d4962bb43ca5e3bfdc9d17640b49 *src/s2/s2point_compression.cc +0261a1c5b0ed5a8fec6b797ec53b5f30 *src/s2/s2point_compression.h +127b57a24cfa7ea2c78bc0fb7c9f446b *src/s2/s2point_index.h +7b88d7113b91cd8d5f0c1ad5bbacc7c4 *src/s2/s2point_region.cc +1c40d79d4102f56ec2301b62d2767a0e *src/s2/s2point_region.h +291c8ec7baef072cb29bc341fb3cb5c1 *src/s2/s2point_span.h +7d75a6945e15b9c78a713a59bd85f199 *src/s2/s2point_vector_shape.h +0475a05e50a35ca03b06e084178cd31a *src/s2/s2pointutil.cc +5d2100a54f4302ee75608d48d586fe3c *src/s2/s2pointutil.h +3272d4e15adf23b8b98b7fe50ca6e43a *src/s2/s2polygon.cc +dad5e0faba176fe983c613c3d6a60d07 *src/s2/s2polygon.h +12ef96f03b504ccf5c7523d15448b882 *src/s2/s2polyline.cc +16ea03d777619f1e59a47cbf83d53071 *src/s2/s2polyline.h +2914934120ec55346427958e5f73fb5e *src/s2/s2polyline_alignment.cc +c0f9b77a9439bc0c84d9dfe2dcbcc25b *src/s2/s2polyline_alignment.h +8a2d7dedff6ee98faf1edcc694a07a19 *src/s2/s2polyline_alignment_internal.h +c206c2fa52e4a5588b1024f45c63595b *src/s2/s2polyline_measures.cc +3b78b40b711bc148a7cc8f01f044d9a7 *src/s2/s2polyline_measures.h +eab00b14b7643e320f8044521eaa16e4 *src/s2/s2polyline_simplifier.cc +a12b87a494ef6e98dd8b19091ebf8cd8 *src/s2/s2polyline_simplifier.h +ff7549a0d6195b34b24867fff4cc6d06 *src/s2/s2predicates.cc +3e580e8e5db3954a6918420df5d50630 *src/s2/s2predicates.h +7dd5ac042f3403c26f10e8bc6bfcc8ac *src/s2/s2predicates_internal.h +7478e1714d9dcaf520159458d10e383b *src/s2/s2projections.cc +fa7c158fb73fced84c86e73bc40a4215 *src/s2/s2projections.h +8baa87c45aa11148a5207f367237e753 *src/s2/s2r2rect.cc +eef1eb7446a79b5b6430af8eede7a698 *src/s2/s2r2rect.h +57778c7c84ae75e67ed68e76f9026ec0 *src/s2/s2region.cc +2234c6d0d865cc61ca6959c01e0015e4 *src/s2/s2region.h +e12692516fc6b52c237e16cf16f11e19 *src/s2/s2region_coverer.cc +fc15e353fdbb3c76de73bf5fa5a713d8 *src/s2/s2region_coverer.h +f622c31665d97c607e1cb5106c295959 *src/s2/s2region_intersection.cc +d146a90111cd2d2b33b1526d5d1eab9e *src/s2/s2region_intersection.h +4bedf29698cbce5a1d755b5d9f0e5312 *src/s2/s2region_term_indexer.cc +ba4411f3cbd9ffe23482ec22e2045b31 *src/s2/s2region_term_indexer.h +f8232c70a590ed11666c93181c9d7343 *src/s2/s2region_union.cc +7acd16e242bad3bf6f8651417f6940c2 *src/s2/s2region_union.h +9e0666e7c03402dbf71fec6da5aa9244 *src/s2/s2shape.h +25ac091ec0963d3216b86e86a050eae9 *src/s2/s2shape_index.cc +1b8ddf576bde4dc6647dcec8341a18a3 *src/s2/s2shape_index.h +906e1e16bc8aa9f5d4e99651a6cdf0fe *src/s2/s2shape_index_buffered_region.cc +2f6f557e792611c362fee453ca16eb88 *src/s2/s2shape_index_buffered_region.h +c0543285953743a9605aa3ef63ad6127 *src/s2/s2shape_index_measures.cc +1caf918a11c8bafc0835861870fbd0ea *src/s2/s2shape_index_measures.h +b095a57fa9ad2adfa3cc31f6b8d6fffd *src/s2/s2shape_index_region.h +23c94c388142e68594cbf29b677d2bbc *src/s2/s2shape_measures.cc +ce9d81e0539f9c3a136e07d12c46cb05 *src/s2/s2shape_measures.h +c5794119535ecf13d27f5ae34c58f8b0 *src/s2/s2shapeutil_build_polygon_boundaries.cc +36f49316d02ac4f5c1124d7a3365419f *src/s2/s2shapeutil_build_polygon_boundaries.h +2b6d1731185a9ff771d2f150cf6c66f2 *src/s2/s2shapeutil_coding.cc +3c7dde9ddf64c467729e08fb30741bbe *src/s2/s2shapeutil_coding.h +afd2720bf07ba0aa5f94dfe9ba0c4d3a *src/s2/s2shapeutil_contains_brute_force.cc +f6d89268f04abb12ccd63236d4362060 *src/s2/s2shapeutil_contains_brute_force.h +1344b756712c76ac8c390abdb5a5886e *src/s2/s2shapeutil_count_edges.h +a699c9e30b9f8eaea6b9308c099c2285 *src/s2/s2shapeutil_edge_iterator.cc +79c3c7d45f832a4b348116e5b9ec993d *src/s2/s2shapeutil_edge_iterator.h +ae2b2addb5cb94cd24c57c195e1b0b49 *src/s2/s2shapeutil_get_reference_point.cc +53160e1333206256c06775aceb56ff6e *src/s2/s2shapeutil_get_reference_point.h +6d1353bb32206db05048ee50a073c0e7 *src/s2/s2shapeutil_range_iterator.cc +d12c9628c73beb0f42bb462060facbe7 *src/s2/s2shapeutil_range_iterator.h +a249fc71604963665d21c311c0e1c391 *src/s2/s2shapeutil_shape_edge.h +bdf516e73992d5a15fdaa074596f8d6f *src/s2/s2shapeutil_shape_edge_id.h +007fda069d8144645da7f2c7462a8359 *src/s2/s2shapeutil_testing.h +2d43581ce7e9638ee9d4390016d422df *src/s2/s2shapeutil_visit_crossing_edge_pairs.cc +f21bfbc04ce2d1d819512176b45954bb *src/s2/s2shapeutil_visit_crossing_edge_pairs.h +489cf62e6b1d3e3ff20b67aa19975130 *src/s2/s2testing.cc +fd1b7e4602b2f79e66496c81232f9cb1 *src/s2/s2testing.h +b504a72745dbeab05a8f198831eebb7d *src/s2/s2text_format.cc +294d80e1f4a307f03d84e66ae11767a1 *src/s2/s2text_format.h +518f25738bd53f6439d3708f56a85435 *src/s2/s2wedge_relations.cc +3c1f74157144820e8e5008fdc433e018 *src/s2/s2wedge_relations.h +d3ffa54eb4cab5a70d23bee5cbc17b3e *src/s2/sequence_lexicon.h +9b6952846f960ce7c8f86d425a6cfc53 *src/s2/strings/ostringstream.cc +a7553d41936a6491b5e79bad88a676b9 *src/s2/strings/ostringstream.h +33d32d088678a9117a40a020df05e8fc *src/s2/strings/serialize.cc +ae996ef21b8b910fc64050493a0595ae *src/s2/strings/serialize.h +c90e431dac905cc9e596e1b1071b7ea1 *src/s2/util/bits/bit-interleave.cc +3bfcb45aa9bfa89827d30afba984f79c *src/s2/util/bits/bit-interleave.h +ce0ab9919f6091c79c677b39c23f8b00 *src/s2/util/bits/bits.cc +37c9da7e1a7d27c20ca04a3a8ef559ec *src/s2/util/bits/bits.h +c0d70edbf2ac3f698c17d89959e326f9 *src/s2/util/coding/coder.cc +92ada6509b11d1a90ec200c2adf9e0dd *src/s2/util/coding/coder.h +c4fe67fd8a294e03bfc46ce9fb1134d1 *src/s2/util/coding/nth-derivative.h +2e711fab16d3ec015ecc12d522b8b0ae *src/s2/util/coding/transforms.h +90d94cda2cd92a02c4851ebb93d7453e *src/s2/util/coding/varint.cc +9a4e580abf363f1e0bb4bded873514cd *src/s2/util/coding/varint.h +9e59dc6aad536cbce24eedfb658d8716 *src/s2/util/endian/endian.h +4525710cda51323baa3327ac57ed3106 *src/s2/util/gtl/btree.h +118e86e589e0fae520e30c18dc1f3e08 *src/s2/util/gtl/btree_container.h +52e6fdc3fa81363e043d7e5a35a697ef *src/s2/util/gtl/btree_map.h +81e2de007193adce24cd9d5f5739bf5e *src/s2/util/gtl/btree_set.h +d3ee6aee2a2c316240faa7a33434a23e *src/s2/util/gtl/compact_array.h +09f417c582f54941a864fcbae7d93397 *src/s2/util/gtl/container_logging.h +6868f5ef7082e686952ee034c62a6cce *src/s2/util/gtl/dense_hash_set.h +e6d5083ac705a0300e77fe096a7a0589 *src/s2/util/gtl/densehashtable.h +fa56660fc40be26b3dbfc7115c56674e *src/s2/util/gtl/hashtable_common.h +8e531532de42f1c1f7cd4adb553fe486 *src/s2/util/gtl/layout.h +8c80e94bcdfb52740a0b8c0561845642 *src/s2/util/gtl/legacy_random_shuffle.h +a36fbccb7e3c064e5a16b4ac1e7842f3 *src/s2/util/hash/mix.h +a77a8f0d9d512bbec62949a0b3db20aa *src/s2/util/math/exactfloat/exactfloat.cc +828d0024dcc62675ab2be8b6f3cc3f93 *src/s2/util/math/exactfloat/exactfloat.h +2037118f434c6f41bc8014de2eb18c92 *src/s2/util/math/mathutil.cc +0c3be461d027e0cccb9c0facb12cd046 *src/s2/util/math/mathutil.h +160830f8d1d1983f1b773f953abff72d *src/s2/util/math/matrix3x3.h +c07f92fe336b33334c46895999a4c62a *src/s2/util/math/vector.h +5b091051cad1329e1e2b535e40e9c142 *src/s2/util/math/vector3_hash.h +2aba55dd12417d944cd91a2f02b1d946 *src/s2/util/units/length-units.cc +1bdf37d9bef7eba2b40d5baad2403f8d *src/s2/util/units/length-units.h +bdfd79c04de7e7893d9cadb9dd9d2f63 *src/s2/util/units/physical-units.h +400af05643ac08d897c2100a0a68f1aa *src/s2/value_lexicon.h +f5e89fcca3521af3ae1984cf3b988541 *src/s2geography.h +d8287075eb3b5a0d59d684a3bbc5b5ba *src/s2geography/accessors-geog.cc +10c803af1a5b075cb974f138658b1f41 *src/s2geography/accessors-geog.h +8e45d5fd12d368bb2989b24b6e158da4 *src/s2geography/accessors.cc +3348390f6d77133281e54b1e8a29c45b *src/s2geography/accessors.h +a3e670ac87785c083206586c8f3f4ca6 *src/s2geography/aggregator.h +78fcf83ac1350f041d585478dd0cc1e0 *src/s2geography/build.cc +fed04d2f5ddae951483be94e78dd8d46 *src/s2geography/build.h +7e63d2ff16c15445e287995ce9376d3a *src/s2geography/constructor.h +26d396867ab4214aaa9c811a3f5076e0 *src/s2geography/coverings.cc +9245d1de9f44acb9dc6f2ea03324761b *src/s2geography/coverings.h +93135f5c6f3755e0b4d2c2c0e40776ed *src/s2geography/distance.cc +ac0bf08dcc4f407205a4d78007006e02 *src/s2geography/distance.h +cb5e8ca81a2ad12cd88d1cc88a2ec00b *src/s2geography/geoarrow-imports.h +be903b4618b334e46916d7c56a3b4cf0 *src/s2geography/geography.cc +f705c43a8173f1b54eb7d391004654c5 *src/s2geography/geography.h +275d807d19673e87935f24d914697a7b *src/s2geography/index.h +520c06d0ec1884663e62ba6e7b183f49 *src/s2geography/linear-referencing.cc +ad9e7623296a977bc6b38f6acd1e29b2 *src/s2geography/linear-referencing.h +d8d87d7c6b89d370f45398ce0bc4b545 *src/s2geography/predicates.cc +a8a420182aad78038f2f87e8b8e0539d *src/s2geography/predicates.h +6622cfeaf8b569e1dbb71209addae76f *src/tests/main.c +4495750656420f7cd041bf0bf27ec020 *src/tests/soname.h +3ff04996080488941707e8fd75d50d02 *src/wk-impl.c +a9ffa55303a6dcba9c5c9c9d322e2009 *tests/area.R +960ec65d25d1018d2084d81ed3324fdd *tests/area.Rout +9a6a22f7e00fd5147781e07d2edd589e *tests/testthat.R +755c4ca0ad5e6f558b9eb7c7e04ded6e *tests/testthat/test-data.R +d8d89b5827614edf67567d7e15e2f213 *tests/testthat/test-plot.R +ba1747c0aca10f3293809487570aa4f4 *tests/testthat/test-s2-accessors.R +6d272b0f1b19c173c19ff225c361cd31 *tests/testthat/test-s2-bounds.R +5c3476c24589151b804b66580c20e9b8 *tests/testthat/test-s2-cell-union.R +adfef75ba2c27c1512e1509dc9949904 *tests/testthat/test-s2-cell.R +2fb5fbb6a8d7d4d13c80c650380a2763 *tests/testthat/test-s2-constructors-formatters.R +532d31b6373a11733f72fda2b4254ef6 *tests/testthat/test-s2-earth.R +6b32984296b69c08ad6f647a13ec683a *tests/testthat/test-s2-geography.R +3f9c9f8b1ee1767381c2f6bde18c35f3 *tests/testthat/test-s2-lnglat.R +b874f311928c98a3056d5628c55422c8 *tests/testthat/test-s2-matrix.R +97bdafc52d85952cf4d72ef1818ba36c *tests/testthat/test-s2-options.R +b8d52c633fa3d389a46c8198804f9757 *tests/testthat/test-s2-point.R +6a0b51e1b62caf1dc1ce91d662c9a424 *tests/testthat/test-s2-predicates.R +5ef809d341a4b5203d675644845e6e94 *tests/testthat/test-s2-transformers.R +a7de1183f03382626ca2b46bf1c4e781 *tests/testthat/test-utils.R +2c9ecaae7d9fe889868e0c69bfecc42c *tests/testthat/test-vctrs.R +eb4ce45137cf0d61472a37c5f493d59b *tests/testthat/test-wk-utils.R +0ec27c181a66ce5b72bc8c3940e9e00e *tools/version.c +497dfd3cbf113981c70269ce4b31c496 *tools/winlibs.R diff --git a/NAMESPACE b/NAMESPACE new file mode 100644 index 0000000..f1213d1 --- /dev/null +++ b/NAMESPACE @@ -0,0 +1,205 @@ +# Generated by roxygen2: do not edit by hand + +S3method("[<-",s2_cell) +S3method("[<-",s2_geography) +S3method("[[<-",s2_cell) +S3method("[[<-",s2_geography) +S3method(Math,s2_cell) +S3method(Ops,s2_cell) +S3method(Summary,s2_cell) +S3method(as.character,s2_cell) +S3method(as.character,s2_cell_union) +S3method(as.character,s2_geography) +S3method(as.list,s2_cell) +S3method(as_s2_cell,character) +S3method(as_s2_cell,s2_cell) +S3method(as_s2_cell,s2_geography) +S3method(as_s2_cell,wk_xy) +S3method(as_s2_cell_union,character) +S3method(as_s2_cell_union,s2_cell) +S3method(as_s2_cell_union,s2_cell_union) +S3method(as_s2_geography,WKB) +S3method(as_s2_geography,blob) +S3method(as_s2_geography,character) +S3method(as_s2_geography,logical) +S3method(as_s2_geography,s2_cell_union) +S3method(as_s2_geography,s2_geography) +S3method(as_s2_geography,wk_wkb) +S3method(as_s2_geography,wk_wkt) +S3method(as_s2_geography,wk_xy) +S3method(as_s2_lnglat,character) +S3method(as_s2_lnglat,default) +S3method(as_s2_lnglat,wk_xy) +S3method(as_s2_lnglat,wk_xyz) +S3method(as_s2_point,character) +S3method(as_s2_point,default) +S3method(as_s2_point,wk_xy) +S3method(as_s2_point,wk_xyz) +S3method(as_wkb,s2_geography) +S3method(as_wkt,s2_geography) +S3method(format,s2_cell) +S3method(format,s2_cell_union) +S3method(format,s2_geography) +S3method(format,s2_point_crs) +S3method(is.na,s2_cell) +S3method(is.na,s2_cell_union) +S3method(is.na,s2_geography) +S3method(is.numeric,s2_cell) +S3method(plot,s2_cell) +S3method(plot,s2_cell_union) +S3method(plot,s2_geography) +S3method(print,s2_cell_union) +S3method(sort,s2_cell) +S3method(str,s2_cell_union) +S3method(unique,s2_cell) +S3method(unlist,s2_cell_union) +S3method(wk_crs,s2_geography) +S3method(wk_handle,s2_geography) +S3method(wk_is_geodesic,s2_geography) +S3method(wk_set_crs,s2_geography) +S3method(wk_set_geodesic,s2_geography) +S3method(wk_writer,s2_geography) +export(as_s2_cell) +export(as_s2_cell_union) +export(as_s2_geography) +export(as_s2_lnglat) +export(as_s2_point) +export(new_s2_cell) +export(s2_area) +export(s2_as_binary) +export(s2_as_text) +export(s2_boundary) +export(s2_bounds_cap) +export(s2_bounds_rect) +export(s2_buffer_cells) +export(s2_cell) +export(s2_cell_area) +export(s2_cell_area_approx) +export(s2_cell_boundary) +export(s2_cell_center) +export(s2_cell_child) +export(s2_cell_common_ancestor_level) +export(s2_cell_common_ancestor_level_agg) +export(s2_cell_contains) +export(s2_cell_debug_string) +export(s2_cell_distance) +export(s2_cell_edge_neighbour) +export(s2_cell_invalid) +export(s2_cell_is_face) +export(s2_cell_is_leaf) +export(s2_cell_is_valid) +export(s2_cell_level) +export(s2_cell_max_distance) +export(s2_cell_may_intersect) +export(s2_cell_parent) +export(s2_cell_polygon) +export(s2_cell_sentinel) +export(s2_cell_to_lnglat) +export(s2_cell_union) +export(s2_cell_union_contains) +export(s2_cell_union_difference) +export(s2_cell_union_intersection) +export(s2_cell_union_intersects) +export(s2_cell_union_normalize) +export(s2_cell_union_union) +export(s2_cell_vertex) +export(s2_centroid) +export(s2_centroid_agg) +export(s2_closest_edges) +export(s2_closest_feature) +export(s2_closest_point) +export(s2_contains) +export(s2_contains_matrix) +export(s2_convex_hull) +export(s2_convex_hull_agg) +export(s2_coverage_union_agg) +export(s2_covered_by) +export(s2_covered_by_matrix) +export(s2_covering_cell_ids) +export(s2_covering_cell_ids_agg) +export(s2_covers) +export(s2_covers_matrix) +export(s2_data_cities) +export(s2_data_countries) +export(s2_data_timezones) +export(s2_difference) +export(s2_dimension) +export(s2_disjoint) +export(s2_disjoint_matrix) +export(s2_distance) +export(s2_distance_matrix) +export(s2_dwithin) +export(s2_dwithin_matrix) +export(s2_earth_radius_meters) +export(s2_equals) +export(s2_equals_matrix) +export(s2_farthest_feature) +export(s2_geog_from_text) +export(s2_geog_from_wkb) +export(s2_geog_point) +export(s2_geography) +export(s2_geography_writer) +export(s2_hemisphere) +export(s2_interpolate) +export(s2_interpolate_normalized) +export(s2_intersection) +export(s2_intersects) +export(s2_intersects_box) +export(s2_intersects_matrix) +export(s2_is_collection) +export(s2_is_empty) +export(s2_is_valid) +export(s2_is_valid_detail) +export(s2_length) +export(s2_lnglat) +export(s2_make_line) +export(s2_make_polygon) +export(s2_max_distance) +export(s2_max_distance_matrix) +export(s2_may_intersect_matrix) +export(s2_minimum_clearance_line_between) +export(s2_num_points) +export(s2_options) +export(s2_perimeter) +export(s2_plot) +export(s2_point) +export(s2_point_crs) +export(s2_point_on_surface) +export(s2_prepared_dwithin) +export(s2_project) +export(s2_project_normalized) +export(s2_projection_mercator) +export(s2_projection_orthographic) +export(s2_projection_plate_carree) +export(s2_rebuild) +export(s2_rebuild_agg) +export(s2_simplify) +export(s2_snap_distance) +export(s2_snap_identity) +export(s2_snap_level) +export(s2_snap_precision) +export(s2_snap_to_grid) +export(s2_sym_difference) +export(s2_tessellate_tol_default) +export(s2_touches) +export(s2_touches_matrix) +export(s2_trans_lnglat) +export(s2_trans_point) +export(s2_union) +export(s2_union_agg) +export(s2_within) +export(s2_within_matrix) +export(s2_world_plate_carree) +export(s2_x) +export(s2_y) +importFrom(Rcpp,sourceCpp) +importFrom(utils,str) +importFrom(wk,as_wkb) +importFrom(wk,as_wkt) +importFrom(wk,wk_crs) +importFrom(wk,wk_handle) +importFrom(wk,wk_is_geodesic) +importFrom(wk,wk_set_crs) +importFrom(wk,wk_set_geodesic) +importFrom(wk,wk_writer) +useDynLib(s2, .registration = TRUE) diff --git a/NEWS.md b/NEWS.md new file mode 100644 index 0000000..28dbbb0 --- /dev/null +++ b/NEWS.md @@ -0,0 +1,131 @@ +# s2 1.1.2 + +- Fixed test for `as.data.frame()` for `s2_cell()` to comply with new wk + version and the latest release of R (#207). +- Fix unary union of an empty multipolygon (#208). +- Added `#include ` to an Abseil header to fix compilation with + gcc13 (#209, #210). +- Update internal Abseil to 20220623.1 LTS (#213). + +# s2 1.1.1 + +- Fix new CRAN check warnings (#202, #203). + +# s2 1.1.0 + +- Fix for s2 build on Windows with R <= 3.6.x (#142) +- Fix for s2 build on MacOS with multiple openssl versions (#142, #145, #146) +- Fix for s2 build on 32-bit openssl (#143, #147) +- Added `s2_convex_hull()` and `s2_convex_hull_agg()` (@spiry34, #150, + #151, #163). +- Added `max_distance` argument to `s2_closest_edges()`, making + distance-constrained k-nearest neighbours possible (#125, #156, #162). +- Added a spherical `s2_point_on_surface()` implementation for polygons + (@kylebutts, #152, #161) +- Added a `s2_cell_union()` vector class to represent cell coverings and + operators to generate them from an s2 geography vector (e.g., + `s2_covering_cell_ids()`). Cell unions are useful as compact representations + of spherical geometry and can be used like a bounding box to determine + a possible intersection with one or more geographies (#85, #94, #164). +- Refactored the simple features compatability layer into a standalone + code base for potential future use in a Python adaptation (#165). +- Migrate input and output to non-deprecated wk package handlers and writers + (#101, #165, #168). +- Make `s2_union_agg()` more efficient using a recursive merge strategy + (#103, #165). +- Fix package build on Raspberry Pi (#169, #171). +- Fix warning on clang14 when compiling with `-O0` (#167, #172). +- Added `s2_prepared_dwithin()` and fixed `s2_dwithin_matrix()` such that it + efficiently uses the index (#157, #174). +- Updated `s2_lnglat()` and `s2_point()` to use `wk::xy()` (a record-style + vctr) to represent point coordinates. This is much faster than the previous + representation which relied on `list()` of external pointers (#181, #159). +- Added arguments `planar` and `tessellate_tol_m` to `s2_as_text()`, + `s2_as_binary()`. Use `planar = TRUE` and set `tessellate_tol_m` to the + maximum error for your use-case to automatically subdivide edges to + preserve or "straight" lines in Plate carree projection on import (#182). +- Added arguments `planar` and `tessellate_tol_m` to `s2_geog_from_text()`, and + `s2_geog_from_wkb()`. Use `planar = TRUE` and set `tessellate_tol_m` to the + maximum error for your use-case to automatically subdivide edges to + ensure or "straight" lines in Plate carree projection on export (#182). + +# s2 1.0.7 + +- Update the internal copy of s2geometry to use updated Abseil, + fixing a compiler warning on gcc-11 (#79, #134). + +# s2 1.0.6 + +- Added support for `STRICT_R_HEADERS` (@eddelbuettel, #118). +- Fixed a bug where the result of `s2_centroid_agg()` did not + behave like a normal point in distance calculations (#119, #121). +- Fixed a Windows UCRT check failure and updated openssl linking + (@jeroen, #122). + +# s2 1.0.5 + +* Added `s2_projection_filter()` and `s2_unprojection_filter()` to + expose the S2 edge tessellator, which can be used to make Cartesian + or great circle assumptions of line segments explicit by adding + points where necessary (#115). +* Added an `s2_cell()` vector class to expose a subset of the S2 + indexing system to R users (#85, #114). +* Added `s2_closest_edges()` to make k-nearest neighbours calculation + possible on the sphere (#111, #112). +* Added `s2_interpolate()`, `s2_interpolate_normalized()`, + `s2_project()`, and `s2_project_normalized()` to provide linear + referencing support on the sphere (#96, #110). +* Fixed import of empty points from WKB (#109). +* Added argument `dimensions` to `s2_options()` to constrain the + output dimensions of a boolean or rebuild operation (#105, #104, #110). +* Added `s2_is_valid()` and `s2_is_valid_detail()` to help find invalid + spherical geometries when importing data into S2 (#100). +* Improved error messages when importing and processing data such that + errors can be debugged more readily (#100, #98). +* The unary version of `s2_union()` can now handle MULTIPOLYGON + geometries with overlapping rings in addition to other invalid + polygons. `s2_union()` can now sanitize + almost any input to be valid spherical geometry with + minimal modification (#100, #99). +* Renamed the existing implementation of `s2_union_agg()` to + `s2_coverage_union_agg()` to make clear that the function only + works when the individual geometries do not have overlapping + interiors. `s2_union_agg()` was replaced with a + true aggregate union that can handle unions of most geometries + (#100, #97). +* Added `s2_rebuild_agg()` to match `s2_union_agg()`. Like + `s2_rebuild()`, `s2_rebuild_agg()` collects the edges in the input + and builds them into a feature, optionally snapping or simplifying + vertices in the process (#100). + +# s2 1.0.4 + +* Fixed errors that resulted from compilation on clang 12.2 (#88, #89). + +# s2 1.0.3 + +* Fixed CRAN check errors (#80). + +# s2 1.0.2 + +* Fixed CRAN check errors (#71, #75, #72). + +# s2 1.0.1 + +* Added layer creation options to `s2_options()`, which now uses strings + rather than numeric codes to specify boolean operation options, geography + construction options, and builder options (#70). +* Added `s2_rebuild()` and `s2_simplify()`, which wrap the S2 C++ `S2Builder` + class to provide simplification and fixing of invalid geographies (#70). +* The s2 package now builds and passes the CMD check on Solaris (#66, #67). +* Renamed `s2_latlng()` to `s2_lnglat()` to keep axis order consistent + throughout the package (#69). +* Added `s2_bounds_cap()` and `s2_bounds_rect()` to compute bounding areas + using geographic coordinates (@edzer, #63). +* `s2_*_matrix()` predicates now efficiently use indexing to compute the + results of many predicate comparisons (#61). + +# s2 1.0.0 + +This version is a complete rewrite of the former s2 CRAN package, entirely +backwards incompatible with previous versions. diff --git a/R/RcppExports.R b/R/RcppExports.R new file mode 100644 index 0000000..bb373a3 --- /dev/null +++ b/R/RcppExports.R @@ -0,0 +1,447 @@ +# Generated by using Rcpp::compileAttributes() -> do not edit by hand +# Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393 + +cpp_s2_init <- function() { + invisible(.Call(`_s2_cpp_s2_init`)) +} + +cpp_s2_is_collection <- function(geog) { + .Call(`_s2_cpp_s2_is_collection`, geog) +} + +cpp_s2_is_valid <- function(geog) { + .Call(`_s2_cpp_s2_is_valid`, geog) +} + +cpp_s2_is_valid_reason <- function(geog) { + .Call(`_s2_cpp_s2_is_valid_reason`, geog) +} + +cpp_s2_dimension <- function(geog) { + .Call(`_s2_cpp_s2_dimension`, geog) +} + +cpp_s2_num_points <- function(geog) { + .Call(`_s2_cpp_s2_num_points`, geog) +} + +cpp_s2_is_empty <- function(geog) { + .Call(`_s2_cpp_s2_is_empty`, geog) +} + +cpp_s2_area <- function(geog) { + .Call(`_s2_cpp_s2_area`, geog) +} + +cpp_s2_length <- function(geog) { + .Call(`_s2_cpp_s2_length`, geog) +} + +cpp_s2_perimeter <- function(geog) { + .Call(`_s2_cpp_s2_perimeter`, geog) +} + +cpp_s2_x <- function(geog) { + .Call(`_s2_cpp_s2_x`, geog) +} + +cpp_s2_y <- function(geog) { + .Call(`_s2_cpp_s2_y`, geog) +} + +cpp_s2_project_normalized <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_project_normalized`, geog1, geog2) +} + +cpp_s2_distance <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_distance`, geog1, geog2) +} + +cpp_s2_max_distance <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_max_distance`, geog1, geog2) +} + +cpp_s2_bounds_cap <- function(geog) { + .Call(`_s2_cpp_s2_bounds_cap`, geog) +} + +cpp_s2_bounds_rect <- function(geog) { + .Call(`_s2_cpp_s2_bounds_rect`, geog) +} + +cpp_s2_cell_union_normalize <- function(cellUnionVector) { + .Call(`_s2_cpp_s2_cell_union_normalize`, cellUnionVector) +} + +cpp_s2_cell_union_is_na <- function(cellUnionVector) { + .Call(`_s2_cpp_s2_cell_union_is_na`, cellUnionVector) +} + +cpp_s2_cell_union_contains <- function(cellUnionVector1, cellUnionVector2) { + .Call(`_s2_cpp_s2_cell_union_contains`, cellUnionVector1, cellUnionVector2) +} + +cpp_s2_cell_union_contains_cell <- function(cellUnionVector, cellIdVector) { + .Call(`_s2_cpp_s2_cell_union_contains_cell`, cellUnionVector, cellIdVector) +} + +cpp_s2_cell_union_intersects <- function(cellUnionVector1, cellUnionVector2) { + .Call(`_s2_cpp_s2_cell_union_intersects`, cellUnionVector1, cellUnionVector2) +} + +cpp_s2_cell_union_intersection <- function(cellUnionVector1, cellUnionVector2) { + .Call(`_s2_cpp_s2_cell_union_intersection`, cellUnionVector1, cellUnionVector2) +} + +cpp_s2_cell_union_union <- function(cellUnionVector1, cellUnionVector2) { + .Call(`_s2_cpp_s2_cell_union_union`, cellUnionVector1, cellUnionVector2) +} + +cpp_s2_cell_union_difference <- function(cellUnionVector1, cellUnionVector2) { + .Call(`_s2_cpp_s2_cell_union_difference`, cellUnionVector1, cellUnionVector2) +} + +cpp_s2_geography_from_cell_union <- function(cellUnionVector) { + .Call(`_s2_cpp_s2_geography_from_cell_union`, cellUnionVector) +} + +cpp_s2_covering_cell_ids <- function(geog, min_level, max_level, max_cells, buffer, interior) { + .Call(`_s2_cpp_s2_covering_cell_ids`, geog, min_level, max_level, max_cells, buffer, interior) +} + +cpp_s2_covering_cell_ids_agg <- function(geog, min_level, max_level, max_cells, buffer, interior, naRm) { + .Call(`_s2_cpp_s2_covering_cell_ids_agg`, geog, min_level, max_level, max_cells, buffer, interior, naRm) +} + +cpp_s2_cell_sentinel <- function() { + .Call(`_s2_cpp_s2_cell_sentinel`) +} + +cpp_s2_cell_from_string <- function(cellString) { + .Call(`_s2_cpp_s2_cell_from_string`, cellString) +} + +cpp_s2_cell_from_lnglat <- function(lnglat) { + .Call(`_s2_cpp_s2_cell_from_lnglat`, lnglat) +} + +cpp_s2_cell_to_lnglat <- function(cellId) { + .Call(`_s2_cpp_s2_cell_to_lnglat`, cellId) +} + +cpp_s2_cell_to_cell_union <- function(cellId) { + .Call(`_s2_cpp_s2_cell_to_cell_union`, cellId) +} + +cpp_s2_cell_is_na <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_is_na`, cellIdVector) +} + +cpp_s2_cell_sort <- function(cellIdVector, decreasing) { + .Call(`_s2_cpp_s2_cell_sort`, cellIdVector, decreasing) +} + +cpp_s2_cell_range <- function(cellIdVector, naRm) { + .Call(`_s2_cpp_s2_cell_range`, cellIdVector, naRm) +} + +cpp_s2_cell_unique <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_unique`, cellIdVector) +} + +cpp_s2_cell_to_string <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_to_string`, cellIdVector) +} + +cpp_s2_cell_debug_string <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_debug_string`, cellIdVector) +} + +cpp_s2_cell_is_valid <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_is_valid`, cellIdVector) +} + +cpp_s2_cell_center <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_center`, cellIdVector) +} + +cpp_s2_cell_polygon <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_polygon`, cellIdVector) +} + +cpp_s2_cell_vertex <- function(cellIdVector, k) { + .Call(`_s2_cpp_s2_cell_vertex`, cellIdVector, k) +} + +cpp_s2_cell_level <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_level`, cellIdVector) +} + +cpp_s2_cell_area <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_area`, cellIdVector) +} + +cpp_s2_cell_area_approx <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_area_approx`, cellIdVector) +} + +cpp_s2_cell_parent <- function(cellIdVector, level) { + .Call(`_s2_cpp_s2_cell_parent`, cellIdVector, level) +} + +cpp_s2_cell_child <- function(cellIdVector, k) { + .Call(`_s2_cpp_s2_cell_child`, cellIdVector, k) +} + +cpp_s2_cell_edge_neighbour <- function(cellIdVector, k) { + .Call(`_s2_cpp_s2_cell_edge_neighbour`, cellIdVector, k) +} + +cpp_s2_cell_cummax <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_cummax`, cellIdVector) +} + +cpp_s2_cell_cummin <- function(cellIdVector) { + .Call(`_s2_cpp_s2_cell_cummin`, cellIdVector) +} + +cpp_s2_cell_eq <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_eq`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_neq <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_neq`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_lt <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_lt`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_lte <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_lte`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_gte <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_gte`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_gt <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_gt`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_contains <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_contains`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_may_intersect <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_may_intersect`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_distance <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_distance`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_max_distance <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_max_distance`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_common_ancestor_level <- function(cellIdVector1, cellIdVector2) { + .Call(`_s2_cpp_s2_cell_common_ancestor_level`, cellIdVector1, cellIdVector2) +} + +cpp_s2_cell_common_ancestor_level_agg <- function(cellId) { + .Call(`_s2_cpp_s2_cell_common_ancestor_level_agg`, cellId) +} + +s2_geography_full <- function(x) { + .Call(`_s2_s2_geography_full`, x) +} + +cpp_s2_geography_is_na <- function(geog) { + .Call(`_s2_cpp_s2_geography_is_na`, geog) +} + +s2_lnglat_from_s2_point <- function(s2_point) { + .Call(`_s2_s2_lnglat_from_s2_point`, s2_point) +} + +s2_point_from_s2_lnglat <- function(s2_lnglat) { + .Call(`_s2_s2_point_from_s2_lnglat`, s2_lnglat) +} + +cpp_s2_closest_feature <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_closest_feature`, geog1, geog2) +} + +cpp_s2_farthest_feature <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_farthest_feature`, geog1, geog2) +} + +cpp_s2_closest_edges <- function(geog1, geog2, n, min_distance, max_distance) { + .Call(`_s2_cpp_s2_closest_edges`, geog1, geog2, n, min_distance, max_distance) +} + +cpp_s2_may_intersect_matrix <- function(geog1, geog2, maxEdgesPerCell, maxFeatureCells, s2options) { + .Call(`_s2_cpp_s2_may_intersect_matrix`, geog1, geog2, maxEdgesPerCell, maxFeatureCells, s2options) +} + +cpp_s2_contains_matrix <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_contains_matrix`, geog1, geog2, s2options) +} + +cpp_s2_within_matrix <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_within_matrix`, geog1, geog2, s2options) +} + +cpp_s2_intersects_matrix <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_intersects_matrix`, geog1, geog2, s2options) +} + +cpp_s2_equals_matrix <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_equals_matrix`, geog1, geog2, s2options) +} + +cpp_s2_touches_matrix <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_touches_matrix`, geog1, geog2, s2options) +} + +cpp_s2_dwithin_matrix <- function(geog1, geog2, distance) { + .Call(`_s2_cpp_s2_dwithin_matrix`, geog1, geog2, distance) +} + +cpp_s2_distance_matrix <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_distance_matrix`, geog1, geog2) +} + +cpp_s2_max_distance_matrix <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_max_distance_matrix`, geog1, geog2) +} + +cpp_s2_contains_matrix_brute_force <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_contains_matrix_brute_force`, geog1, geog2, s2options) +} + +cpp_s2_within_matrix_brute_force <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_within_matrix_brute_force`, geog1, geog2, s2options) +} + +cpp_s2_intersects_matrix_brute_force <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_intersects_matrix_brute_force`, geog1, geog2, s2options) +} + +cpp_s2_disjoint_matrix_brute_force <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_disjoint_matrix_brute_force`, geog1, geog2, s2options) +} + +cpp_s2_equals_matrix_brute_force <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_equals_matrix_brute_force`, geog1, geog2, s2options) +} + +cpp_s2_dwithin_matrix_brute_force <- function(geog1, geog2, distance) { + .Call(`_s2_cpp_s2_dwithin_matrix_brute_force`, geog1, geog2, distance) +} + +cpp_s2_intersects <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_intersects`, geog1, geog2, s2options) +} + +cpp_s2_equals <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_equals`, geog1, geog2, s2options) +} + +cpp_s2_contains <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_contains`, geog1, geog2, s2options) +} + +cpp_s2_touches <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_touches`, geog1, geog2, s2options) +} + +cpp_s2_dwithin <- function(geog1, geog2, distance) { + .Call(`_s2_cpp_s2_dwithin`, geog1, geog2, distance) +} + +cpp_s2_prepared_dwithin <- function(geog1, geog2, distance) { + .Call(`_s2_cpp_s2_prepared_dwithin`, geog1, geog2, distance) +} + +cpp_s2_intersects_box <- function(geog, lng1, lat1, lng2, lat2, detail, s2options) { + .Call(`_s2_cpp_s2_intersects_box`, geog, lng1, lat1, lng2, lat2, detail, s2options) +} + +cpp_s2_intersection <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_intersection`, geog1, geog2, s2options) +} + +cpp_s2_union <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_union`, geog1, geog2, s2options) +} + +cpp_s2_difference <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_difference`, geog1, geog2, s2options) +} + +cpp_s2_sym_difference <- function(geog1, geog2, s2options) { + .Call(`_s2_cpp_s2_sym_difference`, geog1, geog2, s2options) +} + +cpp_s2_coverage_union_agg <- function(geog, s2options, naRm) { + .Call(`_s2_cpp_s2_coverage_union_agg`, geog, s2options, naRm) +} + +cpp_s2_union_agg <- function(geog, s2options, naRm) { + .Call(`_s2_cpp_s2_union_agg`, geog, s2options, naRm) +} + +cpp_s2_centroid_agg <- function(geog, naRm) { + .Call(`_s2_cpp_s2_centroid_agg`, geog, naRm) +} + +cpp_s2_rebuild_agg <- function(geog, s2options, naRm) { + .Call(`_s2_cpp_s2_rebuild_agg`, geog, s2options, naRm) +} + +cpp_s2_closest_point <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_closest_point`, geog1, geog2) +} + +cpp_s2_minimum_clearance_line_between <- function(geog1, geog2) { + .Call(`_s2_cpp_s2_minimum_clearance_line_between`, geog1, geog2) +} + +cpp_s2_centroid <- function(geog) { + .Call(`_s2_cpp_s2_centroid`, geog) +} + +cpp_s2_point_on_surface <- function(geog) { + .Call(`_s2_cpp_s2_point_on_surface`, geog) +} + +cpp_s2_boundary <- function(geog) { + .Call(`_s2_cpp_s2_boundary`, geog) +} + +cpp_s2_rebuild <- function(geog, s2options) { + .Call(`_s2_cpp_s2_rebuild`, geog, s2options) +} + +cpp_s2_unary_union <- function(geog, s2options) { + .Call(`_s2_cpp_s2_unary_union`, geog, s2options) +} + +cpp_s2_interpolate_normalized <- function(geog, distanceNormalized) { + .Call(`_s2_cpp_s2_interpolate_normalized`, geog, distanceNormalized) +} + +cpp_s2_buffer_cells <- function(geog, distance, maxCells, minLevel) { + .Call(`_s2_cpp_s2_buffer_cells`, geog, distance, maxCells, minLevel) +} + +cpp_s2_convex_hull <- function(geog) { + .Call(`_s2_cpp_s2_convex_hull`, geog) +} + +cpp_s2_convex_hull_agg <- function(geog, naRm) { + .Call(`_s2_cpp_s2_convex_hull_agg`, geog, naRm) +} + diff --git a/R/data.R b/R/data.R new file mode 100644 index 0000000..23ee0ea --- /dev/null +++ b/R/data.R @@ -0,0 +1,79 @@ + +#' Low-resolution world boundaries, timezones, and cities +#' +#' Well-known binary versions of the [Natural Earth](https://www.naturalearthdata.com/) +#' low-resolution world boundaries and timezone boundaries. +#' +#' @param name The name of a country, continent, city, or `NULL` +#' for all features. +#' @param utc_offset_min,utc_offset_max Minimum and/or maximum timezone +#' offsets. +#' +#' @format A data.frame with columns `name` (character), and +#' `geometry` (wk_wkb) +#' @source [Natural Earth Data](https://www.naturalearthdata.com/) +#' @examples +#' head(s2_data_countries()) +#' s2_data_countries("Germany") +#' s2_data_countries("Europe") +#' +#' head(s2_data_timezones()) +#' s2_data_timezones(-4) +#' +#' head(s2_data_cities()) +#' s2_data_cities("Cairo") +#' +"s2_data_tbl_countries" + +#' @rdname s2_data_tbl_countries +"s2_data_tbl_timezones" + +#' @rdname s2_data_tbl_countries +"s2_data_tbl_cities" + +#' @rdname s2_data_tbl_countries +#' @export +s2_data_countries <- function(name = NULL) { + df <- s2::s2_data_tbl_countries + if (is.null(name)) { + wkb <- df$geometry + } else { + wkb <- df$geometry[(df$name %in% name) | (df$continent %in% name)] + } + + as_s2_geography(wkb) +} + +#' @rdname s2_data_tbl_countries +#' @export +s2_data_timezones <- function(utc_offset_min = NULL, utc_offset_max = utc_offset_min) { + df <- s2::s2_data_tbl_timezones + if (is.null(utc_offset_min)) { + wkb <- df$geometry + } else { + matches <- (df$utc_offset >= utc_offset_min) & (df$utc_offset <= utc_offset_max) + wkb <- df$geometry[matches] + } + + as_s2_geography(wkb) +} + +#' @rdname s2_data_tbl_countries +#' @export +s2_data_cities <- function(name = NULL) { + df <- s2::s2_data_tbl_cities + if (is.null(name)) { + wkb <- df$geometry + } else { + wkb <- df$geometry[df$name %in% name] + } + + as_s2_geography(wkb) +} + +#' Example Geometries +#' +#' These geometries are toy examples useful for testing various coordinate +#' shuffling operations in the s2 package. +#' +"s2_data_example_wkt" diff --git a/R/plot.R b/R/plot.R new file mode 100644 index 0000000..0117263 --- /dev/null +++ b/R/plot.R @@ -0,0 +1,158 @@ + +#' Plot S2 Geographies +#' +#' @inheritParams wk::wk_plot +#' @param plot_hemisphere Plot the outline of the earth +#' @param centre The longitude/latitude point of the centre of the +#' orthographic projection +#' @param simplify Use `FALSE` to skip the simplification step +#' +#' @return The input, invisibly +#' @export +#' +#' @examples +#' s2_plot(s2_data_countries()) +#' s2_plot(s2_data_cities(), add = TRUE) +#' +s2_plot <- function(x, ..., asp = 1, xlab = "", ylab = "", + rule = "evenodd", add = FALSE, + plot_hemisphere = FALSE, + simplify = TRUE, + centre = NULL) { + x <- as_s2_geography(x) + + if (add) { + last <- last_plot_env$centre + centre <- if (is.null(last)) s2_lnglat(0, 0) else last + } else if (is.null(centre)) { + centre <- s2_centroid_agg(x, na.rm = TRUE) + } + + centre <- as_s2_lnglat(centre) + projection <- s2_projection_orthographic(centre) + hemisphere_bounds_poly <- cap_to_polygon(centre, (pi / 2) - 1e-5) + + if (plot_hemisphere) { + bbox_projected <- wk::rct(-1, -1, 1, 1) + } else if (!add) { + bbox_projected <- wk::wk_handle( + x, + wk::wk_bbox_handler(), + s2_projection = projection, + s2_tessellate_tol = s2_tessellate_tol_default() + ) + } + + wk::wk_plot( + wk::xy(), + bbox = bbox_projected, + asp = asp, xlab = xlab, ylab = ylab, + add = add + ) + + if (!add) { + last_plot_env$centre <- centre + } + + if (plot_hemisphere) { + wk::wk_plot(wk::crc(0, 0, 1), add = TRUE) + } + + # estimate resolution. In user coords, this can be though of in radians + # (at the centre of the plot) + usr <- graphics::par("usr") + usr_x <- usr[1:2] + usr_y <- usr[3:4] + device_x <- graphics::grconvertX(usr_x, to = "device") + device_y <- graphics::grconvertY(usr_y, to = "device") + + scale_x <- diff(device_x) / diff(usr_x) + scale_y <- diff(device_y) / diff(usr_y) + scale <- min(abs(scale_x), abs(scale_y)) + resolution_usr_rad <- 0.25 / scale + + # limit output to dimensions in input + dimensions_in_input <- setdiff(unique(s2_dimension(x)), NA_character_) + + if (simplify) { + x_hemisphere <- s2_intersection( + x, + hemisphere_bounds_poly, + options = s2_options( + snap = s2_snap_distance(resolution_usr_rad), + snap_radius = resolution_usr_rad, + simplify_edge_chains = TRUE, + dimensions = c("point", "polyline", "polygon")[dimensions_in_input + 1] + ) + ) + } else { + x_hemisphere <- s2_intersection( + x, + hemisphere_bounds_poly, + options = s2_options( + dimensions = c("point", "polyline", "polygon")[dimensions_in_input + 1] + ) + ) + } + + x_hemisphere[s2_is_empty(x_hemisphere)] <- as_s2_geography("POINT EMPTY") + + x_hemisphere_planar <- wk::wk_handle( + x_hemisphere, + wk::wkb_writer(), + s2_projection = projection, + # if this is too small we can get a stack overflow since the + # tessellation is recursive + s2_tessellate_tol = max(0.002, resolution_usr_rad * 4) + ) + + wk::wk_plot( + x_hemisphere_planar, + ..., + rule = rule, + add = TRUE + ) + + invisible(x) +} + +#' @export +plot.s2_geography <- function(x, ...) { + s2_plot(x, ...) +} + +#' @export +plot.s2_cell_union <- function(x, ...) { + s2_plot(x, ...) + invisible(x) +} + +#' @export +plot.s2_cell <- function(x, ...) { + if (all(s2_cell_is_leaf(x), na.rm = TRUE)) { + s2_plot(s2_cell_center(x), ...) + } else { + s2_plot(s2_cell_polygon(x), ...) + } + + invisible(x) +} + +cap_to_polygon <- function(centre = s2_lnglat(0, 0), radius_rad) { + centre <- as_s2_lnglat(centre) + rad_proj <- sin(radius_rad) + points <- wk::xy( + c(0, rad_proj, 0, -rad_proj, 0), + c(rad_proj, 0, -rad_proj, 0, rad_proj) + ) + points_s2 <- wk::wk_handle( + points, + s2_geography_writer( + projection = s2_projection_orthographic(centre) + ) + ) + s2_make_polygon(s2_x(points_s2), s2_y(points_s2)) +} + +last_plot_env <- new.env(parent = emptyenv()) +last_plot_env$centre <- NULL diff --git a/R/s2-accessors.R b/R/s2-accessors.R new file mode 100644 index 0000000..961de37 --- /dev/null +++ b/R/s2-accessors.R @@ -0,0 +1,169 @@ + +#' S2 Geography Accessors +#' +#' Accessors extract information about [geography vectors][as_s2_geography]. +#' +#' @param x,y [geography vectors][as_s2_geography]. These inputs +#' are passed to [as_s2_geography()], so you can pass other objects +#' (e.g., character vectors of well-known text) directly. +#' @param radius Radius of the earth. Defaults to the average radius of +#' the earth in meters as defined by [s2_earth_radius_meters()]. +#' +#' @export +#' +#' @seealso +#' BigQuery's geography function reference: +#' +#' - [ST_ISCOLLECTION](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_iscollection) +#' - [ST_DIMENSION](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_dimension) +#' - [ST_NUMPOINTS](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_numpoints) +#' - [ST_ISEMPTY](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_isempty) +#' - [ST_AREA](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_area) +#' - [ST_LENGTH](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_length) +#' - [ST_PERIMETER](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_perimeter) +#' - [ST_X](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_x) +#' - [ST_Y](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_y) +#' - [ST_DISTANCE](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_distance) +#' - [ST_MAXDISTANCE](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_maxdistance) +#' +#' @examples +#' # s2_is_collection() tests for multiple geometries in one feature +#' s2_is_collection(c("POINT (-64 45)", "MULTIPOINT ((-64 45), (8 72))")) +#' +#' # s2_dimension() returns 0 for point, 1 for line, 2 for polygon +#' s2_dimension( +#' c( +#' "GEOMETRYCOLLECTION EMPTY", +#' "POINT (-64 45)", +#' "LINESTRING (-64 45, 8 72)", +#' "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", +#' "GEOMETRYCOLLECTION (POINT (-64 45), LINESTRING (-64 45, 8 72))" +#' ) +#' ) +#' +#' # s2_num_points() counts points +#' s2_num_points(c("POINT (-64 45)", "LINESTRING (-64 45, 8 72)")) +#' +#' # s2_is_empty tests for emptiness +#' s2_is_empty(c("POINT (-64 45)", "POINT EMPTY")) +#' +#' # calculate area, length, and perimeter +#' s2_area("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))") +#' s2_perimeter("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))") +#' s2_length(s2_boundary("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))")) +#' +#' # extract x and y coordinates from points +#' s2_x(c("POINT (-64 45)", "POINT EMPTY")) +#' s2_y(c("POINT (-64 45)", "POINT EMPTY")) +#' +#' # calculate minimum and maximum distance between two geometries +#' s2_distance( +#' "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", +#' "POINT (-64 45)" +#' ) +#' s2_max_distance( +#' "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", +#' "POINT (-64 45)" +#' ) +#' +s2_is_collection <- function(x) { + cpp_s2_is_collection(as_s2_geography(x)) +} + +#' @rdname s2_is_collection +#' @export +s2_is_valid <- function(x) { + cpp_s2_is_valid(as_s2_geography(x, check = FALSE)) +} + +#' @rdname s2_is_collection +#' @export +s2_is_valid_detail <- function(x) { + x <- as_s2_geography(x, check = FALSE) + data.frame( + is_valid = cpp_s2_is_valid(x), + reason = cpp_s2_is_valid_reason(x), + stringsAsFactors = FALSE + ) +} + +#' @rdname s2_is_collection +#' @export +s2_dimension <- function(x) { + cpp_s2_dimension(as_s2_geography(x)) +} + +#' @rdname s2_is_collection +#' @export +s2_num_points <- function(x) { + cpp_s2_num_points(as_s2_geography(x)) +} + +#' @rdname s2_is_collection +#' @export +s2_is_empty <- function(x) { + cpp_s2_is_empty(as_s2_geography(x)) +} + +#' @rdname s2_is_collection +#' @export +s2_area <- function(x, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), radius) + cpp_s2_area(recycled[[1]]) * radius ^ 2 +} + +#' @rdname s2_is_collection +#' @export +s2_length <- function(x, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), radius) + cpp_s2_length(recycled[[1]]) * radius +} + +#' @rdname s2_is_collection +#' @export +s2_perimeter <- function(x, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), radius) + cpp_s2_perimeter(recycled[[1]]) * radius +} + +#' @rdname s2_is_collection +#' @export +s2_x <- function(x) { + cpp_s2_x(as_s2_geography(x)) +} + +#' @rdname s2_is_collection +#' @export +s2_y <- function(x) { + cpp_s2_y(as_s2_geography(x)) +} + +# document these with the other linear referencers +#' @rdname s2_interpolate +#' @export +s2_project <- function(x, y, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y), radius) + length <- cpp_s2_length(recycled[[1]]) * radius + cpp_s2_project_normalized(recycled[[1]], recycled[[2]]) * length +} + +#' @rdname s2_interpolate +#' @export +s2_project_normalized <- function(x, y) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_project_normalized(recycled[[1]], recycled[[2]]) +} + +#' @rdname s2_is_collection +#' @export +s2_distance <- function(x, y, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y), radius) + cpp_s2_distance(recycled[[1]], recycled[[2]]) * radius +} + +#' @rdname s2_is_collection +#' @export +s2_max_distance <- function(x, y, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y), radius) + cpp_s2_max_distance(recycled[[1]], recycled[[2]]) * radius +} diff --git a/R/s2-bounds.R b/R/s2-bounds.R new file mode 100644 index 0000000..6c4f792 --- /dev/null +++ b/R/s2-bounds.R @@ -0,0 +1,35 @@ + +#' Compute feature-wise and aggregate bounds +#' +#' [s2_bounds_rect()] returns a bounding latitude-longitude +#' rectangle that contains the region; [s2_bounds_cap()] returns a bounding circle +#' represented by a centre point (lat, lng) and an angle. The bound may not be tight +#' for points, polylines and geometry collections. The rectangle returned may depend on +#' the order of points or polylines. `lng_lo` values larger than `lng_hi` indicate +#' regions that span the antimeridian, see the Fiji example. +#' +#' @inheritParams s2_is_collection +#' @export +#' @return Both functions return a `data.frame`: +#' +#' - [s2_bounds_rect()]: Columns `minlng`, `minlat`, `maxlng`, `maxlat` (degrees) +#' - [s2_bounds_cap()]: Columns `lng`, `lat`, `angle` (degrees) +#' +#' @examples +#' s2_bounds_cap(s2_data_countries("Antarctica")) +#' s2_bounds_cap(s2_data_countries("Netherlands")) +#' s2_bounds_cap(s2_data_countries("Fiji")) +#' +#' s2_bounds_rect(s2_data_countries("Antarctica")) +#' s2_bounds_rect(s2_data_countries("Netherlands")) +#' s2_bounds_rect(s2_data_countries("Fiji")) +#' +s2_bounds_cap <- function(x) { + cpp_s2_bounds_cap(as_s2_geography(x)) +} + +#' @rdname s2_bounds_cap +#' @export +s2_bounds_rect <- function(x) { + cpp_s2_bounds_rect(as_s2_geography(x)) +} diff --git a/R/s2-cell-union.R b/R/s2-cell-union.R new file mode 100644 index 0000000..3ee5af5 --- /dev/null +++ b/R/s2-cell-union.R @@ -0,0 +1,200 @@ + +#' Create S2 Cell Union vectors +#' +#' @param x A `list()` of [s2_cell()] vectors. +#' @param ... Passed to S3 methods +#' +#' @return An object of class "s2_cell_union". +#' @export +#' +s2_cell_union <- function(x = list()) { + x <- as.list(x) + input_na <- vapply(x, is.null, logical(1)) + union <- vector("list", length(x)) + union[input_na] <- list(NULL) + union[!input_na] <- lapply(x[!input_na], as_s2_cell) + new_s2_cell_union(union) +} + +#' @rdname s2_cell_union +#' @export +as_s2_geography.s2_cell_union <- function(x, ...) { + new_s2_geography(cpp_s2_geography_from_cell_union(as_s2_cell_union(x))) +} + +#' @rdname s2_cell_union +#' @export +as_s2_cell_union <- function(x, ...) { + UseMethod("as_s2_cell_union") +} + +#' @rdname s2_cell_union +#' @export +as_s2_cell_union.s2_cell_union <- function(x, ...) { + x +} + +#' @rdname s2_cell_union +#' @export +as_s2_cell_union.s2_cell <- function(x, ...) { + cpp_s2_cell_to_cell_union(x) +} + +#' @rdname s2_cell_union +#' @export +as_s2_cell_union.character <- function(x, ...) { + split <- strsplit(x, "\\s*;\\s*") + split[is.na(x)] <- list(NULL) + s2_cell_union(split) +} + +new_s2_cell_union <- function(x) { + stopifnot(typeof(x) == "list") + structure(x, class = c("s2_cell_union", "wk_vctr")) +} + +#' @export +is.na.s2_cell_union <- function(x, ...) { + cpp_s2_cell_union_is_na(x) +} + +#' @export +format.s2_cell_union <- function(x, ...) { + formatted <- vapply( + unclass(x), + function(e) paste0(as.character(e), collapse = ";"), + character(1) + ) + + formatted[is.na(x)] <- "" + formatted +} + +#' @export +as.character.s2_cell_union <- function(x, ...) { + formatted <- vapply( + unclass(x), + function(e) paste0(as.character(e), collapse = ";"), + character(1) + ) + + formatted[is.na(x)] <- NA_character_ + formatted +} + +#' @export +print.s2_cell_union <- function(x, ...) { + utils::str(x, ...) + invisible(x) +} + +#' @method unlist s2_cell_union +#' @export +unlist.s2_cell_union <- function(x, recursive = TRUE, use.names = TRUE) { + unlisted <- unlist(unclass(x), recursive = recursive, use.names = use.names) + new_s2_cell(as.double(unlisted)) +} + + +#' @importFrom utils str +#' @export +str.s2_cell_union <- function(object, ..., indent.str = "") { + cat(sprintf("%s\n%s", indent.str, length(object), indent.str)) + str(unclass(object), ..., indent.str = indent.str) + invisible(object) +} + + + +#' S2 cell union operators +#' +#' @param x,y An [s2_geography][as_s2_geography] or [s2_cell_union()]. +#' @param min_level,max_level The minimum and maximum levels to constrain the +#' covering. +#' @param max_cells The maximum number of cells in the covering. Defaults to +#' 8. +#' @param buffer A distance to buffer outside the geography +#' @param interior Use `TRUE` to force the covering inside the geography. +#' @inheritParams s2_cell_is_valid +#' +#' @export +#' +s2_cell_union_normalize <- function(x) { + cpp_s2_cell_union_normalize(as_s2_cell_union(x)) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_cell_union_contains <- function(x, y) { + if (inherits(y, "s2_cell")) { + recycled <- recycle_common(as_s2_cell_union(x), y) + cpp_s2_cell_union_contains_cell(recycled[[1]], recycled[[2]]) + } else { + cpp_s2_cell_union_contains(as_s2_cell_union(x), as_s2_cell_union(y)) + } +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_cell_union_intersects <- function(x, y) { + cpp_s2_cell_union_intersects(as_s2_cell_union(x), as_s2_cell_union(y)) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_cell_union_intersection <- function(x, y) { + cpp_s2_cell_union_intersection(as_s2_cell_union(x), as_s2_cell_union(y)) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_cell_union_union <- function(x, y) { + cpp_s2_cell_union_union(as_s2_cell_union(x), as_s2_cell_union(y)) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_cell_union_difference <- function(x, y) { + cpp_s2_cell_union_difference(as_s2_cell_union(x), as_s2_cell_union(y)) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_covering_cell_ids <- function(x, min_level = 0, max_level = 30, + max_cells = 8, buffer = 0, + interior = FALSE, + radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), buffer / radius) + cpp_s2_covering_cell_ids( + recycled[[1]], + min_level, + max_level, + max_cells, + recycled[[2]], + interior + ) +} + +#' @rdname s2_cell_union_normalize +#' @export +s2_covering_cell_ids_agg <- function(x, min_level = 0, max_level = 30, + max_cells = 8, buffer = 0, + interior = FALSE, + radius = s2_earth_radius_meters(), + na.rm = FALSE) { + distance <- as.numeric(buffer / radius) + stopifnot(length(distance) == 1) + if (is.na(distance)) { + return(new_s2_cell_union(list(NULL))) + } + + cpp_s2_covering_cell_ids_agg( + as_s2_geography(x), + min_level, + max_level, + max_cells, + distance, + interior, + na.rm + ) +} diff --git a/R/s2-cell.R b/R/s2-cell.R new file mode 100644 index 0000000..02a316b --- /dev/null +++ b/R/s2-cell.R @@ -0,0 +1,333 @@ + +#' Create S2 Cell vectors +#' +#' The S2 cell indexing system forms the basis for spatial indexing +#' in the S2 library. On their own, S2 cells can represent points +#' or areas. As a union, a vector of S2 cells can approximate a +#' line or polygon. These functions allow direct access to the +#' S2 cell indexing system and are designed to have minimal overhead +#' such that looping and recursion have acceptable performance +#' when used within R code. +#' +#' Under the hood, S2 cell vectors are represented in R as vectors +#' of type [double()]. This works because S2 cell identifiers are +#' 64 bits wide, as are `double`s on all systems where R runs (The +#' same trick is used by the bit64 package to represent signed +#' 64-bit integers). As a happy accident, `NA_real_` is not a valid +#' or meaningful cell identifier, so missing value support in the +#' way R users might expect is preserved. It is worth noting that +#' the underlying value of `s2_cell_sentinel()` would normally be +#' considered `NA`; however, as it is meaningful and useful when +#' programming with S2 cells, custom `is.na()` and comparison methods +#' are implemented such that `s2_cell_sentinel()` is greater than +#' all valid S2 cells and not considered missing. Users can and should +#' implement compiled code that uses the underlying bytes of the +#' vector, ensuring that the class of any returned object that should +#' be interpreted in this way is constructed with `new_s2_cell()`. +#' +#' @param x The canonical S2 cell identifier as a character vector. +#' @param ... Passed to methods +#' +#' @return An object of class s2_cell +#' @export +#' +#' @examples +#' s2_cell("4b59a0cd83b5de49") +#' as_s2_cell(s2_lnglat(-64, 45)) +#' as_s2_cell(s2_data_cities("Ottawa")) +#' +s2_cell <- function(x = character()) { + new_s2_cell(cpp_s2_cell_from_string(x)) +} + +#' @rdname s2_cell +#' @export +s2_cell_sentinel <- function() { + cpp_s2_cell_sentinel() +} + +#' @rdname s2_cell +#' @export +s2_cell_invalid <- function() { + new_s2_cell(0) +} + +#' @rdname s2_cell +#' @export +s2_cell_sentinel <- function() { + cpp_s2_cell_sentinel() +} + +#' @rdname s2_cell +#' @export +as_s2_cell <- function(x, ...) { + UseMethod("as_s2_cell") +} + +#' @rdname s2_cell +#' @export +as_s2_cell.s2_cell <- function(x, ...) { + x +} + +#' @rdname s2_cell +#' @export +as_s2_cell.character <- function(x, ...) { + s2_cell(x) +} + +#' @rdname s2_cell +#' @export +as_s2_cell.s2_geography <- function(x, ...) { + cpp_s2_cell_from_lnglat(list(s2_x(x), s2_y(x))) +} + +#' @rdname s2_cell +#' @export +as_s2_cell.wk_xy <- function(x, ...) { + cpp_s2_cell_from_lnglat(as_s2_lnglat(x)) +} + +#' @rdname s2_cell +#' @export +new_s2_cell <- function(x) { + stopifnot(is.double(x)) + structure(x, class = c("s2_cell", "wk_vctr")) +} + +#' @export +as.character.s2_cell <- function(x, ...) { + cpp_s2_cell_to_string(x) +} + +#' @export +format.s2_cell <- function(x, ...) { + format(as.character(x), quote = FALSE, ...) +} + +#' @export +as.list.s2_cell <- function(x, ...) { + lapply(NextMethod(), new_s2_cell) +} + +#' @export +`[<-.s2_cell` <- function(x, i, value) { + replacement <- as_s2_cell(value) + x <- unclass(x) + x[i] <- replacement + new_s2_cell(x) +} + +#' @export +`[[<-.s2_cell` <- function(x, i, value) { + x[i] <- value + x +} + +#' @export +unique.s2_cell <- function(x, ...) { + cpp_s2_cell_unique(x) +} + +#' @export +sort.s2_cell <- function(x, decreasing = FALSE, ...) { + cpp_s2_cell_sort(x, decreasing) +} + +#' @export +is.na.s2_cell <- function(x) { + cpp_s2_cell_is_na(x) +} + +#' @export +is.numeric.s2_cell <- function(x, ...) { + FALSE +} + +#' @export +Ops.s2_cell <- function(e1, e2) { + switch( + .Generic, + "==" = cpp_s2_cell_eq(e1, e2), + "!=" = cpp_s2_cell_neq(e1, e2), + "<" = cpp_s2_cell_lt(e1, e2), + "<=" = cpp_s2_cell_lte(e1, e2), + ">=" = cpp_s2_cell_gte(e1, e2), + ">" = cpp_s2_cell_gt(e1, e2), + stop("Arithmetic operations are not meaningful for type 's2_cell'", call. = FALSE) + ) +} + +#' @export +Math.s2_cell <- function(x, ...) { + switch( + .Generic, + "cummax" = cpp_s2_cell_cummax(x), + "cummin" = cpp_s2_cell_cummin(x), + stop("Arithmetic operations are not meaningful for type 's2_cell'", call. = FALSE) + ) +} + +#' @export +Summary.s2_cell <- function(x, ..., na.rm = FALSE) { + switch( + .Generic, + "min" = cpp_s2_cell_range(x, na.rm)[1], + "max" = cpp_s2_cell_range(x, na.rm)[2], + "range" = cpp_s2_cell_range(x, na.rm), + stop("Arithmetic operations are not meaningful for type 's2_cell'", call. = FALSE) + ) +} + +#' S2 cell operators +#' +#' @param x,y An [s2_cell()] vector +#' @param level An integer between 0 and 30, inclusive. +#' @param k An integer between 1 and 4 +#' @param radius The radius to use (e.g., [s2_earth_radius_meters()]) +#' @param na.rm Remove NAs prior to computing aggregate? +#' @export +#' +s2_cell_is_valid <- function(x) { + cpp_s2_cell_is_valid(x) +} + +# exporters + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_debug_string <- function(x) { + cpp_s2_cell_debug_string(x) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_to_lnglat <- function(x) { + lnglat <- cpp_s2_cell_to_lnglat(x) + s2_lnglat(lnglat[[1]], lnglat[[2]]) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_center <- function(x) { + cpp_s2_cell_center(x) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_boundary <- function(x) { + s2_boundary(cpp_s2_cell_polygon(x)) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_polygon <- function(x) { + cpp_s2_cell_polygon(x) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_vertex <- function(x, k) { + recycled <- recycle_common(x, k) + cpp_s2_cell_vertex(recycled[[1]], recycled[[2]]) +} + +# accessors + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_level <- function(x) { + cpp_s2_cell_level(x) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_is_leaf <- function(x) { + s2_cell_level(x) == 30L +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_is_face <- function(x) { + s2_cell_level(x) == 0L +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_area <- function(x, radius = s2_earth_radius_meters()) { + cpp_s2_cell_area(x) * radius ^ 2 +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_area_approx <- function(x, radius = s2_earth_radius_meters()) { + cpp_s2_cell_area_approx(x) * radius ^ 2 +} + +# transversers + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_parent <- function(x, level = -1L) { + recycled <- recycle_common(x, level) + cpp_s2_cell_parent(recycled[[1]], recycled[[2]]) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_child <- function(x, k) { + recycled <- recycle_common(x, k) + cpp_s2_cell_child(recycled[[1]], recycled[[2]]) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_edge_neighbour <- function(x, k) { + recycled <- recycle_common(x, k) + cpp_s2_cell_edge_neighbour(recycled[[1]], recycled[[2]]) +} + +# binary operators + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_contains <- function(x, y) { + cpp_s2_cell_contains(x, y) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_distance <- function(x, y, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(x, y, radius) + cpp_s2_cell_distance(recycled[[1]], recycled[[2]]) * radius +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_max_distance <- function(x, y, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(x, y, radius) + cpp_s2_cell_max_distance(recycled[[1]], recycled[[2]]) * radius +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_may_intersect <- function(x, y) { + cpp_s2_cell_may_intersect(x, y) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_common_ancestor_level <- function(x, y) { + cpp_s2_cell_common_ancestor_level(x, y) +} + +#' @rdname s2_cell_is_valid +#' @export +s2_cell_common_ancestor_level_agg <- function(x, na.rm = FALSE) { + x_na <- is.na(x) + if (any(x_na) && !na.rm) { + return(NA_integer_) + } + + cpp_s2_cell_common_ancestor_level_agg(x[!x_na]) +} diff --git a/R/s2-constructors-formatters.R b/R/s2-constructors-formatters.R new file mode 100644 index 0000000..e76501d --- /dev/null +++ b/R/s2-constructors-formatters.R @@ -0,0 +1,210 @@ + +#' Create and Format Geography Vectors +#' +#' These functions create and export [geography vectors][as_s2_geography]. +#' Unlike the BigQuery geography constructors, these functions do not sanitize +#' invalid or redundant input using [s2_union()]. Note that when creating polygons +#' using [s2_make_polygon()], rings can be open or closed. +#' +#' @inheritParams s2_is_collection +#' @inheritParams as_s2_geography +#' @param precision The number of significant digits to export when +#' writing well-known text. If `trim = FALSE`, the number of +#' digits after the decimal place. +#' @param trim Should trailing zeroes be included after the decimal place? +#' @param endian The endian-ness of the well-known binary. See [wk::wkb_translate_wkb()]. +#' @param longitude,latitude Vectors of latitude and longitude +#' @param wkt_string Well-known text +#' @param wkb_bytes A `list()` of `raw()` +#' @param planar Use `TRUE` to force planar edges in import or export. +#' @param tessellate_tol_m The maximum number of meters to that a point must +#' be moved to satisfy the planar edge constraint. +#' @param feature_id,ring_id Vectors for which a change in +#' sequential values indicates a new feature or ring. Use [factor()] +#' to convert from a character vector. +#' +#' @export +#' +#' @seealso +#' See [as_s2_geography()] for other ways to construct geography vectors. +#' +#' BigQuery's geography function reference: +#' +#' - [ST_GEOGPOINT](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogpoint) +#' - [ST_MAKELINE](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_makeline) +#' - [ST_MAKEPOLYGON](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_makepolygon) +#' - [ST_GEOGFROMTEXT](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogfromtext) +#' - [ST_GEOGFROMWKB](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogfromwkb) +#' - [ST_ASTEXT](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_astext) +#' - [ST_ASBINARY](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_asbinary) +#' +#' @examples +#' # create point geographies using coordinate values: +#' s2_geog_point(-64, 45) +#' +#' # create line geographies using coordinate values: +#' s2_make_line(c(-64, 8), c(45, 71)) +#' +#' # optionally, separate features using feature_id: +#' s2_make_line( +#' c(-64, 8, -27, -27), c(45, 71, 0, 45), +#' feature_id = c(1, 1, 2, 2) +#' ) +#' +#' # create polygon geographies using coordinate values: +#' # (rings can be open or closed) +#' s2_make_polygon(c(-45, 8, 0), c(64, 71, 90)) +#' +#' # optionally, separate rings and/or features using +#' # ring_id and/or feature_id +#' s2_make_polygon( +#' c(20, 10, 10, 30, 45, 30, 20, 20, 40, 20, 45), +#' c(35, 30, 10, 5, 20, 20, 15, 25, 40, 45, 30), +#' feature_id = c(rep(1, 8), rep(2, 3)), +#' ring_id = c(1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1) +#' ) +#' +#' # import and export well-known text +#' (geog <- s2_geog_from_text("POINT (-64 45)")) +#' s2_as_text(geog) +#' +#' # import and export well-known binary +#' (geog <- s2_geog_from_wkb(wk::as_wkb("POINT (-64 45)"))) +#' s2_as_binary(geog) +#' +#' # import geometry from planar space +#' s2_geog_from_text( +#' "POLYGON ((0 0, 1 0, 0 1, 0 0))", +#' planar = TRUE, +#' tessellate_tol_m = 1 +#' ) +#' +#' # export geographies into planar space +#' geog <- s2_make_polygon(c(179, -179, 179), c(10, 10, 11)) +#' s2_as_text(geog, planar = TRUE) +#' +#' # polygons containing a pole need an extra step +#' geog <- s2_data_countries("Antarctica") +#' geom <- s2_as_text( +#' s2_intersection(geog, s2_world_plate_carree()), +#' planar = TRUE +#' ) +#' +s2_geog_point <- function(longitude, latitude) { + wk::wk_handle(wk::xy(longitude, latitude), s2_geography_writer()) +} + +#' @rdname s2_geog_point +#' @export +s2_make_line <- function(longitude, latitude, feature_id = 1L) { + wk::wk_handle( + wk::xy(longitude, latitude), + wk::wk_linestring_filter( + s2_geography_writer(), + feature_id = as.integer(feature_id) + ) + ) +} + +#' @rdname s2_geog_point +#' @export +s2_make_polygon <- function(longitude, latitude, feature_id = 1L, ring_id = 1L, + oriented = FALSE, check = TRUE) { + wk::wk_handle( + wk::xy(longitude, latitude), + wk::wk_polygon_filter( + s2_geography_writer(oriented = oriented, check = check), + feature_id = as.integer(feature_id), + ring_id = as.integer(ring_id) + ) + ) +} + +#' @rdname s2_geog_point +#' @export +s2_geog_from_text <- function(wkt_string, oriented = FALSE, check = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default()) { + attributes(wkt_string) <- NULL + wkt <- wk::new_wk_wkt(wkt_string, geodesic = TRUE) + wk::validate_wk_wkt(wkt) + + wk::wk_handle( + wkt, + s2_geography_writer( + oriented = oriented, + check = check, + tessellate_tol = if (planar) { + tessellate_tol_m / s2_earth_radius_meters() + } else { + Inf + } + ) + ) +} + +#' @rdname s2_geog_point +#' @export +s2_geog_from_wkb <- function(wkb_bytes, oriented = FALSE, check = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default()) { + attributes(wkb_bytes) <- NULL + wkb <- wk::new_wk_wkb(wkb_bytes) + wk::validate_wk_wkb(wkb) + wk::wk_handle( + wkb, + s2_geography_writer( + oriented = oriented, + check = check, + tessellate_tol = if (planar) { + tessellate_tol_m / s2_earth_radius_meters() + } else { + Inf + } + ) + ) +} + +#' @rdname s2_geog_point +#' @export +s2_as_text <- function(x, precision = 16, trim = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default()) { + wkt <- wk::wk_handle( + as_s2_geography(x), + wk::wkt_writer(precision = precision, trim = trim), + s2_tessellate_tol = if (planar) { + tessellate_tol_m / s2_earth_radius_meters() + } else { + Inf + } + ) + + attributes(wkt) <- NULL + wkt +} + +#' @rdname s2_geog_point +#' @export +s2_as_binary <- function(x, endian = wk::wk_platform_endian(), + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default()) { + structure( + wk::wk_handle( + as_s2_geography(x), + wk::wkb_writer(endian = endian), + s2_tessellate_tol = if (planar) { + tessellate_tol_m / s2_earth_radius_meters() + } else { + Inf + } + ), + class = "blob" + ) +} + +#' @rdname s2_geog_point +#' @export +s2_tessellate_tol_default <- function() { + 100 +} diff --git a/R/s2-earth.R b/R/s2-earth.R new file mode 100644 index 0000000..4814130 --- /dev/null +++ b/R/s2-earth.R @@ -0,0 +1,23 @@ + +#' Earth Constants +#' +#' According to Yoder (1995), the radius of the earth is +#' 6371.01 km. These functions are used to set the +#' default radis for functions that return a distance +#' or accept a distance as input +#' (e.g., [s2_distance()] and [s2_dwithin()]). +#' +#' @export +#' +#' @references +#' Yoder, C.F. 1995. "Astrometric and Geodetic Properties of Earth and the +#' Solar System" in Global Earth Physics, A Handbook of Physical Constants, +#' AGU Reference Shelf 1, American Geophysical Union, Table 2. +#' \doi{10.1029/RF001p0001} +#' +#' @examples +#' s2_earth_radius_meters() +#' +s2_earth_radius_meters <- function() { + 6371.01 * 1000 +} diff --git a/R/s2-geography.R b/R/s2-geography.R new file mode 100644 index 0000000..df1ea5b --- /dev/null +++ b/R/s2-geography.R @@ -0,0 +1,215 @@ + +#' Create an S2 Geography Vector +#' +#' Geography vectors are arrays of points, lines, polygons, and/or collections +#' of these. Geography vectors assume coordinates are longitude and latitude +#' on a perfect sphere. +#' +#' The coercion function [as_s2_geography()] is used to wrap the input +#' of most functions in the s2 package so that you can use other objects with +#' an unambiguious interpretation as a geography vector. Geography vectors +#' have a minimal [vctrs][vctrs::vctrs-package] implementation, so you can +#' use these objects in tibble, dplyr, and other packages that use the vctrs +#' framework. +#' +#' @param x An object that can be converted to an s2_geography vector +#' @param oriented TRUE if polygon ring directions are known to be correct +#' (i.e., exterior rings are defined counter clockwise and interior +#' rings are defined clockwise). +#' @param check Use `check = FALSE` to skip error on invalid geometries +#' @param ... Unused +#' +#' @return An object with class s2_geography +#' @export +#' +#' @seealso +#' [s2_geog_from_wkb()], [s2_geog_from_text()], [s2_geog_point()], +#' [s2_make_line()], [s2_make_polygon()] for other ways to +#' create geography vectors, and [s2_as_binary()] and [s2_as_text()] +#' for other ways to export them. +#' +as_s2_geography <- function(x, ...) { + UseMethod("as_s2_geography") +} + +#' @rdname as_s2_geography +#' @export +s2_geography <- function() { + new_s2_geography(list()) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.s2_geography <- function(x, ...) { + x +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.wk_xy <- function(x, ...) { + x <- as_s2_lnglat(x) + df <- unclass(x) + s2_geog_point(df[[1]], df[[2]]) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.wk_wkb <- function(x, ..., oriented = FALSE, check = TRUE) { + if (identical(wk::wk_is_geodesic(x), FALSE)) { + + # points and an empty vector are OK and shouldn't trigger an error + meta <- wk::wk_meta(x) + if (!all(meta$geometry_type %in% c(1, 4, NA), na.rm = TRUE)) { + stop( + paste0( + "Can't create s2_geography from Cartesian wkb().\n", + "Use `wk_set_geodesic(x, TRUE)` to assert that edges can be\n", + "interpolated along the sphere." + ), + call. = FALSE + ) + } + } + + wk::wk_handle( + x, + s2_geography_writer(oriented = oriented, check = check) + ) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.WKB <- function(x, ..., oriented = FALSE, check = TRUE) { + s2_geog_from_wkb(x, oriented = oriented, check = check) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.blob <- function(x, ..., oriented = FALSE, check = TRUE) { + s2_geog_from_wkb(x, oriented = oriented, check = check) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.wk_wkt <- function(x, ..., oriented = FALSE, check = TRUE) { + if (identical(wk::wk_is_geodesic(x), FALSE)) { + + # points and an empty vector are OK and shouldn't trigger an error + meta <- wk::wk_meta(x) + if (!all(meta$geometry_type %in% c(1, 4, NA), na.rm = TRUE)) { + stop( + paste0( + "Can't create s2_geography from Cartesian wkt().\n", + "Use `wk_set_geodesic(x, TRUE)` to assert that edges can be\n", + "interpolated along the sphere." + ), + call. = FALSE + ) + } + } + + wk::wk_handle( + x, + s2_geography_writer(oriented = oriented, check = check) + ) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.character <- function(x, ..., oriented = FALSE, check = TRUE) { + s2_geog_from_text(x, oriented = oriented, check = check) +} + +#' @rdname as_s2_geography +#' @export +as_s2_geography.logical <- function(x, ...) { + stopifnot(isTRUE(x)) + new_s2_geography(s2_geography_full(TRUE)) +} + +#' @importFrom wk as_wkb +#' @rdname as_s2_geography +#' @export +as_wkb.s2_geography <- function(x, ...) { + wkb <- wk::wk_handle(x, wk::wkb_writer()) + wk::wk_is_geodesic(wkb) <- TRUE + wk::wk_crs(wkb) <- wk::wk_crs_longlat() + wkb +} + +#' @importFrom wk as_wkt +#' @rdname as_s2_geography +#' @export +as_wkt.s2_geography <- function(x, ...) { + wkt <- wk::wk_handle(x, wk::wkt_writer()) + wk::wk_is_geodesic(wkt) <- TRUE + wk::wk_crs(wkt) <- wk::wk_crs_longlat() + wkt +} + +#' @importFrom wk wk_crs +#' @export +wk_crs.s2_geography <- function(x) { + wk::wk_crs_longlat() +} + +#' @importFrom wk wk_set_crs +#' @export +wk_set_crs.s2_geography <- function(x, crs) { + if (!wk::wk_crs_equal(crs, wk::wk_crs(x))) { + warning("Setting the crs of s2_geography() is not supported") + } + + x +} + +#' @importFrom wk wk_is_geodesic +#' @export +wk_is_geodesic.s2_geography <- function(x) { + TRUE +} + +#' @importFrom wk wk_set_geodesic +#' @export +wk_set_geodesic.s2_geography <- function(x, geodesic) { + if (!isTRUE(geodesic)) { + stop("Can't set geodesic of s2_geography to FALSE") + } + + x +} + +new_s2_geography <- function(x) { + structure(x, class = c("s2_geography", "wk_vctr")) +} + +#' @export +is.na.s2_geography <- function(x) { + cpp_s2_geography_is_na(x) +} + +#' @export +`[<-.s2_geography` <- function(x, i, value) { + x <- unclass(x) + x[i] <- as_s2_geography(value) + new_s2_geography(x) +} + +#' @export +`[[<-.s2_geography` <- function(x, i, value) { + x <- unclass(x) + x[i] <- as_s2_geography(value) + new_s2_geography(x) +} + +#' @export +format.s2_geography <- function(x, ..., max_coords = 5, precision = 9, trim = TRUE) { + wk::wk_format(x, precision = precision, max_coords = max_coords, trim = trim) +} + +# this is what gets called by the RStudio viewer, for which +# format() is best suited (s2_as_text() is more explicit for WKT output) +#' @export +as.character.s2_geography <- function(x, ..., max_coords = 5, precision = 9, trim = TRUE) { + format(x, ..., max_coords = max_coords, precision = precision, trim = trim) +} diff --git a/R/s2-lnglat.R b/R/s2-lnglat.R new file mode 100644 index 0000000..6f39ce2 --- /dev/null +++ b/R/s2-lnglat.R @@ -0,0 +1,60 @@ + +#' Create an S2 LngLat Vector +#' +#' This class represents a latitude and longitude on the Earth's surface. +#' Most calculations in S2 convert this to a [as_s2_point()], which is a +#' unit vector representation of this value. +#' +#' @param lat,lng Vectors of latitude and longitude values in degrees. +#' @param x A [s2_lnglat()] vector or an object that can be coerced to one. +#' @param ... Unused +#' +#' @return An object with class s2_lnglat +#' @export +#' +#' @examples +#' s2_lnglat(45, -64) # Halifax, Nova Scotia! +#' as.data.frame(s2_lnglat(45, -64)) +#' +s2_lnglat <- function(lng, lat) { + wk::xy(lng, lat, crs = wk::wk_crs_longlat()) +} + +#' @rdname s2_lnglat +#' @export +as_s2_lnglat <- function(x, ...) { + UseMethod("as_s2_lnglat") +} + +#' @rdname s2_lnglat +#' @export +as_s2_lnglat.default <- function(x, ...) { + as_s2_lnglat(wk::as_xy(x)) +} + +#' @rdname s2_lnglat +#' @export +as_s2_lnglat.wk_xy <- function(x, ...) { + wk::wk_set_crs( + wk::as_xy(x, dims = c("x", "y")), + wk::wk_crs_longlat() + ) +} + +#' @rdname s2_lnglat +#' @export +as_s2_lnglat.wk_xyz <- function(x, ...) { + if (wk::wk_crs_equal(wk::wk_crs(x), s2_point_crs())) { + wk::new_wk_xy( + s2_lnglat_from_s2_point(x), + crs = wk::wk_crs_longlat() + ) + } else { + NextMethod() + } +} + +#' @export +as_s2_lnglat.character <- function(x, ...) { + as_s2_lnglat(wk::new_wk_wkt(x)) +} diff --git a/R/s2-matrix.R b/R/s2-matrix.R new file mode 100644 index 0000000..48ab320 --- /dev/null +++ b/R/s2-matrix.R @@ -0,0 +1,205 @@ + +#' Matrix Functions +#' +#' These functions are similar to accessors and predicates, but instead of +#' recycling `x` and `y` to a common length and returning a vector of that +#' length, these functions return a vector of length `x` with each element +#' `i` containing information about how the entire vector `y` relates to +#' the feature at `x[i]`. +#' +#' @inheritParams s2_is_collection +#' @inheritParams s2_contains +#' @param x,y Geography vectors, coerced using [as_s2_geography()]. +#' `x` is considered the source, where as `y` is considered the target. +#' @param k The number of closest edges to consider when searching. Note +#' that in S2 a point is also considered an edge. +#' @param min_distance The minimum distance to consider when searching for +#' edges. This filter is applied after the search is complete (i.e., +#' may cause fewer than `k` values to be returned). +#' @param max_distance The maximum distance to consider when searching for +#' edges. This filter is applied before the search. +#' @param max_edges_per_cell For [s2_may_intersect_matrix()], +#' this values controls the nature of the index on `y`, with higher values +#' leading to coarser index. Values should be between 10 and 50; the default +#' of 50 is adequate for most use cases, but for specialized operations users +#' may wish to use a lower value to increase performance. +#' @param max_feature_cells For [s2_may_intersect_matrix()], this value +#' controls the approximation of `x` used to identify potential intersections +#' on `y`. The default value of 4 gives the best performance for most operations, +#' but for specialized operations users may wish to use a higher value to increase +#' performance. +#' +#' @return A vector of length `x`. +#' @export +#' +#' @seealso +#' See pairwise predicate functions (e.g., [s2_intersects()]). +#' +#' @examples +#' city_names <- c("Vatican City", "San Marino", "Luxembourg") +#' cities <- s2_data_cities(city_names) +#' country_names <- s2_data_tbl_countries$name +#' countries <- s2_data_countries() +#' +#' # closest feature returns y indices of the closest feature +#' # for each feature in x +#' country_names[s2_closest_feature(cities, countries)] +#' +#' # farthest feature returns y indices of the farthest feature +#' # for each feature in x +#' country_names[s2_farthest_feature(cities, countries)] +#' +#' # use s2_closest_edges() to find the k-nearest neighbours +#' nearest <- s2_closest_edges(cities, cities, k = 2, min_distance = 0) +#' city_names +#' city_names[unlist(nearest)] +#' +#' # predicate matrices +#' country_names[s2_intersects_matrix(cities, countries)[[1]]] +#' +#' # distance matrices +#' s2_distance_matrix(cities, cities) +#' s2_max_distance_matrix(cities, countries[1:4]) +#' +s2_closest_feature <- function(x, y) { + cpp_s2_closest_feature(as_s2_geography(x), as_s2_geography(y)) +} + +#' @rdname s2_closest_feature +#' @export +s2_closest_edges <- function(x, y, k, min_distance = -1, max_distance = Inf, + radius = s2_earth_radius_meters()) { + stopifnot(k >= 1) + cpp_s2_closest_edges( + as_s2_geography(x), + as_s2_geography(y), + k, + min_distance / radius, + max_distance / radius + ) +} + +#' @rdname s2_closest_feature +#' @export +s2_farthest_feature <- function(x, y) { + cpp_s2_farthest_feature(as_s2_geography(x), as_s2_geography(y)) +} + +#' @rdname s2_closest_feature +#' @export +s2_distance_matrix <- function(x, y, radius = s2_earth_radius_meters()) { + cpp_s2_distance_matrix(as_s2_geography(x), as_s2_geography(y)) * radius +} + +#' @rdname s2_closest_feature +#' @export +s2_max_distance_matrix <- function(x, y, radius = s2_earth_radius_meters()) { + cpp_s2_max_distance_matrix(as_s2_geography(x), as_s2_geography(y)) * radius +} + +#' @rdname s2_closest_feature +#' @export +s2_contains_matrix <- function(x, y, options = s2_options(model = "open")) { + cpp_s2_contains_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_within_matrix <- function(x, y, options = s2_options(model = "open")) { + cpp_s2_within_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_covers_matrix <- function(x, y, options = s2_options(model = "closed")) { + cpp_s2_contains_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_covered_by_matrix <- function(x, y, options = s2_options(model = "closed")) { + cpp_s2_within_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_intersects_matrix <- function(x, y, options = s2_options()) { + cpp_s2_intersects_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_disjoint_matrix <- function(x, y, options = s2_options()) { + # disjoint is the odd one out, in that it requires a negation of intersects + # this is inconvenient to do on the C++ level, and is easier to maintain + # with setdiff() here (unless somebody complains that this is slow) + intersection <- cpp_s2_intersects_matrix(as_s2_geography(x), as_s2_geography(y), options) + Map(setdiff, list(seq_along(y)), intersection) +} + +#' @rdname s2_closest_feature +#' @export +s2_equals_matrix <- function(x, y, options = s2_options()) { + cpp_s2_equals_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_touches_matrix <- function(x, y, options = s2_options()) { + cpp_s2_touches_matrix(as_s2_geography(x), as_s2_geography(y), options) +} + +#' @rdname s2_closest_feature +#' @export +s2_dwithin_matrix <- function(x, y, distance, radius = s2_earth_radius_meters()) { + cpp_s2_dwithin_matrix(as_s2_geography(x), as_s2_geography(y), distance / radius) +} + +#' @rdname s2_closest_feature +#' @export +s2_may_intersect_matrix <- function(x, y, max_edges_per_cell = 50, max_feature_cells = 4) { + cpp_s2_may_intersect_matrix( + as_s2_geography(x), as_s2_geography(y), + max_edges_per_cell, max_feature_cells, + s2_options() + ) +} + +# ------- for testing, non-indexed versions of matrix operators ------- + +s2_contains_matrix_brute_force <- function(x, y, options = s2_options()) { + cpp_s2_contains_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_within_matrix_brute_force <- function(x, y, options = s2_options()) { + cpp_s2_within_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_covers_matrix_brute_force <- function(x, y, options = s2_options(model = "closed")) { + cpp_s2_contains_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_covered_by_matrix_brute_force <- function(x, y, options = s2_options(model = "closed")) { + cpp_s2_within_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_intersects_matrix_brute_force <- function(x, y, options = s2_options()) { + cpp_s2_intersects_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_disjoint_matrix_brute_force <- function(x, y, options = s2_options()) { + cpp_s2_disjoint_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_equals_matrix_brute_force <- function(x, y, options = s2_options()) { + cpp_s2_equals_matrix_brute_force(as_s2_geography(x), as_s2_geography(y), options) +} + +s2_dwithin_matrix_brute_force <- function(x, y, distance, + radius = s2_earth_radius_meters()) { + cpp_s2_dwithin_matrix_brute_force( + as_s2_geography(x), + as_s2_geography(y), + distance / radius + ) +} diff --git a/R/s2-options.R b/R/s2-options.R new file mode 100644 index 0000000..f8b3672 --- /dev/null +++ b/R/s2-options.R @@ -0,0 +1,147 @@ + +#' Geography Operation Options +#' +#' These functions specify defaults for options used to perform operations +#' and construct geometries. These are used in predicates (e.g., [s2_intersects()]), +#' and boolean operations (e.g., [s2_intersection()]) to specify the model for +#' containment and how new geometries should be constructed. +#' +#' @param model One of 'open', 'semi-open' (default for polygons), +#' or 'closed' (default for polylines). See section 'Model' +#' @param snap Use `s2_snap_identity()`, `s2_snap_distance()`, `s2_snap_level()`, +#' or `s2_snap_precision()` to specify how or if coordinate rounding should +#' occur. +#' @param snap_radius As opposed to the snap function, which specifies +#' the maximum distance a vertex should move, the snap radius (in radians) sets +#' the minimum distance between vertices of the output that don't cause vertices +#' to move more than the distance specified by the snap function. This can be used +#' to simplify the result of a boolean operation. Use -1 to specify that any +#' minimum distance is acceptable. +#' @param duplicate_edges Use `TRUE` to keep duplicate edges (e.g., duplicate +#' points). +#' @param edge_type One of 'directed' (default) or 'undirected'. +#' @param polyline_type One of 'path' (default) or 'walk'. If 'walk', +#' polylines that backtrack are preserved. +#' @param polyline_sibling_pairs One of 'discard' (default) or 'keep'. +#' @param simplify_edge_chains Use `TRUE` to remove vertices that are within +#' `snap_radius` of the original vertex. +#' @param split_crossing_edges Use `TRUE` to split crossing polyline edges +#' when creating geometries. +#' @param idempotent Use `FALSE` to apply snap even if snapping is not necessary +#' to satisfy vertex constraints. +#' @param validate Use `TRUE` to validate the result from the builder. +#' @param level A value from 0 to 30 corresponding to the cell level +#' at which snapping should occur. +#' @param distance A distance (in radians) denoting the maximum +#' distance a vertex should move in the snapping process. +#' @param precision A number by which coordinates should be multiplied +#' before being rounded. Rounded to the nearest exponent of 10. +#' @param dimensions A combination of 'point', 'polyline', and/or 'polygon' +#' that can used to constrain the output of [s2_rebuild()] or a +#' boolean operation. +#' +#' @section Model: +#' The geometry model indicates whether or not a geometry includes its boundaries. +#' Boundaries of line geometries are its end points. +#' OPEN geometries do not contain their boundary (`model = "open"`); CLOSED +#' geometries (`model = "closed"`) contain their boundary; SEMI-OPEN geometries +#' (`model = "semi-open"`) contain half of their boundaries, such that when two polygons +#' do not overlap or two lines do not cross, no point exist that belong to +#' more than one of the geometries. (This latter form, half-closed, is +#' not present in the OpenGIS "simple feature access" (SFA) standard nor DE9-IM on +#' which that is based). The default values for [s2_contains()] (open) +#' and covers/covered_by (closed) correspond to the SFA standard specification +#' of these operators. +#' +#' @export +#' +#' @examples +#' # use s2_options() to specify containment models, snap level +#' # layer creation options, and builder options +#' s2_options(model = "closed", snap = s2_snap_level(30)) +#' +s2_options <- function(model = NULL, + snap = s2_snap_identity(), + snap_radius = -1, + duplicate_edges = FALSE, + edge_type = "directed", + validate = FALSE, + polyline_type = "path", + polyline_sibling_pairs = "keep", + simplify_edge_chains = FALSE, + split_crossing_edges = FALSE, + idempotent = FALSE, + dimensions = c("point", "polyline", "polygon")) { + # check snap radius (passing in a huge snap radius can cause problems) + if (snap_radius > 3) { + stop( + "Snap radius is too large. Did you pass in a snap radius in meters instead of radians?", + call. = FALSE + ) + } + + structure( + list( + # model needs to be "unset" by default because there are differences in polygon + # and polyline handling by default that are good defaults to preserve + model = if (is.null(model)) -1 else match_option(model[1], c("open", "semi-open", "closed"), "model"), + snap = snap, + snap_radius = snap_radius, + duplicate_edges = duplicate_edges, + edge_type = match_option(edge_type[1], c("directed", "undirected"), "edge_type"), + validate = validate, + polyline_type = match_option(polyline_type[1], c("path", "walk"), "polyline_type"), + polyline_sibling_pairs = match_option( + polyline_sibling_pairs, + c("discard", "keep"), + "polyline_sibling_pairs" + ), + simplify_edge_chains = simplify_edge_chains, + split_crossing_edges = split_crossing_edges, + idempotent = idempotent, + dimensions = match_option(dimensions, c("point", "polyline", "polygon"), "dimensions") + ), + class = "s2_options" + ) +} + +#' @rdname s2_options +#' @export +s2_snap_identity <- function() { + structure(list(), class = "snap_identity") +} + +#' @rdname s2_options +#' @export +s2_snap_level <- function(level) { + if (level > 30) { + stop("`level` must be an intger between 1 and 30", call. = FALSE) + } + + structure(list(level = level), class = "snap_level") +} + +#' @rdname s2_options +#' @export +s2_snap_precision <- function(precision) { + structure(list(exponent = round(log10(precision))), class = "snap_precision") +} + +#' @rdname s2_options +#' @export +s2_snap_distance <- function(distance) { + structure(list(distance = distance), class = "snap_distance") +} + + +match_option <- function(x, options, arg) { + result <- match(x, options) + if (any(is.na(result))) { + stop( + sprintf("`%s` must be one of %s", arg, paste0('"', options, '"', collapse = ", ")), + call. = FALSE + ) + } + + result +} diff --git a/R/s2-package.R b/R/s2-package.R new file mode 100644 index 0000000..6c757a5 --- /dev/null +++ b/R/s2-package.R @@ -0,0 +1,10 @@ +#' @keywords internal +"_PACKAGE" + +# The following block is used by usethis to automatically manage +# roxygen namespace tags. Modify with care! +## usethis namespace: start +#' @useDynLib s2, .registration = TRUE +#' @importFrom Rcpp sourceCpp +## usethis namespace: end +NULL diff --git a/R/s2-point.R b/R/s2-point.R new file mode 100644 index 0000000..26e8adf --- /dev/null +++ b/R/s2-point.R @@ -0,0 +1,68 @@ + +#' Create an S2 Point Vector +#' +#' In S2 terminology, a "point" is a 3-dimensional unit vector representation +#' of an [s2_point()]. Internally, all s2 objects are stored as +#' 3-dimensional unit vectors. +#' +#' @param x,y,z Vectors of latitude and longitude values in degrees. +#' @param ... Unused +#' +#' @return An object with class s2_point +#' @export +#' +#' @examples +#' point <- s2_lnglat(-64, 45) # Halifax, Nova Scotia! +#' as_s2_point(point) +#' as.data.frame(as_s2_point(point)) +#' +s2_point <- function(x, y, z) { + wk::xyz(x, y, z, crs = s2_point_crs()) +} + +#' @rdname s2_point +#' @export +s2_point_crs <- function() { + structure(list(), class = "s2_point_crs") +} + +#' @export +format.s2_point_crs <- function(x, ...) { + "s2_point_crs" +} + +#' @rdname s2_point +#' @export +as_s2_point <- function(x, ...) { + UseMethod("as_s2_point") +} + +#' @rdname s2_point +#' @export +as_s2_point.default <- function(x, ...) { + as_s2_point(wk::as_xy(x)) +} + +#' @rdname s2_point +#' @export +as_s2_point.wk_xy <- function(x, ...) { + stopifnot(wk::wk_crs_equal(wk::wk_crs(x), wk::wk_crs_longlat())) + wk::new_wk_xyz( + s2_point_from_s2_lnglat(x), + crs = s2_point_crs() + ) +} + +#' @rdname s2_point +#' @export +as_s2_point.wk_xyz <- function(x, ...) { + wk::wk_set_crs( + wk::as_xy(x, dims = c("x", "y", "z")), + s2_point_crs() + ) +} + +#' @export +as_s2_point.character <- function(x, ...) { + as_s2_point(wk::new_wk_wkt(x)) +} diff --git a/R/s2-predicates.R b/R/s2-predicates.R new file mode 100644 index 0000000..26a5738 --- /dev/null +++ b/R/s2-predicates.R @@ -0,0 +1,178 @@ + +#' S2 Geography Predicates +#' +#' These functions operate two geography vectors (pairwise), and return +#' a logical vector. +#' +#' @inheritParams s2_is_collection +#' @inheritParams s2_boundary +#' @param distance A distance on the surface of the earth in the same units +#' as `radius`. +#' @param lng1,lat1,lng2,lat2 A latitude/longitude range +#' @param detail The number of points with which to approximate +#' non-geodesic edges. +#' +#' @inheritSection s2_options Model +#' +#' @export +#' +#' @seealso +#' Matrix versions of these predicates (e.g., [s2_intersects_matrix()]). +#' +#' BigQuery's geography function reference: +#' +#' - [ST_CONTAINS](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_contains) +#' - [ST_COVEREDBY](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_coveredby) +#' - [ST_COVERS](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_covers) +#' - [ST_DISJOINT](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_disjoint) +#' - [ST_EQUALS](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_equals) +#' - [ST_INTERSECTS](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersects) +#' - [ST_INTERSECTSBOX](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersectsbox) +#' - [ST_TOUCHES](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_touches) +#' - [ST_WITHIN](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_within) +#' - [ST_DWITHIN](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_dwithin) +#' +#' @examples +#' s2_contains( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_within( +#' c("POINT (5 5)", "POINT (-1 1)"), +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))" +#' ) +#' +#' s2_covered_by( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_covers( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_disjoint( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_intersects( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_equals( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((10 0, 10 10, 0 10, 0 0, 10 0))", +#' "POLYGON ((-1 -1, 10 0, 10 10, 0 10, -1 -1))" +#' ) +#' ) +#' +#' s2_intersects( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)") +#' ) +#' +#' s2_intersects_box( +#' c("POINT (5 5)", "POINT (-1 1)"), +#' 0, 0, 10, 10 +#' ) +#' +#' s2_touches( +#' "POLYGON ((0 0, 0 1, 1 1, 0 0))", +#' c("POINT (0 0)", "POINT (0.5 0.75)", "POINT (0 0.5)") +#' ) +#' +#' s2_dwithin( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)"), +#' 0 # distance in meters +#' ) +#' +#' s2_dwithin( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' c("POINT (5 5)", "POINT (-1 1)"), +#' 1e6 # distance in meters +#' ) +#' +s2_contains <- function(x, y, options = s2_options(model = "open")) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_contains(recycled[[1]], recycled[[2]], options) +} + +#' @rdname s2_contains +#' @export +s2_within <- function(x, y, options = s2_options(model = "open")) { + s2_contains(y, x, options) +} + +#' @rdname s2_contains +#' @export +s2_covered_by <- function(x, y, options = s2_options(model = "closed")) { + s2_covers(y, x, options) +} + +#' @rdname s2_contains +#' @export +s2_covers <- function(x, y, options = s2_options(model = "closed")) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_contains(recycled[[1]], recycled[[2]], options) +} + +#' @rdname s2_contains +#' @export +s2_disjoint <- function(x, y, options = s2_options()) { + !s2_intersects(x, y, options) +} + +#' @rdname s2_contains +#' @export +s2_intersects <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_intersects(recycled[[1]], recycled[[2]], options) +} + +#' @rdname s2_contains +#' @export +s2_equals <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_equals(recycled[[1]], recycled[[2]], options) +} + +#' @rdname s2_contains +#' @export +s2_intersects_box <- function(x, lng1, lat1, lng2, lat2, detail = 1000, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), lng1, lat1, lng2, lat2, detail) + cpp_s2_intersects_box( + recycled[[1]], + recycled[[2]], recycled[[3]], + recycled[[4]], recycled[[5]], + detail = recycled[[6]], + s2options = options + ) +} + +#' @rdname s2_contains +#' @export +s2_touches <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + cpp_s2_touches(recycled[[1]], recycled[[2]], options) +} + +#' @rdname s2_contains +#' @export +s2_dwithin <- function(x, y, distance, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y), distance / radius) + cpp_s2_dwithin(recycled[[1]], recycled[[2]], recycled[[3]]) +} + +#' @rdname s2_contains +#' @export +s2_prepared_dwithin <- function(x, y, distance, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y), distance / radius) + cpp_s2_prepared_dwithin(recycled[[1]], recycled[[2]], recycled[[3]]) +} diff --git a/R/s2-transformers.R b/R/s2-transformers.R new file mode 100644 index 0000000..d5f64e5 --- /dev/null +++ b/R/s2-transformers.R @@ -0,0 +1,298 @@ + +#' S2 Geography Transformations +#' +#' These functions operate on one or more geography vectors and +#' return a geography vector. +#' +#' @inheritParams s2_is_collection +#' @param na.rm For aggregate calculations use `na.rm = TRUE` +#' to drop missing values. +#' @param grid_size The grid size to which coordinates should be snapped; +#' will be rounded to the nearest power of 10. +#' @param options An [s2_options()] object describing the polygon/polyline +#' model to use and the snap level. +#' @param distance The distance to buffer, in units of `radius`. +#' @param max_cells The maximum number of cells to approximate a buffer. +#' @param min_level The minimum cell level used to approximate a buffer +#' (1 - 30). Setting this value too high will result in unnecessarily +#' large geographies, but may help improve buffers along long, narrow +#' regions. +#' @param tolerance The minimum distance between vertexes to use when +#' simplifying a geography. +#' +#' @inheritSection s2_options Model +#' +#' @export +#' +#' @seealso +#' BigQuery's geography function reference: +#' +#' - [ST_BOUNDARY](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_boundary) +#' - [ST_CENTROID](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_centroid) +#' - [ST_CLOSESTPOINT](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_closestpoint) +#' - [ST_DIFFERENCE](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_difference) +#' - [ST_INTERSECTION](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersection) +#' - [ST_UNION](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_union) +#' - [ST_SNAPTOGRID](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_snaptogrid) +#' - [ST_SIMPLIFY](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_simplify) +#' - [ST_UNION_AGG](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_union_agg) +#' - [ST_CENTROID_AGG](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#s2_centroid_agg) +#' +#' @examples +#' # returns the boundary: +#' # empty for point, endpoints of a linestring, +#' # perimeter of a polygon +#' s2_boundary("POINT (-64 45)") +#' s2_boundary("LINESTRING (0 0, 10 0)") +#' s2_boundary("POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))") +#' +#' # returns the area-weighted centroid, element-wise +#' s2_centroid("POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))") +#' s2_centroid("LINESTRING (0 0, 10 0)") +#' +#' # s2_point_on_surface guarantees a point on surface +#' # Note: this is not the same as st_point_on_surface +#' s2_centroid("POLYGON ((0 0, 10 0, 1 1, 0 10, 0 0))") +#' s2_point_on_surface("POLYGON ((0 0, 10 0, 1 1, 0 10, 0 0))") +#' +#' # returns the unweighted centroid of the entire input +#' s2_centroid_agg(c("POINT (0 0)", "POINT (10 0)")) +#' +#' # returns the closest point on x to y +#' s2_closest_point( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POINT (0 90)" # north pole! +#' ) +#' +#' # returns the shortest possible line between x and y +#' s2_minimum_clearance_line_between( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POINT (0 90)" # north pole! +#' ) +#' +#' # binary operations: difference, symmetric difference, intersection and union +#' s2_difference( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", +#' # 32 bit platforms may need to set snap rounding +#' s2_options(snap = s2_snap_level(30)) +#' ) +#' +#' s2_sym_difference( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", +#' # 32 bit platforms may need to set snap rounding +#' s2_options(snap = s2_snap_level(30)) +#' ) +#' +#' s2_intersection( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", +#' # 32 bit platforms may need to set snap rounding +#' s2_options(snap = s2_snap_level(30)) +#' ) +#' +#' s2_union( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", +#' # 32 bit platforms may need to set snap rounding +#' s2_options(snap = s2_snap_level(30)) +#' ) +#' +#' # s2_convex_hull_agg builds the convex hull of a list of geometries +#' s2_convex_hull_agg( +#' c( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))" +#' ) +#' ) +#' +#' # use s2_union_agg() to aggregate geographies in a vector +#' s2_coverage_union_agg( +#' c( +#' "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", +#' "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))" +#' ), +#' # 32 bit platforms may need to set snap rounding +#' s2_options(snap = s2_snap_level(30)) +#' ) +#' +#' # snap to grid rounds coordinates to a specified grid size +#' s2_snap_to_grid("POINT (0.333333333333 0.666666666666)", 1e-2) +#' +#' +s2_boundary <- function(x) { + new_s2_geography(cpp_s2_boundary(as_s2_geography(x))) +} + +#' @rdname s2_boundary +#' @export +s2_centroid <- function(x) { + new_s2_geography(cpp_s2_centroid(as_s2_geography(x))) +} + +#' @rdname s2_boundary +#' @export +s2_closest_point <- function(x, y) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + new_s2_geography(cpp_s2_closest_point(recycled[[1]], recycled[[2]])) +} + +#' @rdname s2_boundary +#' @export +s2_minimum_clearance_line_between <- function(x, y) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + new_s2_geography(cpp_s2_minimum_clearance_line_between(recycled[[1]], recycled[[2]])) +} + +#' @rdname s2_boundary +#' @export +s2_difference <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + new_s2_geography(cpp_s2_difference(recycled[[1]], recycled[[2]], options)) +} + +#' @rdname s2_boundary +#' @export +s2_sym_difference <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + new_s2_geography(cpp_s2_sym_difference(recycled[[1]], recycled[[2]], options)) +} + +#' @rdname s2_boundary +#' @export +s2_intersection <- function(x, y, options = s2_options()) { + recycled <- recycle_common(as_s2_geography(x), as_s2_geography(y)) + new_s2_geography(cpp_s2_intersection(recycled[[1]], recycled[[2]], options)) +} + +#' @rdname s2_boundary +#' @export +s2_union <- function(x, y = NULL, options = s2_options()) { + x <- as_s2_geography(x) + + if (is.null(y)) { + new_s2_geography(cpp_s2_unary_union(x, options)) + } else { + recycled <- recycle_common(x, as_s2_geography(y)) + new_s2_geography(cpp_s2_union(recycled[[1]], recycled[[2]], options)) + } +} + +#' @rdname s2_boundary +#' @export +s2_snap_to_grid <- function(x, grid_size) { + s2_rebuild( + x, + options = s2_options( + snap = s2_snap_precision(10^(-log10(grid_size))), + duplicate_edges = TRUE + ) + ) +} + +#' @rdname s2_boundary +#' @export +s2_simplify <- function(x, tolerance, radius = s2_earth_radius_meters()) { + s2_rebuild(x, options = s2_options(snap_radius = tolerance / radius, simplify_edge_chains = TRUE)) +} + +#' @rdname s2_boundary +#' @export +s2_rebuild <- function(x, options = s2_options()) { + new_s2_geography(cpp_s2_rebuild(as_s2_geography(x), options)) +} + +#' @rdname s2_boundary +#' @export +s2_buffer_cells <- function(x, distance, max_cells = 1000, min_level = -1, + radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), distance / radius) + new_s2_geography(cpp_s2_buffer_cells(recycled[[1]], recycled[[2]], max_cells, min_level)) +} + +#' @rdname s2_boundary +#' @export +s2_convex_hull <- function(x) { + new_s2_geography(cpp_s2_convex_hull(as_s2_geography(x))) +} + +#' @rdname s2_boundary +#' @export +s2_centroid_agg <- function(x, na.rm = FALSE) { + new_s2_geography(cpp_s2_centroid_agg(as_s2_geography(x), naRm = na.rm)) +} + +#' @rdname s2_boundary +#' @export +s2_coverage_union_agg <- function(x, options = s2_options(), na.rm = FALSE) { + new_s2_geography(cpp_s2_coverage_union_agg(as_s2_geography(x), options, na.rm)) +} + +#' @rdname s2_boundary +#' @export +s2_rebuild_agg <- function(x, options = s2_options(), na.rm = FALSE) { + new_s2_geography(cpp_s2_rebuild_agg(as_s2_geography(x), options, na.rm)) +} + +#' @rdname s2_boundary +#' @export +s2_union_agg <- function(x, options = s2_options(), na.rm = FALSE) { + new_s2_geography(cpp_s2_union_agg(s2_union(x, options = options), options, na.rm)) +} + +#' @rdname s2_boundary +#' @export +s2_convex_hull_agg <- function(x, na.rm = FALSE) { + new_s2_geography(cpp_s2_convex_hull_agg(as_s2_geography(x), na.rm)) +} + +#' Linear referencing +#' +#' @param x A simple polyline geography vector +#' @param y A simple point geography vector. The point will be +#' snapped to the nearest point on `x` for the purposes of +#' interpolation. +#' @param distance A distance along `x` in `radius` units. +#' @param distance_normalized A `distance` normalized to [s2_length()] of +#' `x`. +#' @inheritParams s2_is_collection +#' +#' @return +#' - `s2_interpolate()` returns the point on `x`, `distance` meters +#' along the line. +#' - `s2_interpolate_normalized()` returns the point on `x` interpolated +#' to a fraction along the line. +#' - `s2_project()` returns the `distance` that `point` occurs along `x`. +#' - `s2_project_normalized()` returns the `distance_normalized` along `x` +#' where `point` occurs. +#' @export +#' +#' @examples +#' s2_project_normalized("LINESTRING (0 0, 0 90)", "POINT (0 22.5)") +#' s2_project("LINESTRING (0 0, 0 90)", "POINT (0 22.5)") +#' s2_interpolate_normalized("LINESTRING (0 0, 0 90)", 0.25) +#' s2_interpolate("LINESTRING (0 0, 0 90)", 2501890) +#' +s2_interpolate <- function(x, distance, radius = s2_earth_radius_meters()) { + recycled <- recycle_common(as_s2_geography(x), distance / radius) + length <- cpp_s2_length(recycled[[1]]) + new_s2_geography( + cpp_s2_interpolate_normalized(recycled[[1]], distance / radius / length) + ) +} + +#' @rdname s2_interpolate +#' @export +s2_interpolate_normalized <- function(x, distance_normalized) { + recycled <- recycle_common(as_s2_geography(x), distance_normalized) + new_s2_geography( + cpp_s2_interpolate_normalized(recycled[[1]], distance_normalized) + ) +} + +#' @rdname s2_boundary +#' @export +s2_point_on_surface <- function(x, na.rm = FALSE) { + new_s2_geography(cpp_s2_point_on_surface(as_s2_geography(x))) +} diff --git a/R/utils.R b/R/utils.R new file mode 100644 index 0000000..f12af48 --- /dev/null +++ b/R/utils.R @@ -0,0 +1,88 @@ + +new_data_frame <- function(x) { + structure(x, row.names = c(NA, length(x[[1]])), class = "data.frame") +} + +recycle_common <- function(...) { + dots <- list(...) + lengths <- vapply(dots, length, integer(1)) + non_constant_lengths <- unique(lengths[lengths != 1]) + if (length(non_constant_lengths) == 0) { + final_length <- 1 + } else if(length(non_constant_lengths) == 1) { + final_length <- non_constant_lengths + } else { + lengths_label <- paste0(non_constant_lengths, collapse = ", ") + stop(sprintf("Incompatible lengths: %s", lengths_label)) + } + + lapply(dots, rep_len, final_length) +} + +# The problems object is generated when building or processing an s2_geography(): +# instead of attaching to the object as an attribute, this function is +# called from Rcpp if there were any problems to format them in a +# human-readable way. Theoretically one could change this to only warn +# instead of stop (error values are set to NA/NULL). +stop_problems_create <- function(feature_id, problem) { + n <- length(feature_id) + feature_label <- if (n != 1) "features" else "feature" + + stop_problems( + feature_id, + problem, + sprintf("Found %d %s with invalid spherical geometry.", n, feature_label) + ) +} + +stop_problems_process <- function(feature_id, problem) { + n <- length(feature_id) + error_label <- if (n != 1) "errors" else "error" + + stop_problems( + feature_id, + problem, + sprintf("Encountered %d processing %s.", n, error_label) + ) +} + +stop_problems <- function(feature_id, problem, header) { + n <- length(feature_id) + + if (n > 10) { + feature_id <- feature_id[1:10] + problem <- problem[1:10] + more <- sprintf("\n...and %s more", n - 10) + } else { + more <- "" + } + + msg <- paste0( + header, "\n", + paste0("[", feature_id + 1, "] ", problem , collapse = "\n"), + more + ) + + stop(msg, call. = FALSE) +} + +expect_wkt_equal <- function(x, y, precision = 16) { + testthat::expect_equal( + wk::wk_format( + as_s2_geography(x), + precision = precision, + trim = TRUE, + max_coords = .Machine$integer.max + ), + wk::wk_format( + as_s2_geography(y), + precision = precision, + trim = TRUE, + max_coords = .Machine$integer.max + ) + ) +} + +expect_near <- function(x, y, epsilon) { + testthat::expect_true(abs(y - x) < epsilon) +} diff --git a/R/vctrs.R b/R/vctrs.R new file mode 100644 index 0000000..fd985f5 --- /dev/null +++ b/R/vctrs.R @@ -0,0 +1,36 @@ + +vec_proxy.s2_geography <- function(x, ...) { + unclass(x) +} + +vec_restore.s2_geography <- function(x, ...) { + new_s2_geography(x) +} + +vec_ptype_abbr.s2_geography <- function(x, ...) { + "s2_geography" +} + +vec_proxy.s2_cell <- function(x, ...) { + unclass(x) +} + +vec_restore.s2_cell <- function(x, ...) { + new_s2_cell(x) +} + +vec_ptype_abbr.s2_cell <- function(x, ...) { + "s2cell" +} + +vec_proxy.s2_cell_union <- function(x, ...) { + unclass(x) +} + +vec_restore.s2_cell_union <- function(x, ...) { + new_s2_cell_union(x) +} + +vec_ptype_abbr.s2_cell_union <- function(x, ...) { + "s2cellunion" +} diff --git a/R/wk-utils.R b/R/wk-utils.R new file mode 100644 index 0000000..df6a6ca --- /dev/null +++ b/R/wk-utils.R @@ -0,0 +1,129 @@ + +#' Low-level wk filters and handlers +#' +#' @inheritParams wk::wk_handle +#' @param projection,s2_projection One of [s2_projection_plate_carree()] or +#' [s2_projection_mercator()] +#' @param tessellate_tol,s2_tessellate_tol An angle in radians. +#' Points will not be added if a line segment is within this +#' distance of a point. +#' @param x_scale The maximum x value of the projection +#' @param centre The center point of the orthographic projection +#' @param epsilon_east_west,epsilon_north_south Use a positive number to +#' define the edges of a Cartesian world slightly inward from -180, -90, +#' 180, 90. This may be used to define a world outline for a projection where +#' projecting at the extreme edges of the earth results in a non-finite value. +#' @inheritParams as_s2_geography +#' +#' @return +#' - `s2_projection_plate_carree()`, `s2_projection_mercator()`: An external pointer +#' to an S2 projection. +#' @importFrom wk wk_handle +#' @export +#' +wk_handle.s2_geography <- function(handleable, handler, ..., + s2_projection = s2_projection_plate_carree(), + s2_tessellate_tol = Inf) { + stopifnot(is.null(s2_projection) || inherits(s2_projection, "s2_projection")) + attr(handleable, "s2_projection") <- s2_projection + + if (identical(s2_tessellate_tol, Inf)) { + .Call(c_s2_handle_geography, handleable, wk::as_wk_handler(handler)) + } else { + attr(handleable, "s2_tessellate_tol") <- as.double(s2_tessellate_tol)[1] + .Call(c_s2_handle_geography_tessellated, handleable, wk::as_wk_handler(handler)) + } +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_geography_writer <- function(oriented = FALSE, check = TRUE, + projection = s2_projection_plate_carree(), + tessellate_tol = Inf) { + stopifnot(is.null(projection) || inherits(projection, "s2_projection")) + + wk::new_wk_handler( + .Call( + c_s2_geography_writer_new, + as.logical(oriented)[1], + as.logical(check)[1], + projection, + as.double(tessellate_tol[1]) + ), + "s2_geography_writer" + ) +} + +#' @rdname wk_handle.s2_geography +#' @importFrom wk wk_writer +#' @method wk_writer s2_geography +#' @export +wk_writer.s2_geography <- function(handleable, ...) { + s2_geography_writer() +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_trans_point <- function() { + wk::new_wk_trans(.Call(c_s2_trans_s2_point_new)) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_trans_lnglat <- function() { + wk::new_wk_trans(.Call(c_s2_trans_s2_lnglat_new)) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_projection_plate_carree <- function(x_scale = 180) { + structure( + .Call(c_s2_projection_plate_carree, as.double(x_scale)[1]), + class = "s2_projection" + ) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_projection_mercator <- function(x_scale = 20037508.3427892) { + structure( + .Call(c_s2_projection_mercator, as.double(x_scale)[1]), + class = "s2_projection" + ) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_hemisphere <- function(centre) { + cap_to_polygon(centre, pi / 2) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_world_plate_carree <- function(epsilon_east_west = 0, epsilon_north_south = 0) { + s2_make_polygon( + c( + -180 + epsilon_east_west, 0, 180 - epsilon_east_west, + 180 - epsilon_east_west, 180 - epsilon_east_west, 0, + -180 + epsilon_east_west, -180 + epsilon_east_west + ), + c( + -90 + epsilon_north_south, -90 + epsilon_north_south, + -90 + epsilon_north_south, 0, 90 - epsilon_north_south, + 90 - epsilon_north_south, 90 - epsilon_north_south, 0 + ), + oriented = TRUE + ) +} + +#' @rdname wk_handle.s2_geography +#' @export +s2_projection_orthographic <- function(centre = s2_lnglat(0, 0)) { + centre <- as_s2_lnglat(centre) + centre <- as.matrix(centre) + + structure( + .Call(c_s2_projection_orthographic, centre[1:2]), + class = "s2_projection" + ) +} diff --git a/R/zzz.R b/R/zzz.R new file mode 100644 index 0000000..acd3ba8 --- /dev/null +++ b/R/zzz.R @@ -0,0 +1,72 @@ + +# nocov start +.onLoad <- function(...) { + # call c++ init + cpp_s2_init() + + # dynamically register vctrs dependencies + for (cls in c("s2_geography", "s2_cell", "s2_cell_union")) { + s3_register("vctrs::vec_proxy", cls) + s3_register("vctrs::vec_restore", cls) + s3_register("vctrs::vec_ptype_abbr", cls) + } +} + +s3_register <- function(generic, class, method = NULL) { + stopifnot(is.character(generic), length(generic) == 1) + stopifnot(is.character(class), length(class) == 1) + + pieces <- strsplit(generic, "::")[[1]] + stopifnot(length(pieces) == 2) + package <- pieces[[1]] + generic <- pieces[[2]] + + caller <- parent.frame() + + get_method_env <- function() { + top <- topenv(caller) + if (isNamespace(top)) { + asNamespace(environmentName(top)) + } else { + caller + } + } + get_method <- function(method, env) { + if (is.null(method)) { + get(paste0(generic, ".", class), envir = get_method_env()) + } else { + method + } + } + + method_fn <- get_method(method) + stopifnot(is.function(method_fn)) + + # Always register hook in case package is later unloaded & reloaded + setHook( + packageEvent(package, "onLoad"), + function(...) { + ns <- asNamespace(package) + + # Refresh the method, it might have been updated by `devtools::load_all()` + method_fn <- get_method(method) + + registerS3method(generic, class, method_fn, envir = ns) + } + ) + + # Avoid registration failures during loading (pkgload or regular) + if (!isNamespaceLoaded(package)) { + return(invisible()) + } + + envir <- asNamespace(package) + + # Only register if generic can be accessed + if (exists(generic, envir)) { + registerS3method(generic, class, method_fn, envir = envir) + } + + invisible() +} +# nocov end diff --git a/README.md b/README.md new file mode 100644 index 0000000..b798e46 --- /dev/null +++ b/README.md @@ -0,0 +1,184 @@ + + + +# s2 + + + +![R-CMD-check](https://github.com/r-spatial/s2/workflows/R-CMD-check/badge.svg) +[![codecov](https://codecov.io/gh/r-spatial/s2/branch/master/graph/badge.svg)](https://app.codecov.io/gh/r-spatial/s2) +[![CRAN](http://www.r-pkg.org/badges/version/s2)](https://cran.r-project.org/package=s2) +[![Downloads](http://cranlogs.r-pkg.org/badges/s2?color=brightgreen)](https://www.r-pkg.org/pkg/s2) + + +The s2 R package provides bindings to Google’s +[S2Geometry](https://s2geometry.io) library. The package exposes an API +similar to Google’s [BigQuery Geography +API](https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions), +whose functions also operate on spherical geometries. Package +[sf](https://cran.r-project.org/package=sf) uses this package by default +for nearly all its geometrical operations on objects with ellipsoidal +(unprojected) coordinates; in cases where it doesn’t, such as +`st_relate()`, it emits a warning. + +This package is a complete rewrite of an earlier CRAN package s2 with +versions up to 0.4-2, for which the sources are found +[here](https://github.com/spatstat/s2/). + +## Installation + +You can install the released version of s2 from +[CRAN](https://CRAN.R-project.org) with: + +``` r +install.packages("s2") +``` + +And the development version from [GitHub](https://github.com/) with: + +``` r +# install.packages("remotes") +remotes::install_github("r-spatial/s2") +``` + +## Example + +The s2 package provides geometry transformers and predicates similar to +those found in [GEOS](https://trac.osgeo.org/geos/), except instead of +assuming a planar geometry, s2’s functions work in latitude and +longitude and assume a spherical geometry: + +``` r +library(s2) + +s2_contains( + # polygon containing much of the northern hemisphere + "POLYGON ((-63.5 44.6, -149.75 61.20, 116.4 40.2, 13.5 52.51, -63.5 44.6))", + # ...should contain the north pole + "POINT (0 90)" +) +#> [1] TRUE +``` + +The [sf package](https://r-spatial.github.io/sf/) uses s2 for geographic +coordinates with `sf::sf_use_s2(TRUE)`, and will become the default +after sf version 1.0.0. The sf package also supports creating s2 vectors +using `as_s2_geography()`: + +``` r +library(dplyr) +library(sf) + +nc_s2 <- read_sf(system.file("shape/nc.shp", package = "sf")) %>% + mutate(geometry = as_s2_geography(geometry)) %>% + as_tibble() %>% + select(NAME, geometry) + +nc_s2 +#> # A tibble: 100 × 2 +#> NAME geometry +#> +#> 1 Ashe POLYGON ((-81.4528885 36.2395859, -81.4310379 36.2607193, -81.41… +#> 2 Alleghany POLYGON ((-81.1766739 36.4154434, -81.1533661 36.4247398, -81.13… +#> 3 Surry POLYGON ((-80.4530106 36.2570877, -80.4353104 36.5510445, -80.61… +#> 4 Currituck MULTIPOLYGON (((-75.9419327 36.2943382, -75.9575119 36.2594528, … +#> 5 Northampton POLYGON ((-77.1419601 36.4170647, -77.1393204 36.4564781, -77.12… +#> 6 Hertford POLYGON ((-76.7074966 36.2661324, -76.7413483 36.3151665, -76.92… +#> 7 Camden POLYGON ((-76.0173492 36.3377304, -76.0328751 36.3359756, -76.04… +#> 8 Gates POLYGON ((-76.46035 36.3738976, -76.5024643 36.4522858, -76.4983… +#> 9 Warren POLYGON ((-78.1347198 36.2365837, -78.1096268 36.2135086, -78.05… +#> 10 Stokes POLYGON ((-80.0240555 36.5450249, -80.0480957 36.5471344, -80.43… +#> # … with 90 more rows +``` + +Use accessors to extract information about geometries: + +``` r +nc_s2 %>% + mutate( + area = s2_area(geometry), + perimeter = s2_perimeter(geometry) + ) +#> # A tibble: 100 × 4 +#> NAME geometry area perimeter +#> +#> 1 Ashe POLYGON ((-81.4528885 36.2395859, -81.4310379 3… 1.14e9 141627. +#> 2 Alleghany POLYGON ((-81.1766739 36.4154434, -81.1533661 3… 6.11e8 119876. +#> 3 Surry POLYGON ((-80.4530106 36.2570877, -80.4353104 3… 1.42e9 160458. +#> 4 Currituck MULTIPOLYGON (((-75.9419327 36.2943382, -75.957… 6.94e8 301644. +#> 5 Northampton POLYGON ((-77.1419601 36.4170647, -77.1393204 3… 1.52e9 211794. +#> 6 Hertford POLYGON ((-76.7074966 36.2661324, -76.7413483 3… 9.68e8 160780. +#> 7 Camden POLYGON ((-76.0173492 36.3377304, -76.0328751 3… 6.16e8 150430. +#> 8 Gates POLYGON ((-76.46035 36.3738976, -76.5024643 36.… 9.03e8 123170. +#> 9 Warren POLYGON ((-78.1347198 36.2365837, -78.1096268 3… 1.18e9 141073. +#> 10 Stokes POLYGON ((-80.0240555 36.5450249, -80.0480957 3… 1.23e9 140583. +#> # … with 90 more rows +``` + +Use predicates to subset vectors: + +``` r +nc_s2 %>% + filter(s2_contains(geometry, "POINT (-80.9313 35.6196)")) +#> # A tibble: 1 × 2 +#> NAME geometry +#> +#> 1 Catawba POLYGON ((-80.9312744 35.6195908, -81.0035782 35.6970558, -81.0547791… +``` + +Use transformers to create new geometries: + +``` r +nc_s2 %>% + mutate(geometry = s2_boundary(geometry)) +#> # A tibble: 100 × 2 +#> NAME geometry +#> +#> 1 Ashe LINESTRING (-81.4528885 36.2395859, -81.4310379 36.2607193, -81.… +#> 2 Alleghany LINESTRING (-81.1766739 36.4154434, -81.1533661 36.4247398, -81.… +#> 3 Surry LINESTRING (-80.4530106 36.2570877, -80.4353104 36.5510445, -80.… +#> 4 Currituck MULTILINESTRING ((-75.9419327 36.2943382, -75.9575119 36.2594528… +#> 5 Northampton LINESTRING (-77.1419601 36.4170647, -77.1393204 36.4564781, -77.… +#> 6 Hertford LINESTRING (-76.7074966 36.2661324, -76.7413483 36.3151665, -76.… +#> 7 Camden LINESTRING (-76.0173492 36.3377304, -76.0328751 36.3359756, -76.… +#> 8 Gates LINESTRING (-76.46035 36.3738976, -76.5024643 36.4522858, -76.49… +#> 9 Warren LINESTRING (-78.1347198 36.2365837, -78.1096268 36.2135086, -78.… +#> 10 Stokes LINESTRING (-80.0240555 36.5450249, -80.0480957 36.5471344, -80.… +#> # … with 90 more rows +``` + +Finally, use the WKB or WKT exporters to export to sf or some other +package: + +``` r +nc_s2 %>% + mutate(geometry = st_as_sfc(s2_as_binary(geometry))) %>% + st_as_sf() +#> Simple feature collection with 100 features and 1 field +#> Geometry type: GEOMETRY +#> Dimension: XY +#> Bounding box: xmin: -84.32385 ymin: 33.88199 xmax: -75.45698 ymax: 36.58965 +#> CRS: NA +#> # A tibble: 100 × 2 +#> NAME geometry +#> +#> 1 Ashe POLYGON ((-81.45289 36.23959, -81.43104 36.26072, -81.41233 36.2… +#> 2 Alleghany POLYGON ((-81.17667 36.41544, -81.15337 36.42474, -81.1384 36.41… +#> 3 Surry POLYGON ((-80.45301 36.25709, -80.43531 36.55104, -80.61105 36.5… +#> 4 Currituck MULTIPOLYGON (((-75.94193 36.29434, -75.95751 36.25945, -75.9137… +#> 5 Northampton POLYGON ((-77.14196 36.41706, -77.13932 36.45648, -77.12733 36.4… +#> 6 Hertford POLYGON ((-76.7075 36.26613, -76.74135 36.31517, -76.92408 36.39… +#> 7 Camden POLYGON ((-76.01735 36.33773, -76.03288 36.33598, -76.04395 36.3… +#> 8 Gates POLYGON ((-76.46035 36.3739, -76.50246 36.45229, -76.49834 36.50… +#> 9 Warren POLYGON ((-78.13472 36.23658, -78.10963 36.21351, -78.05835 36.2… +#> 10 Stokes POLYGON ((-80.02406 36.54502, -80.0481 36.54713, -80.43531 36.55… +#> # … with 90 more rows +``` + +## Acknowledgment + +This project gratefully acknowledges financial +[support](https://www.r-consortium.org/projects) from the + + + diff --git a/build/partial.rdb b/build/partial.rdb new file mode 100644 index 0000000..d53b128 Binary files /dev/null and b/build/partial.rdb differ diff --git a/cleanup b/cleanup new file mode 100755 index 0000000..3fd9cd2 --- /dev/null +++ b/cleanup @@ -0,0 +1,3 @@ +#!/bin/sh +rm -f src/Makevars configure.log autobrew +rm `find src -name *.o` diff --git a/configure b/configure new file mode 100755 index 0000000..fbf909f --- /dev/null +++ b/configure @@ -0,0 +1,125 @@ +# Anticonf (tm) script by Jeroen Ooms (2020) +# This script will query 'pkg-config' for the required cflags and ldflags. +# If pkg-config is unavailable or does not find the library, try setting +# INCLUDE_DIR and LIB_DIR manually via e.g: +# R CMD INSTALL --configure-vars='INCLUDE_DIR=/.../include LIB_DIR=/.../lib' + +# Library settings +PKG_CONFIG_NAME="openssl" +PKG_DEB_NAME="libssl-dev" +PKG_RPM_NAME="openssl-devel" +PKG_CSW_NAME="libssl_dev" +PKG_BREW_NAME="openssl@1.1" +PKG_TEST_FILE="tools/version.c" +PKG_LIBS="-lssl -lcrypto" +PKG_CFLAGS="" + +# Use pkg-config if available +pkg-config ${PKG_CONFIG_NAME} --atleast-version=1.0 2>/dev/null +if [ $? -eq 0 ]; then + PKGCONFIG_CFLAGS=`pkg-config --cflags ${PKG_CONFIG_NAME}` + PKGCONFIG_LIBS=`pkg-config --libs ${PKG_CONFIG_NAME}` +fi + +# Note that cflags may be empty in case of success +if [ "$INCLUDE_DIR" ] || [ "$LIB_DIR" ]; then + echo "Found INCLUDE_DIR and/or LIB_DIR!" + PKG_CFLAGS="-I$INCLUDE_DIR $PKG_CFLAGS" + PKG_LIBS="-L$LIB_DIR $PKG_LIBS" +elif [ "$PKGCONFIG_CFLAGS" ] || [ "$PKGCONFIG_LIBS" ]; then + echo "Found pkg-config cflags and libs!" + PKG_CFLAGS=${PKGCONFIG_CFLAGS} + PKG_LIBS=${PKGCONFIG_LIBS} +elif [ `uname` = "Darwin" ]; then + test ! "$CI" && brew --version 2>/dev/null + if [ $? -eq 0 ]; then + BREWDIR=`brew --prefix` + PKG_CFLAGS="-I$BREWDIR/opt/openssl/include -I$BREWDIR/opt/openssl@1.1/include" + PKG_LIBS="-L$BREWDIR/opt/openssl/lib -L$BREWDIR/opt/openssl@1.1/lib $PKG_LIBS" + else + curl -sfL "https://autobrew.github.io/scripts/$PKG_BREW_NAME" > autobrew + . ./autobrew + fi +fi + +# Find compiler +CC=`${R_HOME}/bin/R CMD config CC` +CFLAGS=`${R_HOME}/bin/R CMD config CFLAGS` +CPPFLAGS=`${R_HOME}/bin/R CMD config CPPFLAGS` + +# For debugging +echo "Testing compiler using PKG_CFLAGS=$PKG_CFLAGS" + +# Test configuration +${CC} ${CPPFLAGS} ${PKG_CFLAGS} ${CFLAGS} -E ${PKG_TEST_FILE} >/dev/null 2>configure.log + +# Customize the error +if [ $? -ne 0 ]; then + echo "--------------------------- [ANTICONF] --------------------------------" + echo "Configuration failed because $PKG_CONFIG_NAME was not found. Try installing:" + echo " * deb: $PKG_DEB_NAME (Debian, Ubuntu, etc)" + echo " * rpm: $PKG_RPM_NAME (Fedora, CentOS, RHEL)" + echo " * csw: $PKG_CSW_NAME (Solaris)" + echo " * brew: $PKG_BREW_NAME (Mac OSX)" + echo "If $PKG_CONFIG_NAME is already installed, check that 'pkg-config' is in your" + echo "PATH and PKG_CONFIG_PATH contains a $PKG_CONFIG_NAME.pc file. If pkg-config" + echo "is unavailable you can set INCLUDE_DIR and LIB_DIR manually via:" + echo "R CMD INSTALL --configure-vars='INCLUDE_DIR=... LIB_DIR=...'" + echo "-------------------------- [ERROR MESSAGE] ---------------------------" + cat configure.log + echo "--------------------------------------------------------------------" + exit 1 +fi + +# Try to link against the correct OpenSSL version +if [ -z "$AUTOBREW" ]; then +SONAME=`${CC} -E ${PKG_CFLAGS} src/tests/soname.h | sh | xargs` +if [ "$SONAME" ]; then +if [ `uname` = "Darwin" ]; then + PKG_LIBS_VERSIONED=`echo "${PKG_LIBS}" | sed "s/-lssl/-lssl.${SONAME}/" | sed "s/-lcrypto/-lcrypto.${SONAME}/"` +else + PKG_LIBS_VERSIONED=`echo "${PKG_LIBS}" | sed "s/-lssl/-l:libssl.so.${SONAME}/" | sed "s/-lcrypto/-l:libcrypto.so.${SONAME}/"` +fi + +# Test if versioned linking works +${CC} ${PKG_CFLAGS} src/tests/main.c ${PKG_LIBS_VERSIONED} -o src/main.exe 2>/dev/null +if [ $? -eq 0 ]; then PKG_LIBS="${PKG_LIBS_VERSIONED}"; fi +rm src/main.exe || true + +# Suppress opensslv3 warnings for now +if [ "$SONAME" = "3" ]; then +PKG_CFLAGS="$PKG_CFLAGS -DOPENSSL_SUPPRESS_DEPRECATED" +fi + +fi #SONAME +fi #AUTOBREW + +# Define system endianness (compile-time endianness using system/compiler +# defines isn't detected on Solaris) +# based on endian detection from the feather package by @hadley +R_ENDIAN=`${R_HOME}/bin/Rscript -e 'cat(.Platform$endian)'` +# Trim off any warning messages that Rscript appends in front of the platform endianness +R_ENDIAN=`expr "$R_ENDIAN" : '.*\(little\)$'` +SYS_ENDIAN="" +if [ "$R_ENDIAN" = "little" ]; then + PKG_CFLAGS="$PKG_CFLAGS -DIS_LITTLE_ENDIAN" +else + PKG_CFLAGS="$PKG_CFLAGS -DIS_BIG_ENDIAN" +fi + +# From apache/arrow/r/configure: +# If on Raspberry Pi, need to manually link against latomic +# See: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81358 for similar example +if grep raspbian /etc/os-release >/dev/null 2>&1; then + PKG_LIBS="-latomic $PKG_LIBS" +fi + +echo "Using PKG_LIBS=$PKG_LIBS" +echo "Using PKG_CFLAGS=$PKG_CFLAGS" + + +# Write to Makevars +sed -e "s|@cflags@|$PKG_CFLAGS|" -e "s|@libs@|$PKG_LIBS|" src/Makevars.in > src/Makevars + +# Success +exit 0 diff --git a/configure.win b/configure.win new file mode 100755 index 0000000..e69de29 diff --git a/data/s2_data_example_wkt.rda b/data/s2_data_example_wkt.rda new file mode 100644 index 0000000..23820bb Binary files /dev/null and b/data/s2_data_example_wkt.rda differ diff --git a/data/s2_data_tbl_cities.rda b/data/s2_data_tbl_cities.rda new file mode 100644 index 0000000..1362234 Binary files /dev/null and b/data/s2_data_tbl_cities.rda differ diff --git a/data/s2_data_tbl_countries.rda b/data/s2_data_tbl_countries.rda new file mode 100644 index 0000000..119290e Binary files /dev/null and b/data/s2_data_tbl_countries.rda differ diff --git a/data/s2_data_tbl_timezones.rda b/data/s2_data_tbl_timezones.rda new file mode 100644 index 0000000..584b1dc Binary files /dev/null and b/data/s2_data_tbl_timezones.rda differ diff --git a/inst/extdata/emptyfile b/inst/extdata/emptyfile new file mode 100644 index 0000000..e69de29 diff --git a/man/as_s2_geography.Rd b/man/as_s2_geography.Rd new file mode 100644 index 0000000..8fa0402 --- /dev/null +++ b/man/as_s2_geography.Rd @@ -0,0 +1,74 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-geography.R +\name{as_s2_geography} +\alias{as_s2_geography} +\alias{s2_geography} +\alias{as_s2_geography.s2_geography} +\alias{as_s2_geography.wk_xy} +\alias{as_s2_geography.wk_wkb} +\alias{as_s2_geography.WKB} +\alias{as_s2_geography.blob} +\alias{as_s2_geography.wk_wkt} +\alias{as_s2_geography.character} +\alias{as_s2_geography.logical} +\alias{as_wkb.s2_geography} +\alias{as_wkt.s2_geography} +\title{Create an S2 Geography Vector} +\usage{ +as_s2_geography(x, ...) + +s2_geography() + +\method{as_s2_geography}{s2_geography}(x, ...) + +\method{as_s2_geography}{wk_xy}(x, ...) + +\method{as_s2_geography}{wk_wkb}(x, ..., oriented = FALSE, check = TRUE) + +\method{as_s2_geography}{WKB}(x, ..., oriented = FALSE, check = TRUE) + +\method{as_s2_geography}{blob}(x, ..., oriented = FALSE, check = TRUE) + +\method{as_s2_geography}{wk_wkt}(x, ..., oriented = FALSE, check = TRUE) + +\method{as_s2_geography}{character}(x, ..., oriented = FALSE, check = TRUE) + +\method{as_s2_geography}{logical}(x, ...) + +\method{as_wkb}{s2_geography}(x, ...) + +\method{as_wkt}{s2_geography}(x, ...) +} +\arguments{ +\item{x}{An object that can be converted to an s2_geography vector} + +\item{...}{Unused} + +\item{oriented}{TRUE if polygon ring directions are known to be correct +(i.e., exterior rings are defined counter clockwise and interior +rings are defined clockwise).} + +\item{check}{Use \code{check = FALSE} to skip error on invalid geometries} +} +\value{ +An object with class s2_geography +} +\description{ +Geography vectors are arrays of points, lines, polygons, and/or collections +of these. Geography vectors assume coordinates are longitude and latitude +on a perfect sphere. +} +\details{ +The coercion function \code{\link[=as_s2_geography]{as_s2_geography()}} is used to wrap the input +of most functions in the s2 package so that you can use other objects with +an unambiguious interpretation as a geography vector. Geography vectors +have a minimal \link[vctrs:vctrs-package]{vctrs} implementation, so you can +use these objects in tibble, dplyr, and other packages that use the vctrs +framework. +} +\seealso{ +\code{\link[=s2_geog_from_wkb]{s2_geog_from_wkb()}}, \code{\link[=s2_geog_from_text]{s2_geog_from_text()}}, \code{\link[=s2_geog_point]{s2_geog_point()}}, +\code{\link[=s2_make_line]{s2_make_line()}}, \code{\link[=s2_make_polygon]{s2_make_polygon()}} for other ways to +create geography vectors, and \code{\link[=s2_as_binary]{s2_as_binary()}} and \code{\link[=s2_as_text]{s2_as_text()}} +for other ways to export them. +} diff --git a/man/figures/rc300.png b/man/figures/rc300.png new file mode 100644 index 0000000..cdb066a Binary files /dev/null and b/man/figures/rc300.png differ diff --git a/man/s2-package.Rd b/man/s2-package.Rd new file mode 100644 index 0000000..980c7dc --- /dev/null +++ b/man/s2-package.Rd @@ -0,0 +1,37 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-package.R +\docType{package} +\name{s2-package} +\alias{s2} +\alias{s2-package} +\title{s2: Spherical Geometry Operators Using the S2 Geometry Library} +\description{ +Provides R bindings for Google's s2 library for geometric calculations on the sphere. High-performance constructors and exporters provide high compatibility with existing spatial packages, transformers construct new geometries from existing geometries, predicates provide a means to select geometries based on spatial relationships, and accessors extract information about geometries. +} +\seealso{ +Useful links: +\itemize{ + \item \url{https://r-spatial.github.io/s2/} + \item \url{https://github.com/r-spatial/s2} + \item \url{https://s2geometry.io/} + \item Report bugs at \url{https://github.com/r-spatial/s2/issues} +} + +} +\author{ +\strong{Maintainer}: Edzer Pebesma \email{edzer.pebesma@uni-muenster.de} (\href{https://orcid.org/0000-0001-8049-7069}{ORCID}) + +Authors: +\itemize{ + \item Dewey Dunnington \email{dewey@fishandwhistle.net} (\href{https://orcid.org/0000-0002-9415-4582}{ORCID}) + \item Ege Rubak \email{rubak@math.aau.dk} +} + +Other contributors: +\itemize{ + \item Jeroen Ooms \email{jeroen.ooms@stat.ucla.edu} (configure script) [contributor] + \item Google, Inc. (Original s2geometry.io source code) [copyright holder] +} + +} +\keyword{internal} diff --git a/man/s2_boundary.Rd b/man/s2_boundary.Rd new file mode 100644 index 0000000..d6a5665 --- /dev/null +++ b/man/s2_boundary.Rd @@ -0,0 +1,215 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-transformers.R +\name{s2_boundary} +\alias{s2_boundary} +\alias{s2_centroid} +\alias{s2_closest_point} +\alias{s2_minimum_clearance_line_between} +\alias{s2_difference} +\alias{s2_sym_difference} +\alias{s2_intersection} +\alias{s2_union} +\alias{s2_snap_to_grid} +\alias{s2_simplify} +\alias{s2_rebuild} +\alias{s2_buffer_cells} +\alias{s2_convex_hull} +\alias{s2_centroid_agg} +\alias{s2_coverage_union_agg} +\alias{s2_rebuild_agg} +\alias{s2_union_agg} +\alias{s2_convex_hull_agg} +\alias{s2_point_on_surface} +\title{S2 Geography Transformations} +\usage{ +s2_boundary(x) + +s2_centroid(x) + +s2_closest_point(x, y) + +s2_minimum_clearance_line_between(x, y) + +s2_difference(x, y, options = s2_options()) + +s2_sym_difference(x, y, options = s2_options()) + +s2_intersection(x, y, options = s2_options()) + +s2_union(x, y = NULL, options = s2_options()) + +s2_snap_to_grid(x, grid_size) + +s2_simplify(x, tolerance, radius = s2_earth_radius_meters()) + +s2_rebuild(x, options = s2_options()) + +s2_buffer_cells( + x, + distance, + max_cells = 1000, + min_level = -1, + radius = s2_earth_radius_meters() +) + +s2_convex_hull(x) + +s2_centroid_agg(x, na.rm = FALSE) + +s2_coverage_union_agg(x, options = s2_options(), na.rm = FALSE) + +s2_rebuild_agg(x, options = s2_options(), na.rm = FALSE) + +s2_union_agg(x, options = s2_options(), na.rm = FALSE) + +s2_convex_hull_agg(x, na.rm = FALSE) + +s2_point_on_surface(x, na.rm = FALSE) +} +\arguments{ +\item{x, y}{\link[=as_s2_geography]{geography vectors}. These inputs +are passed to \code{\link[=as_s2_geography]{as_s2_geography()}}, so you can pass other objects +(e.g., character vectors of well-known text) directly.} + +\item{options}{An \code{\link[=s2_options]{s2_options()}} object describing the polygon/polyline +model to use and the snap level.} + +\item{grid_size}{The grid size to which coordinates should be snapped; +will be rounded to the nearest power of 10.} + +\item{tolerance}{The minimum distance between vertexes to use when +simplifying a geography.} + +\item{radius}{Radius of the earth. Defaults to the average radius of +the earth in meters as defined by \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}}.} + +\item{distance}{The distance to buffer, in units of \code{radius}.} + +\item{max_cells}{The maximum number of cells to approximate a buffer.} + +\item{min_level}{The minimum cell level used to approximate a buffer +(1 - 30). Setting this value too high will result in unnecessarily +large geographies, but may help improve buffers along long, narrow +regions.} + +\item{na.rm}{For aggregate calculations use \code{na.rm = TRUE} +to drop missing values.} +} +\description{ +These functions operate on one or more geography vectors and +return a geography vector. +} +\section{Model}{ + +The geometry model indicates whether or not a geometry includes its boundaries. +Boundaries of line geometries are its end points. +OPEN geometries do not contain their boundary (\code{model = "open"}); CLOSED +geometries (\code{model = "closed"}) contain their boundary; SEMI-OPEN geometries +(\code{model = "semi-open"}) contain half of their boundaries, such that when two polygons +do not overlap or two lines do not cross, no point exist that belong to +more than one of the geometries. (This latter form, half-closed, is +not present in the OpenGIS "simple feature access" (SFA) standard nor DE9-IM on +which that is based). The default values for \code{\link[=s2_contains]{s2_contains()}} (open) +and covers/covered_by (closed) correspond to the SFA standard specification +of these operators. +} + +\examples{ +# returns the boundary: +# empty for point, endpoints of a linestring, +# perimeter of a polygon +s2_boundary("POINT (-64 45)") +s2_boundary("LINESTRING (0 0, 10 0)") +s2_boundary("POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))") + +# returns the area-weighted centroid, element-wise +s2_centroid("POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))") +s2_centroid("LINESTRING (0 0, 10 0)") + +# s2_point_on_surface guarantees a point on surface +# Note: this is not the same as st_point_on_surface +s2_centroid("POLYGON ((0 0, 10 0, 1 1, 0 10, 0 0))") +s2_point_on_surface("POLYGON ((0 0, 10 0, 1 1, 0 10, 0 0))") + +# returns the unweighted centroid of the entire input +s2_centroid_agg(c("POINT (0 0)", "POINT (10 0)")) + +# returns the closest point on x to y +s2_closest_point( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POINT (0 90)" # north pole! +) + +# returns the shortest possible line between x and y +s2_minimum_clearance_line_between( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POINT (0 90)" # north pole! +) + +# binary operations: difference, symmetric difference, intersection and union +s2_difference( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", + # 32 bit platforms may need to set snap rounding + s2_options(snap = s2_snap_level(30)) +) + +s2_sym_difference( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", + # 32 bit platforms may need to set snap rounding + s2_options(snap = s2_snap_level(30)) +) + +s2_intersection( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", + # 32 bit platforms may need to set snap rounding + s2_options(snap = s2_snap_level(30)) +) + +s2_union( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))", + # 32 bit platforms may need to set snap rounding + s2_options(snap = s2_snap_level(30)) +) + +# s2_convex_hull_agg builds the convex hull of a list of geometries +s2_convex_hull_agg( + c( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))" + ) +) + +# use s2_union_agg() to aggregate geographies in a vector +s2_coverage_union_agg( + c( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((5 5, 15 5, 15 15, 5 15, 5 5))" + ), + # 32 bit platforms may need to set snap rounding + s2_options(snap = s2_snap_level(30)) +) + +# snap to grid rounds coordinates to a specified grid size +s2_snap_to_grid("POINT (0.333333333333 0.666666666666)", 1e-2) + + +} +\seealso{ +BigQuery's geography function reference: +\itemize{ +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_boundary}{ST_BOUNDARY} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_centroid}{ST_CENTROID} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_closestpoint}{ST_CLOSESTPOINT} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_difference}{ST_DIFFERENCE} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersection}{ST_INTERSECTION} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_union}{ST_UNION} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_snaptogrid}{ST_SNAPTOGRID} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_simplify}{ST_SIMPLIFY} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_union_agg}{ST_UNION_AGG} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#s2_centroid_agg}{ST_CENTROID_AGG} +} +} diff --git a/man/s2_bounds_cap.Rd b/man/s2_bounds_cap.Rd new file mode 100644 index 0000000..e4bd419 --- /dev/null +++ b/man/s2_bounds_cap.Rd @@ -0,0 +1,41 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-bounds.R +\name{s2_bounds_cap} +\alias{s2_bounds_cap} +\alias{s2_bounds_rect} +\title{Compute feature-wise and aggregate bounds} +\usage{ +s2_bounds_cap(x) + +s2_bounds_rect(x) +} +\arguments{ +\item{x}{\link[=as_s2_geography]{geography vectors}. These inputs +are passed to \code{\link[=as_s2_geography]{as_s2_geography()}}, so you can pass other objects +(e.g., character vectors of well-known text) directly.} +} +\value{ +Both functions return a \code{data.frame}: +\itemize{ +\item \code{\link[=s2_bounds_rect]{s2_bounds_rect()}}: Columns \code{minlng}, \code{minlat}, \code{maxlng}, \code{maxlat} (degrees) +\item \code{\link[=s2_bounds_cap]{s2_bounds_cap()}}: Columns \code{lng}, \code{lat}, \code{angle} (degrees) +} +} +\description{ +\code{\link[=s2_bounds_rect]{s2_bounds_rect()}} returns a bounding latitude-longitude +rectangle that contains the region; \code{\link[=s2_bounds_cap]{s2_bounds_cap()}} returns a bounding circle +represented by a centre point (lat, lng) and an angle. The bound may not be tight +for points, polylines and geometry collections. The rectangle returned may depend on +the order of points or polylines. \code{lng_lo} values larger than \code{lng_hi} indicate +regions that span the antimeridian, see the Fiji example. +} +\examples{ +s2_bounds_cap(s2_data_countries("Antarctica")) +s2_bounds_cap(s2_data_countries("Netherlands")) +s2_bounds_cap(s2_data_countries("Fiji")) + +s2_bounds_rect(s2_data_countries("Antarctica")) +s2_bounds_rect(s2_data_countries("Netherlands")) +s2_bounds_rect(s2_data_countries("Fiji")) + +} diff --git a/man/s2_cell.Rd b/man/s2_cell.Rd new file mode 100644 index 0000000..d525599 --- /dev/null +++ b/man/s2_cell.Rd @@ -0,0 +1,74 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-cell.R +\name{s2_cell} +\alias{s2_cell} +\alias{s2_cell_sentinel} +\alias{s2_cell_invalid} +\alias{as_s2_cell} +\alias{as_s2_cell.s2_cell} +\alias{as_s2_cell.character} +\alias{as_s2_cell.s2_geography} +\alias{as_s2_cell.wk_xy} +\alias{new_s2_cell} +\title{Create S2 Cell vectors} +\usage{ +s2_cell(x = character()) + +s2_cell_sentinel() + +s2_cell_invalid() + +s2_cell_sentinel() + +as_s2_cell(x, ...) + +\method{as_s2_cell}{s2_cell}(x, ...) + +\method{as_s2_cell}{character}(x, ...) + +\method{as_s2_cell}{s2_geography}(x, ...) + +\method{as_s2_cell}{wk_xy}(x, ...) + +new_s2_cell(x) +} +\arguments{ +\item{x}{The canonical S2 cell identifier as a character vector.} + +\item{...}{Passed to methods} +} +\value{ +An object of class s2_cell +} +\description{ +The S2 cell indexing system forms the basis for spatial indexing +in the S2 library. On their own, S2 cells can represent points +or areas. As a union, a vector of S2 cells can approximate a +line or polygon. These functions allow direct access to the +S2 cell indexing system and are designed to have minimal overhead +such that looping and recursion have acceptable performance +when used within R code. +} +\details{ +Under the hood, S2 cell vectors are represented in R as vectors +of type \code{\link[=double]{double()}}. This works because S2 cell identifiers are +64 bits wide, as are \code{double}s on all systems where R runs (The +same trick is used by the bit64 package to represent signed +64-bit integers). As a happy accident, \code{NA_real_} is not a valid +or meaningful cell identifier, so missing value support in the +way R users might expect is preserved. It is worth noting that +the underlying value of \code{s2_cell_sentinel()} would normally be +considered \code{NA}; however, as it is meaningful and useful when +programming with S2 cells, custom \code{is.na()} and comparison methods +are implemented such that \code{s2_cell_sentinel()} is greater than +all valid S2 cells and not considered missing. Users can and should +implement compiled code that uses the underlying bytes of the +vector, ensuring that the class of any returned object that should +be interpreted in this way is constructed with \code{new_s2_cell()}. +} +\examples{ +s2_cell("4b59a0cd83b5de49") +as_s2_cell(s2_lnglat(-64, 45)) +as_s2_cell(s2_data_cities("Ottawa")) + +} diff --git a/man/s2_cell_is_valid.Rd b/man/s2_cell_is_valid.Rd new file mode 100644 index 0000000..63b6668 --- /dev/null +++ b/man/s2_cell_is_valid.Rd @@ -0,0 +1,82 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-cell.R +\name{s2_cell_is_valid} +\alias{s2_cell_is_valid} +\alias{s2_cell_debug_string} +\alias{s2_cell_to_lnglat} +\alias{s2_cell_center} +\alias{s2_cell_boundary} +\alias{s2_cell_polygon} +\alias{s2_cell_vertex} +\alias{s2_cell_level} +\alias{s2_cell_is_leaf} +\alias{s2_cell_is_face} +\alias{s2_cell_area} +\alias{s2_cell_area_approx} +\alias{s2_cell_parent} +\alias{s2_cell_child} +\alias{s2_cell_edge_neighbour} +\alias{s2_cell_contains} +\alias{s2_cell_distance} +\alias{s2_cell_max_distance} +\alias{s2_cell_may_intersect} +\alias{s2_cell_common_ancestor_level} +\alias{s2_cell_common_ancestor_level_agg} +\title{S2 cell operators} +\usage{ +s2_cell_is_valid(x) + +s2_cell_debug_string(x) + +s2_cell_to_lnglat(x) + +s2_cell_center(x) + +s2_cell_boundary(x) + +s2_cell_polygon(x) + +s2_cell_vertex(x, k) + +s2_cell_level(x) + +s2_cell_is_leaf(x) + +s2_cell_is_face(x) + +s2_cell_area(x, radius = s2_earth_radius_meters()) + +s2_cell_area_approx(x, radius = s2_earth_radius_meters()) + +s2_cell_parent(x, level = -1L) + +s2_cell_child(x, k) + +s2_cell_edge_neighbour(x, k) + +s2_cell_contains(x, y) + +s2_cell_distance(x, y, radius = s2_earth_radius_meters()) + +s2_cell_max_distance(x, y, radius = s2_earth_radius_meters()) + +s2_cell_may_intersect(x, y) + +s2_cell_common_ancestor_level(x, y) + +s2_cell_common_ancestor_level_agg(x, na.rm = FALSE) +} +\arguments{ +\item{x, y}{An \code{\link[=s2_cell]{s2_cell()}} vector} + +\item{k}{An integer between 1 and 4} + +\item{radius}{The radius to use (e.g., \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}})} + +\item{level}{An integer between 0 and 30, inclusive.} + +\item{na.rm}{Remove NAs prior to computing aggregate?} +} +\description{ +S2 cell operators +} diff --git a/man/s2_cell_union.Rd b/man/s2_cell_union.Rd new file mode 100644 index 0000000..4af519a --- /dev/null +++ b/man/s2_cell_union.Rd @@ -0,0 +1,34 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-cell-union.R +\name{s2_cell_union} +\alias{s2_cell_union} +\alias{as_s2_geography.s2_cell_union} +\alias{as_s2_cell_union} +\alias{as_s2_cell_union.s2_cell_union} +\alias{as_s2_cell_union.s2_cell} +\alias{as_s2_cell_union.character} +\title{Create S2 Cell Union vectors} +\usage{ +s2_cell_union(x = list()) + +\method{as_s2_geography}{s2_cell_union}(x, ...) + +as_s2_cell_union(x, ...) + +\method{as_s2_cell_union}{s2_cell_union}(x, ...) + +\method{as_s2_cell_union}{s2_cell}(x, ...) + +\method{as_s2_cell_union}{character}(x, ...) +} +\arguments{ +\item{x}{A \code{list()} of \code{\link[=s2_cell]{s2_cell()}} vectors.} + +\item{...}{Passed to S3 methods} +} +\value{ +An object of class "s2_cell_union". +} +\description{ +Create S2 Cell Union vectors +} diff --git a/man/s2_cell_union_normalize.Rd b/man/s2_cell_union_normalize.Rd new file mode 100644 index 0000000..208da9a --- /dev/null +++ b/man/s2_cell_union_normalize.Rd @@ -0,0 +1,66 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-cell-union.R +\name{s2_cell_union_normalize} +\alias{s2_cell_union_normalize} +\alias{s2_cell_union_contains} +\alias{s2_cell_union_intersects} +\alias{s2_cell_union_intersection} +\alias{s2_cell_union_union} +\alias{s2_cell_union_difference} +\alias{s2_covering_cell_ids} +\alias{s2_covering_cell_ids_agg} +\title{S2 cell union operators} +\usage{ +s2_cell_union_normalize(x) + +s2_cell_union_contains(x, y) + +s2_cell_union_intersects(x, y) + +s2_cell_union_intersection(x, y) + +s2_cell_union_union(x, y) + +s2_cell_union_difference(x, y) + +s2_covering_cell_ids( + x, + min_level = 0, + max_level = 30, + max_cells = 8, + buffer = 0, + interior = FALSE, + radius = s2_earth_radius_meters() +) + +s2_covering_cell_ids_agg( + x, + min_level = 0, + max_level = 30, + max_cells = 8, + buffer = 0, + interior = FALSE, + radius = s2_earth_radius_meters(), + na.rm = FALSE +) +} +\arguments{ +\item{x, y}{An \link[=as_s2_geography]{s2_geography} or \code{\link[=s2_cell_union]{s2_cell_union()}}.} + +\item{min_level, max_level}{The minimum and maximum levels to constrain the +covering.} + +\item{max_cells}{The maximum number of cells in the covering. Defaults to +8.} + +\item{buffer}{A distance to buffer outside the geography} + +\item{interior}{Use \code{TRUE} to force the covering inside the geography.} + +\item{radius}{The radius to use (e.g., \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}})} + +\item{na.rm}{Remove NAs prior to computing aggregate?} +} +\description{ +S2 cell union operators +} diff --git a/man/s2_closest_feature.Rd b/man/s2_closest_feature.Rd new file mode 100644 index 0000000..093109e --- /dev/null +++ b/man/s2_closest_feature.Rd @@ -0,0 +1,132 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-matrix.R +\name{s2_closest_feature} +\alias{s2_closest_feature} +\alias{s2_closest_edges} +\alias{s2_farthest_feature} +\alias{s2_distance_matrix} +\alias{s2_max_distance_matrix} +\alias{s2_contains_matrix} +\alias{s2_within_matrix} +\alias{s2_covers_matrix} +\alias{s2_covered_by_matrix} +\alias{s2_intersects_matrix} +\alias{s2_disjoint_matrix} +\alias{s2_equals_matrix} +\alias{s2_touches_matrix} +\alias{s2_dwithin_matrix} +\alias{s2_may_intersect_matrix} +\title{Matrix Functions} +\usage{ +s2_closest_feature(x, y) + +s2_closest_edges( + x, + y, + k, + min_distance = -1, + max_distance = Inf, + radius = s2_earth_radius_meters() +) + +s2_farthest_feature(x, y) + +s2_distance_matrix(x, y, radius = s2_earth_radius_meters()) + +s2_max_distance_matrix(x, y, radius = s2_earth_radius_meters()) + +s2_contains_matrix(x, y, options = s2_options(model = "open")) + +s2_within_matrix(x, y, options = s2_options(model = "open")) + +s2_covers_matrix(x, y, options = s2_options(model = "closed")) + +s2_covered_by_matrix(x, y, options = s2_options(model = "closed")) + +s2_intersects_matrix(x, y, options = s2_options()) + +s2_disjoint_matrix(x, y, options = s2_options()) + +s2_equals_matrix(x, y, options = s2_options()) + +s2_touches_matrix(x, y, options = s2_options()) + +s2_dwithin_matrix(x, y, distance, radius = s2_earth_radius_meters()) + +s2_may_intersect_matrix(x, y, max_edges_per_cell = 50, max_feature_cells = 4) +} +\arguments{ +\item{x, y}{Geography vectors, coerced using \code{\link[=as_s2_geography]{as_s2_geography()}}. +\code{x} is considered the source, where as \code{y} is considered the target.} + +\item{k}{The number of closest edges to consider when searching. Note +that in S2 a point is also considered an edge.} + +\item{min_distance}{The minimum distance to consider when searching for +edges. This filter is applied after the search is complete (i.e., +may cause fewer than \code{k} values to be returned).} + +\item{max_distance}{The maximum distance to consider when searching for +edges. This filter is applied before the search.} + +\item{radius}{Radius of the earth. Defaults to the average radius of +the earth in meters as defined by \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}}.} + +\item{options}{An \code{\link[=s2_options]{s2_options()}} object describing the polygon/polyline +model to use and the snap level.} + +\item{distance}{A distance on the surface of the earth in the same units +as \code{radius}.} + +\item{max_edges_per_cell}{For \code{\link[=s2_may_intersect_matrix]{s2_may_intersect_matrix()}}, +this values controls the nature of the index on \code{y}, with higher values +leading to coarser index. Values should be between 10 and 50; the default +of 50 is adequate for most use cases, but for specialized operations users +may wish to use a lower value to increase performance.} + +\item{max_feature_cells}{For \code{\link[=s2_may_intersect_matrix]{s2_may_intersect_matrix()}}, this value +controls the approximation of \code{x} used to identify potential intersections +on \code{y}. The default value of 4 gives the best performance for most operations, +but for specialized operations users may wish to use a higher value to increase +performance.} +} +\value{ +A vector of length \code{x}. +} +\description{ +These functions are similar to accessors and predicates, but instead of +recycling \code{x} and \code{y} to a common length and returning a vector of that +length, these functions return a vector of length \code{x} with each element +\code{i} containing information about how the entire vector \code{y} relates to +the feature at \code{x[i]}. +} +\examples{ +city_names <- c("Vatican City", "San Marino", "Luxembourg") +cities <- s2_data_cities(city_names) +country_names <- s2_data_tbl_countries$name +countries <- s2_data_countries() + +# closest feature returns y indices of the closest feature +# for each feature in x +country_names[s2_closest_feature(cities, countries)] + +# farthest feature returns y indices of the farthest feature +# for each feature in x +country_names[s2_farthest_feature(cities, countries)] + +# use s2_closest_edges() to find the k-nearest neighbours +nearest <- s2_closest_edges(cities, cities, k = 2, min_distance = 0) +city_names +city_names[unlist(nearest)] + +# predicate matrices +country_names[s2_intersects_matrix(cities, countries)[[1]]] + +# distance matrices +s2_distance_matrix(cities, cities) +s2_max_distance_matrix(cities, countries[1:4]) + +} +\seealso{ +See pairwise predicate functions (e.g., \code{\link[=s2_intersects]{s2_intersects()}}). +} diff --git a/man/s2_contains.Rd b/man/s2_contains.Rd new file mode 100644 index 0000000..514d01e --- /dev/null +++ b/man/s2_contains.Rd @@ -0,0 +1,169 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-predicates.R +\name{s2_contains} +\alias{s2_contains} +\alias{s2_within} +\alias{s2_covered_by} +\alias{s2_covers} +\alias{s2_disjoint} +\alias{s2_intersects} +\alias{s2_equals} +\alias{s2_intersects_box} +\alias{s2_touches} +\alias{s2_dwithin} +\alias{s2_prepared_dwithin} +\title{S2 Geography Predicates} +\usage{ +s2_contains(x, y, options = s2_options(model = "open")) + +s2_within(x, y, options = s2_options(model = "open")) + +s2_covered_by(x, y, options = s2_options(model = "closed")) + +s2_covers(x, y, options = s2_options(model = "closed")) + +s2_disjoint(x, y, options = s2_options()) + +s2_intersects(x, y, options = s2_options()) + +s2_equals(x, y, options = s2_options()) + +s2_intersects_box( + x, + lng1, + lat1, + lng2, + lat2, + detail = 1000, + options = s2_options() +) + +s2_touches(x, y, options = s2_options()) + +s2_dwithin(x, y, distance, radius = s2_earth_radius_meters()) + +s2_prepared_dwithin(x, y, distance, radius = s2_earth_radius_meters()) +} +\arguments{ +\item{x, y}{\link[=as_s2_geography]{geography vectors}. These inputs +are passed to \code{\link[=as_s2_geography]{as_s2_geography()}}, so you can pass other objects +(e.g., character vectors of well-known text) directly.} + +\item{options}{An \code{\link[=s2_options]{s2_options()}} object describing the polygon/polyline +model to use and the snap level.} + +\item{lng1, lat1, lng2, lat2}{A latitude/longitude range} + +\item{detail}{The number of points with which to approximate +non-geodesic edges.} + +\item{distance}{A distance on the surface of the earth in the same units +as \code{radius}.} + +\item{radius}{Radius of the earth. Defaults to the average radius of +the earth in meters as defined by \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}}.} +} +\description{ +These functions operate two geography vectors (pairwise), and return +a logical vector. +} +\section{Model}{ + +The geometry model indicates whether or not a geometry includes its boundaries. +Boundaries of line geometries are its end points. +OPEN geometries do not contain their boundary (\code{model = "open"}); CLOSED +geometries (\code{model = "closed"}) contain their boundary; SEMI-OPEN geometries +(\code{model = "semi-open"}) contain half of their boundaries, such that when two polygons +do not overlap or two lines do not cross, no point exist that belong to +more than one of the geometries. (This latter form, half-closed, is +not present in the OpenGIS "simple feature access" (SFA) standard nor DE9-IM on +which that is based). The default values for \code{\link[=s2_contains]{s2_contains()}} (open) +and covers/covered_by (closed) correspond to the SFA standard specification +of these operators. +} + +\examples{ +s2_contains( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_within( + c("POINT (5 5)", "POINT (-1 1)"), + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))" +) + +s2_covered_by( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_covers( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_disjoint( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_intersects( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_equals( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + "POLYGON ((10 0, 10 10, 0 10, 0 0, 10 0))", + "POLYGON ((-1 -1, 10 0, 10 10, 0 10, -1 -1))" + ) +) + +s2_intersects( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)") +) + +s2_intersects_box( + c("POINT (5 5)", "POINT (-1 1)"), + 0, 0, 10, 10 +) + +s2_touches( + "POLYGON ((0 0, 0 1, 1 1, 0 0))", + c("POINT (0 0)", "POINT (0.5 0.75)", "POINT (0 0.5)") +) + +s2_dwithin( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)"), + 0 # distance in meters +) + +s2_dwithin( + "POLYGON ((0 0, 10 0, 10 10, 0 10, 0 0))", + c("POINT (5 5)", "POINT (-1 1)"), + 1e6 # distance in meters +) + +} +\seealso{ +Matrix versions of these predicates (e.g., \code{\link[=s2_intersects_matrix]{s2_intersects_matrix()}}). + +BigQuery's geography function reference: +\itemize{ +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_contains}{ST_CONTAINS} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_coveredby}{ST_COVEREDBY} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_covers}{ST_COVERS} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_disjoint}{ST_DISJOINT} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_equals}{ST_EQUALS} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersects}{ST_INTERSECTS} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_intersectsbox}{ST_INTERSECTSBOX} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_touches}{ST_TOUCHES} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_within}{ST_WITHIN} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_dwithin}{ST_DWITHIN} +} +} diff --git a/man/s2_data_example_wkt.Rd b/man/s2_data_example_wkt.Rd new file mode 100644 index 0000000..1dea278 --- /dev/null +++ b/man/s2_data_example_wkt.Rd @@ -0,0 +1,17 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/data.R +\docType{data} +\name{s2_data_example_wkt} +\alias{s2_data_example_wkt} +\title{Example Geometries} +\format{ +An object of class \code{list} of length 29. +} +\usage{ +s2_data_example_wkt +} +\description{ +These geometries are toy examples useful for testing various coordinate +shuffling operations in the s2 package. +} +\keyword{datasets} diff --git a/man/s2_data_tbl_countries.Rd b/man/s2_data_tbl_countries.Rd new file mode 100644 index 0000000..a2a6ce8 --- /dev/null +++ b/man/s2_data_tbl_countries.Rd @@ -0,0 +1,59 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/data.R +\docType{data} +\name{s2_data_tbl_countries} +\alias{s2_data_tbl_countries} +\alias{s2_data_tbl_timezones} +\alias{s2_data_tbl_cities} +\alias{s2_data_countries} +\alias{s2_data_timezones} +\alias{s2_data_cities} +\title{Low-resolution world boundaries, timezones, and cities} +\format{ +A data.frame with columns \code{name} (character), and +\code{geometry} (wk_wkb) + +An object of class \code{data.frame} with 120 rows and 2 columns. + +An object of class \code{data.frame} with 243 rows and 3 columns. +} +\source{ +\href{https://www.naturalearthdata.com/}{Natural Earth Data} +} +\usage{ +s2_data_tbl_countries + +s2_data_tbl_timezones + +s2_data_tbl_cities + +s2_data_countries(name = NULL) + +s2_data_timezones(utc_offset_min = NULL, utc_offset_max = utc_offset_min) + +s2_data_cities(name = NULL) +} +\arguments{ +\item{name}{The name of a country, continent, city, or \code{NULL} +for all features.} + +\item{utc_offset_min, utc_offset_max}{Minimum and/or maximum timezone +offsets.} +} +\description{ +Well-known binary versions of the \href{https://www.naturalearthdata.com/}{Natural Earth} +low-resolution world boundaries and timezone boundaries. +} +\examples{ +head(s2_data_countries()) +s2_data_countries("Germany") +s2_data_countries("Europe") + +head(s2_data_timezones()) +s2_data_timezones(-4) + +head(s2_data_cities()) +s2_data_cities("Cairo") + +} +\keyword{datasets} diff --git a/man/s2_earth_radius_meters.Rd b/man/s2_earth_radius_meters.Rd new file mode 100644 index 0000000..249e411 --- /dev/null +++ b/man/s2_earth_radius_meters.Rd @@ -0,0 +1,25 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-earth.R +\name{s2_earth_radius_meters} +\alias{s2_earth_radius_meters} +\title{Earth Constants} +\usage{ +s2_earth_radius_meters() +} +\description{ +According to Yoder (1995), the radius of the earth is +6371.01 km. These functions are used to set the +default radis for functions that return a distance +or accept a distance as input +(e.g., \code{\link[=s2_distance]{s2_distance()}} and \code{\link[=s2_dwithin]{s2_dwithin()}}). +} +\examples{ +s2_earth_radius_meters() + +} +\references{ +Yoder, C.F. 1995. "Astrometric and Geodetic Properties of Earth and the +Solar System" in Global Earth Physics, A Handbook of Physical Constants, +AGU Reference Shelf 1, American Geophysical Union, Table 2. +\doi{10.1029/RF001p0001} +} diff --git a/man/s2_geog_point.Rd b/man/s2_geog_point.Rd new file mode 100644 index 0000000..d72ae62 --- /dev/null +++ b/man/s2_geog_point.Rd @@ -0,0 +1,164 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-constructors-formatters.R +\name{s2_geog_point} +\alias{s2_geog_point} +\alias{s2_make_line} +\alias{s2_make_polygon} +\alias{s2_geog_from_text} +\alias{s2_geog_from_wkb} +\alias{s2_as_text} +\alias{s2_as_binary} +\alias{s2_tessellate_tol_default} +\title{Create and Format Geography Vectors} +\usage{ +s2_geog_point(longitude, latitude) + +s2_make_line(longitude, latitude, feature_id = 1L) + +s2_make_polygon( + longitude, + latitude, + feature_id = 1L, + ring_id = 1L, + oriented = FALSE, + check = TRUE +) + +s2_geog_from_text( + wkt_string, + oriented = FALSE, + check = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default() +) + +s2_geog_from_wkb( + wkb_bytes, + oriented = FALSE, + check = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default() +) + +s2_as_text( + x, + precision = 16, + trim = TRUE, + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default() +) + +s2_as_binary( + x, + endian = wk::wk_platform_endian(), + planar = FALSE, + tessellate_tol_m = s2_tessellate_tol_default() +) + +s2_tessellate_tol_default() +} +\arguments{ +\item{longitude, latitude}{Vectors of latitude and longitude} + +\item{feature_id, ring_id}{Vectors for which a change in +sequential values indicates a new feature or ring. Use \code{\link[=factor]{factor()}} +to convert from a character vector.} + +\item{oriented}{TRUE if polygon ring directions are known to be correct +(i.e., exterior rings are defined counter clockwise and interior +rings are defined clockwise).} + +\item{check}{Use \code{check = FALSE} to skip error on invalid geometries} + +\item{wkt_string}{Well-known text} + +\item{planar}{Use \code{TRUE} to force planar edges in import or export.} + +\item{tessellate_tol_m}{The maximum number of meters to that a point must +be moved to satisfy the planar edge constraint.} + +\item{wkb_bytes}{A \code{list()} of \code{raw()}} + +\item{x}{An object that can be converted to an s2_geography vector} + +\item{precision}{The number of significant digits to export when +writing well-known text. If \code{trim = FALSE}, the number of +digits after the decimal place.} + +\item{trim}{Should trailing zeroes be included after the decimal place?} + +\item{endian}{The endian-ness of the well-known binary. See \code{\link[wk:deprecated]{wk::wkb_translate_wkb()}}.} +} +\description{ +These functions create and export \link[=as_s2_geography]{geography vectors}. +Unlike the BigQuery geography constructors, these functions do not sanitize +invalid or redundant input using \code{\link[=s2_union]{s2_union()}}. Note that when creating polygons +using \code{\link[=s2_make_polygon]{s2_make_polygon()}}, rings can be open or closed. +} +\examples{ +# create point geographies using coordinate values: +s2_geog_point(-64, 45) + +# create line geographies using coordinate values: +s2_make_line(c(-64, 8), c(45, 71)) + +# optionally, separate features using feature_id: +s2_make_line( + c(-64, 8, -27, -27), c(45, 71, 0, 45), + feature_id = c(1, 1, 2, 2) +) + +# create polygon geographies using coordinate values: +# (rings can be open or closed) +s2_make_polygon(c(-45, 8, 0), c(64, 71, 90)) + +# optionally, separate rings and/or features using +# ring_id and/or feature_id +s2_make_polygon( + c(20, 10, 10, 30, 45, 30, 20, 20, 40, 20, 45), + c(35, 30, 10, 5, 20, 20, 15, 25, 40, 45, 30), + feature_id = c(rep(1, 8), rep(2, 3)), + ring_id = c(1, 1, 1, 1, 1, 2, 2, 2, 1, 1, 1) +) + +# import and export well-known text +(geog <- s2_geog_from_text("POINT (-64 45)")) +s2_as_text(geog) + +# import and export well-known binary +(geog <- s2_geog_from_wkb(wk::as_wkb("POINT (-64 45)"))) +s2_as_binary(geog) + +# import geometry from planar space +s2_geog_from_text( + "POLYGON ((0 0, 1 0, 0 1, 0 0))", + planar = TRUE, + tessellate_tol_m = 1 +) + +# export geographies into planar space +geog <- s2_make_polygon(c(179, -179, 179), c(10, 10, 11)) +s2_as_text(geog, planar = TRUE) + +# polygons containing a pole need an extra step +geog <- s2_data_countries("Antarctica") +geom <- s2_as_text( + s2_intersection(geog, s2_world_plate_carree()), + planar = TRUE +) + +} +\seealso{ +See \code{\link[=as_s2_geography]{as_s2_geography()}} for other ways to construct geography vectors. + +BigQuery's geography function reference: +\itemize{ +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogpoint}{ST_GEOGPOINT} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_makeline}{ST_MAKELINE} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_makepolygon}{ST_MAKEPOLYGON} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogfromtext}{ST_GEOGFROMTEXT} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_geogfromwkb}{ST_GEOGFROMWKB} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_astext}{ST_ASTEXT} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_asbinary}{ST_ASBINARY} +} +} diff --git a/man/s2_interpolate.Rd b/man/s2_interpolate.Rd new file mode 100644 index 0000000..f59eb83 --- /dev/null +++ b/man/s2_interpolate.Rd @@ -0,0 +1,53 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-accessors.R, R/s2-transformers.R +\name{s2_project} +\alias{s2_project} +\alias{s2_project_normalized} +\alias{s2_interpolate} +\alias{s2_interpolate_normalized} +\title{Linear referencing} +\usage{ +s2_project(x, y, radius = s2_earth_radius_meters()) + +s2_project_normalized(x, y) + +s2_interpolate(x, distance, radius = s2_earth_radius_meters()) + +s2_interpolate_normalized(x, distance_normalized) +} +\arguments{ +\item{x}{A simple polyline geography vector} + +\item{y}{A simple point geography vector. The point will be +snapped to the nearest point on \code{x} for the purposes of +interpolation.} + +\item{radius}{Radius of the earth. Defaults to the average radius of +the earth in meters as defined by \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}}.} + +\item{distance}{A distance along \code{x} in \code{radius} units.} + +\item{distance_normalized}{A \code{distance} normalized to \code{\link[=s2_length]{s2_length()}} of +\code{x}.} +} +\value{ +\itemize{ +\item \code{s2_interpolate()} returns the point on \code{x}, \code{distance} meters +along the line. +\item \code{s2_interpolate_normalized()} returns the point on \code{x} interpolated +to a fraction along the line. +\item \code{s2_project()} returns the \code{distance} that \code{point} occurs along \code{x}. +\item \code{s2_project_normalized()} returns the \code{distance_normalized} along \code{x} +where \code{point} occurs. +} +} +\description{ +Linear referencing +} +\examples{ +s2_project_normalized("LINESTRING (0 0, 0 90)", "POINT (0 22.5)") +s2_project("LINESTRING (0 0, 0 90)", "POINT (0 22.5)") +s2_interpolate_normalized("LINESTRING (0 0, 0 90)", 0.25) +s2_interpolate("LINESTRING (0 0, 0 90)", 2501890) + +} diff --git a/man/s2_is_collection.Rd b/man/s2_is_collection.Rd new file mode 100644 index 0000000..bec93c0 --- /dev/null +++ b/man/s2_is_collection.Rd @@ -0,0 +1,112 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-accessors.R +\name{s2_is_collection} +\alias{s2_is_collection} +\alias{s2_is_valid} +\alias{s2_is_valid_detail} +\alias{s2_dimension} +\alias{s2_num_points} +\alias{s2_is_empty} +\alias{s2_area} +\alias{s2_length} +\alias{s2_perimeter} +\alias{s2_x} +\alias{s2_y} +\alias{s2_distance} +\alias{s2_max_distance} +\title{S2 Geography Accessors} +\usage{ +s2_is_collection(x) + +s2_is_valid(x) + +s2_is_valid_detail(x) + +s2_dimension(x) + +s2_num_points(x) + +s2_is_empty(x) + +s2_area(x, radius = s2_earth_radius_meters()) + +s2_length(x, radius = s2_earth_radius_meters()) + +s2_perimeter(x, radius = s2_earth_radius_meters()) + +s2_x(x) + +s2_y(x) + +s2_distance(x, y, radius = s2_earth_radius_meters()) + +s2_max_distance(x, y, radius = s2_earth_radius_meters()) +} +\arguments{ +\item{x, y}{\link[=as_s2_geography]{geography vectors}. These inputs +are passed to \code{\link[=as_s2_geography]{as_s2_geography()}}, so you can pass other objects +(e.g., character vectors of well-known text) directly.} + +\item{radius}{Radius of the earth. Defaults to the average radius of +the earth in meters as defined by \code{\link[=s2_earth_radius_meters]{s2_earth_radius_meters()}}.} +} +\description{ +Accessors extract information about \link[=as_s2_geography]{geography vectors}. +} +\examples{ +# s2_is_collection() tests for multiple geometries in one feature +s2_is_collection(c("POINT (-64 45)", "MULTIPOINT ((-64 45), (8 72))")) + +# s2_dimension() returns 0 for point, 1 for line, 2 for polygon +s2_dimension( + c( + "GEOMETRYCOLLECTION EMPTY", + "POINT (-64 45)", + "LINESTRING (-64 45, 8 72)", + "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", + "GEOMETRYCOLLECTION (POINT (-64 45), LINESTRING (-64 45, 8 72))" + ) +) + +# s2_num_points() counts points +s2_num_points(c("POINT (-64 45)", "LINESTRING (-64 45, 8 72)")) + +# s2_is_empty tests for emptiness +s2_is_empty(c("POINT (-64 45)", "POINT EMPTY")) + +# calculate area, length, and perimeter +s2_area("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))") +s2_perimeter("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))") +s2_length(s2_boundary("POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))")) + +# extract x and y coordinates from points +s2_x(c("POINT (-64 45)", "POINT EMPTY")) +s2_y(c("POINT (-64 45)", "POINT EMPTY")) + +# calculate minimum and maximum distance between two geometries +s2_distance( + "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", + "POINT (-64 45)" +) +s2_max_distance( + "POLYGON ((0 0, 0 10, 10 10, 10 0, 0 0))", + "POINT (-64 45)" +) + +} +\seealso{ +BigQuery's geography function reference: +\itemize{ +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_iscollection}{ST_ISCOLLECTION} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_dimension}{ST_DIMENSION} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_numpoints}{ST_NUMPOINTS} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_isempty}{ST_ISEMPTY} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_area}{ST_AREA} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_length}{ST_LENGTH} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_perimeter}{ST_PERIMETER} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_x}{ST_X} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_y}{ST_Y} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_distance}{ST_DISTANCE} +\item \href{https://cloud.google.com/bigquery/docs/reference/standard-sql/geography_functions#st_maxdistance}{ST_MAXDISTANCE} +} +} diff --git a/man/s2_lnglat.Rd b/man/s2_lnglat.Rd new file mode 100644 index 0000000..53bc070 --- /dev/null +++ b/man/s2_lnglat.Rd @@ -0,0 +1,40 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-lnglat.R +\name{s2_lnglat} +\alias{s2_lnglat} +\alias{as_s2_lnglat} +\alias{as_s2_lnglat.default} +\alias{as_s2_lnglat.wk_xy} +\alias{as_s2_lnglat.wk_xyz} +\title{Create an S2 LngLat Vector} +\usage{ +s2_lnglat(lng, lat) + +as_s2_lnglat(x, ...) + +\method{as_s2_lnglat}{default}(x, ...) + +\method{as_s2_lnglat}{wk_xy}(x, ...) + +\method{as_s2_lnglat}{wk_xyz}(x, ...) +} +\arguments{ +\item{lat, lng}{Vectors of latitude and longitude values in degrees.} + +\item{x}{A \code{\link[=s2_lnglat]{s2_lnglat()}} vector or an object that can be coerced to one.} + +\item{...}{Unused} +} +\value{ +An object with class s2_lnglat +} +\description{ +This class represents a latitude and longitude on the Earth's surface. +Most calculations in S2 convert this to a \code{\link[=as_s2_point]{as_s2_point()}}, which is a +unit vector representation of this value. +} +\examples{ +s2_lnglat(45, -64) # Halifax, Nova Scotia! +as.data.frame(s2_lnglat(45, -64)) + +} diff --git a/man/s2_options.Rd b/man/s2_options.Rd new file mode 100644 index 0000000..d1b4865 --- /dev/null +++ b/man/s2_options.Rd @@ -0,0 +1,109 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-options.R +\name{s2_options} +\alias{s2_options} +\alias{s2_snap_identity} +\alias{s2_snap_level} +\alias{s2_snap_precision} +\alias{s2_snap_distance} +\title{Geography Operation Options} +\usage{ +s2_options( + model = NULL, + snap = s2_snap_identity(), + snap_radius = -1, + duplicate_edges = FALSE, + edge_type = "directed", + validate = FALSE, + polyline_type = "path", + polyline_sibling_pairs = "keep", + simplify_edge_chains = FALSE, + split_crossing_edges = FALSE, + idempotent = FALSE, + dimensions = c("point", "polyline", "polygon") +) + +s2_snap_identity() + +s2_snap_level(level) + +s2_snap_precision(precision) + +s2_snap_distance(distance) +} +\arguments{ +\item{model}{One of 'open', 'semi-open' (default for polygons), +or 'closed' (default for polylines). See section 'Model'} + +\item{snap}{Use \code{s2_snap_identity()}, \code{s2_snap_distance()}, \code{s2_snap_level()}, +or \code{s2_snap_precision()} to specify how or if coordinate rounding should +occur.} + +\item{snap_radius}{As opposed to the snap function, which specifies +the maximum distance a vertex should move, the snap radius (in radians) sets +the minimum distance between vertices of the output that don't cause vertices +to move more than the distance specified by the snap function. This can be used +to simplify the result of a boolean operation. Use -1 to specify that any +minimum distance is acceptable.} + +\item{duplicate_edges}{Use \code{TRUE} to keep duplicate edges (e.g., duplicate +points).} + +\item{edge_type}{One of 'directed' (default) or 'undirected'.} + +\item{validate}{Use \code{TRUE} to validate the result from the builder.} + +\item{polyline_type}{One of 'path' (default) or 'walk'. If 'walk', +polylines that backtrack are preserved.} + +\item{polyline_sibling_pairs}{One of 'discard' (default) or 'keep'.} + +\item{simplify_edge_chains}{Use \code{TRUE} to remove vertices that are within +\code{snap_radius} of the original vertex.} + +\item{split_crossing_edges}{Use \code{TRUE} to split crossing polyline edges +when creating geometries.} + +\item{idempotent}{Use \code{FALSE} to apply snap even if snapping is not necessary +to satisfy vertex constraints.} + +\item{dimensions}{A combination of 'point', 'polyline', and/or 'polygon' +that can used to constrain the output of \code{\link[=s2_rebuild]{s2_rebuild()}} or a +boolean operation.} + +\item{level}{A value from 0 to 30 corresponding to the cell level +at which snapping should occur.} + +\item{precision}{A number by which coordinates should be multiplied +before being rounded. Rounded to the nearest exponent of 10.} + +\item{distance}{A distance (in radians) denoting the maximum +distance a vertex should move in the snapping process.} +} +\description{ +These functions specify defaults for options used to perform operations +and construct geometries. These are used in predicates (e.g., \code{\link[=s2_intersects]{s2_intersects()}}), +and boolean operations (e.g., \code{\link[=s2_intersection]{s2_intersection()}}) to specify the model for +containment and how new geometries should be constructed. +} +\section{Model}{ + +The geometry model indicates whether or not a geometry includes its boundaries. +Boundaries of line geometries are its end points. +OPEN geometries do not contain their boundary (\code{model = "open"}); CLOSED +geometries (\code{model = "closed"}) contain their boundary; SEMI-OPEN geometries +(\code{model = "semi-open"}) contain half of their boundaries, such that when two polygons +do not overlap or two lines do not cross, no point exist that belong to +more than one of the geometries. (This latter form, half-closed, is +not present in the OpenGIS "simple feature access" (SFA) standard nor DE9-IM on +which that is based). The default values for \code{\link[=s2_contains]{s2_contains()}} (open) +and covers/covered_by (closed) correspond to the SFA standard specification +of these operators. +} + +\examples{ +# use s2_options() to specify containment models, snap level +# layer creation options, and builder options +s2_options(model = "closed", snap = s2_snap_level(30)) + +} diff --git a/man/s2_plot.Rd b/man/s2_plot.Rd new file mode 100644 index 0000000..8de0c1b --- /dev/null +++ b/man/s2_plot.Rd @@ -0,0 +1,52 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/plot.R +\name{s2_plot} +\alias{s2_plot} +\title{Plot S2 Geographies} +\usage{ +s2_plot( + x, + ..., + asp = 1, + xlab = "", + ylab = "", + rule = "evenodd", + add = FALSE, + plot_hemisphere = FALSE, + simplify = TRUE, + centre = NULL +) +} +\arguments{ +\item{x}{A \code{\link[wk:wkb]{wkb()}} or \code{\link[wk:wkt]{wkt()}}} + +\item{...}{Passed to plotting functions for features: \code{\link[graphics:points]{graphics::points()}} +for point and multipoint geometries, \code{\link[graphics:lines]{graphics::lines()}} for linestring +and multilinestring geometries, and \code{\link[graphics:polypath]{graphics::polypath()}} for polygon +and multipolygon geometries.} + +\item{asp, xlab, ylab}{Passed to \code{\link[graphics:plot.default]{graphics::plot()}}} + +\item{rule}{The rule to use for filling polygons (see \code{\link[graphics:polypath]{graphics::polypath()}})} + +\item{add}{Should a new plot be created, or should \code{handleable} be added to the +existing plot?} + +\item{plot_hemisphere}{Plot the outline of the earth} + +\item{simplify}{Use \code{FALSE} to skip the simplification step} + +\item{centre}{The longitude/latitude point of the centre of the +orthographic projection} +} +\value{ +The input, invisibly +} +\description{ +Plot S2 Geographies +} +\examples{ +s2_plot(s2_data_countries()) +s2_plot(s2_data_cities(), add = TRUE) + +} diff --git a/man/s2_point.Rd b/man/s2_point.Rd new file mode 100644 index 0000000..2046478 --- /dev/null +++ b/man/s2_point.Rd @@ -0,0 +1,42 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/s2-point.R +\name{s2_point} +\alias{s2_point} +\alias{s2_point_crs} +\alias{as_s2_point} +\alias{as_s2_point.default} +\alias{as_s2_point.wk_xy} +\alias{as_s2_point.wk_xyz} +\title{Create an S2 Point Vector} +\usage{ +s2_point(x, y, z) + +s2_point_crs() + +as_s2_point(x, ...) + +\method{as_s2_point}{default}(x, ...) + +\method{as_s2_point}{wk_xy}(x, ...) + +\method{as_s2_point}{wk_xyz}(x, ...) +} +\arguments{ +\item{x, y, z}{Vectors of latitude and longitude values in degrees.} + +\item{...}{Unused} +} +\value{ +An object with class s2_point +} +\description{ +In S2 terminology, a "point" is a 3-dimensional unit vector representation +of an \code{\link[=s2_point]{s2_point()}}. Internally, all s2 objects are stored as +3-dimensional unit vectors. +} +\examples{ +point <- s2_lnglat(-64, 45) # Halifax, Nova Scotia! +as_s2_point(point) +as.data.frame(as_s2_point(point)) + +} diff --git a/man/wk_handle.s2_geography.Rd b/man/wk_handle.s2_geography.Rd new file mode 100644 index 0000000..88ff6ec --- /dev/null +++ b/man/wk_handle.s2_geography.Rd @@ -0,0 +1,85 @@ +% Generated by roxygen2: do not edit by hand +% Please edit documentation in R/wk-utils.R +\name{wk_handle.s2_geography} +\alias{wk_handle.s2_geography} +\alias{s2_geography_writer} +\alias{wk_writer.s2_geography} +\alias{s2_trans_point} +\alias{s2_trans_lnglat} +\alias{s2_projection_plate_carree} +\alias{s2_projection_mercator} +\alias{s2_hemisphere} +\alias{s2_world_plate_carree} +\alias{s2_projection_orthographic} +\title{Low-level wk filters and handlers} +\usage{ +\method{wk_handle}{s2_geography}( + handleable, + handler, + ..., + s2_projection = s2_projection_plate_carree(), + s2_tessellate_tol = Inf +) + +s2_geography_writer( + oriented = FALSE, + check = TRUE, + projection = s2_projection_plate_carree(), + tessellate_tol = Inf +) + +\method{wk_writer}{s2_geography}(handleable, ...) + +s2_trans_point() + +s2_trans_lnglat() + +s2_projection_plate_carree(x_scale = 180) + +s2_projection_mercator(x_scale = 20037508.3427892) + +s2_hemisphere(centre) + +s2_world_plate_carree(epsilon_east_west = 0, epsilon_north_south = 0) + +s2_projection_orthographic(centre = s2_lnglat(0, 0)) +} +\arguments{ +\item{handleable}{A geometry vector (e.g., \code{\link[wk:wkb]{wkb()}}, \code{\link[wk:wkt]{wkt()}}, \code{\link[wk:xy]{xy()}}, +\code{\link[wk:rct]{rct()}}, or \code{\link[sf:sfc]{sf::st_sfc()}}) for which \code{\link[wk:wk_handle]{wk_handle()}} is defined.} + +\item{handler}{A \link[wk:wk_handle]{wk_handler} object.} + +\item{...}{Passed to the \code{\link[wk:wk_handle]{wk_handle()}} method.} + +\item{oriented}{TRUE if polygon ring directions are known to be correct +(i.e., exterior rings are defined counter clockwise and interior +rings are defined clockwise).} + +\item{check}{Use \code{check = FALSE} to skip error on invalid geometries} + +\item{projection, s2_projection}{One of \code{\link[=s2_projection_plate_carree]{s2_projection_plate_carree()}} or +\code{\link[=s2_projection_mercator]{s2_projection_mercator()}}} + +\item{tessellate_tol, s2_tessellate_tol}{An angle in radians. +Points will not be added if a line segment is within this +distance of a point.} + +\item{x_scale}{The maximum x value of the projection} + +\item{centre}{The center point of the orthographic projection} + +\item{epsilon_east_west, epsilon_north_south}{Use a positive number to +define the edges of a Cartesian world slightly inward from -180, -90, +180, 90. This may be used to define a world outline for a projection where +projecting at the extreme edges of the earth results in a non-finite value.} +} +\value{ +\itemize{ +\item \code{s2_projection_plate_carree()}, \code{s2_projection_mercator()}: An external pointer +to an S2 projection. +} +} +\description{ +Low-level wk filters and handlers +} diff --git a/src/Makevars.in b/src/Makevars.in new file mode 100644 index 0000000..db45424 --- /dev/null +++ b/src/Makevars.in @@ -0,0 +1,218 @@ +PKG_CPPFLAGS = -I../src -DSTRICT_R_HEADERS +PKG_LIBS = @libs@ +PKG_CXXFLAGS = @cflags@ -pthread +CXX_STD = CXX11 + +ABSL_LIBS = absl/base/internal/cycleclock.o \ + absl/base/internal/low_level_alloc.o \ + absl/base/internal/raw_logging.o \ + absl/base/internal/scoped_set_env.o \ + absl/base/internal/spinlock_wait.o \ + absl/base/internal/spinlock.o \ + absl/base/internal/strerror.o \ + absl/base/internal/sysinfo.o \ + absl/base/internal/thread_identity.o \ + absl/base/internal/throw_delegate.o \ + absl/base/internal/unscaledcycleclock.o \ + absl/base/log_severity.o \ + absl/container/internal/hashtablez_sampler_force_weak_definition.o \ + absl/container/internal/hashtablez_sampler.o \ + absl/container/internal/raw_hash_set.o \ + absl/debugging/failure_signal_handler.o \ + absl/debugging/internal/address_is_readable.o \ + absl/debugging/internal/demangle.o \ + absl/debugging/internal/elf_mem_image.o \ + absl/debugging/internal/examine_stack.o \ + absl/debugging/internal/stack_consumption.o \ + absl/debugging/internal/vdso_support.o \ + absl/debugging/leak_check.o \ + absl/debugging/stacktrace.o \ + absl/debugging/symbolize.o \ + absl/numeric/int128.o \ + absl/profiling/internal/exponential_biased.o \ + absl/profiling/internal/periodic_sampler.o \ + absl/strings/ascii.o \ + absl/strings/charconv.o \ + absl/strings/cord_analysis.o \ + absl/strings/cord_buffer.o \ + absl/strings/cord.o \ + absl/strings/escaping.o \ + absl/strings/internal/charconv_bigint.o \ + absl/strings/internal/charconv_parse.o \ + absl/strings/internal/cord_internal.o \ + absl/strings/internal/cord_rep_btree_navigator.o \ + absl/strings/internal/cord_rep_btree_reader.o \ + absl/strings/internal/cord_rep_btree.o \ + absl/strings/internal/cord_rep_consume.o \ + absl/strings/internal/cord_rep_crc.o \ + absl/strings/internal/cord_rep_ring.o \ + absl/strings/internal/cordz_functions.o \ + absl/strings/internal/cordz_handle.o \ + absl/strings/internal/cordz_info.o \ + absl/strings/internal/cordz_sample_token.o \ + absl/strings/internal/escaping.o \ + absl/strings/internal/memutil.o \ + absl/strings/internal/ostringstream.o \ + absl/strings/internal/pow10_helper.o \ + absl/strings/internal/str_format/arg.o \ + absl/strings/internal/str_format/bind.o \ + absl/strings/internal/str_format/extension.o \ + absl/strings/internal/str_format/float_conversion.o \ + absl/strings/internal/str_format/output.o \ + absl/strings/internal/str_format/parser.o \ + absl/strings/internal/utf8.o \ + absl/strings/match.o \ + absl/strings/numbers.o \ + absl/strings/str_cat.o \ + absl/strings/str_replace.o \ + absl/strings/str_split.o \ + absl/strings/string_view.o \ + absl/strings/substitute.o \ + absl/synchronization/barrier.o \ + absl/synchronization/blocking_counter.o \ + absl/synchronization/internal/create_thread_identity.o \ + absl/synchronization/internal/graphcycles.o \ + absl/synchronization/internal/per_thread_sem.o \ + absl/synchronization/internal/waiter.o \ + absl/synchronization/mutex.o \ + absl/synchronization/notification.o \ + absl/time/civil_time.o \ + absl/time/clock.o \ + absl/time/duration.o \ + absl/time/format.o \ + absl/time/internal/cctz/src/civil_time_detail.o \ + absl/time/internal/cctz/src/time_zone_fixed.o \ + absl/time/internal/cctz/src/time_zone_format.o \ + absl/time/internal/cctz/src/time_zone_if.o \ + absl/time/internal/cctz/src/time_zone_impl.o \ + absl/time/internal/cctz/src/time_zone_info.o \ + absl/time/internal/cctz/src/time_zone_libc.o \ + absl/time/internal/cctz/src/time_zone_lookup.o \ + absl/time/internal/cctz/src/time_zone_posix.o \ + absl/time/internal/cctz/src/zone_info_source.o \ + absl/time/time.o \ + absl/types/bad_any_cast.o \ + absl/types/bad_optional_access.o \ + absl/types/bad_variant_access.o + +OBJECTS = $(ABSL_LIBS) \ + cpp-compat.o \ + s2-accessors.o \ + s2-bounds.o \ + s2-cell.o \ + s2-cell-union.o \ + s2-constructors-formatters.o \ + s2-predicates.o \ + s2-transformers.o \ + init.o \ + RcppExports.o \ + s2-geography.o \ + s2-lnglat.o \ + s2-matrix.o \ + wk-impl.o \ + s2geography/accessors.o \ + s2geography/accessors-geog.o \ + s2geography/linear-referencing.o \ + s2geography/distance.o \ + s2geography/build.o \ + s2geography/coverings.o \ + s2geography/geography.o \ + s2geography/predicates.o \ + s2/base/stringprintf.o \ + s2/base/strtoint.o \ + s2/encoded_s2cell_id_vector.o \ + s2/encoded_s2point_vector.o \ + s2/encoded_s2shape_index.o \ + s2/encoded_string_vector.o \ + s2/id_set_lexicon.o \ + s2/mutable_s2shape_index.o \ + s2/r2rect.o \ + s2/s1angle.o \ + s2/s1chord_angle.o \ + s2/s1interval.o \ + s2/s2boolean_operation.o \ + s2/s2builder_graph.o \ + s2/s2builder.o \ + s2/s2builderutil_closed_set_normalizer.o \ + s2/s2builderutil_find_polygon_degeneracies.o \ + s2/s2builderutil_lax_polygon_layer.o \ + s2/s2builderutil_s2point_vector_layer.o \ + s2/s2builderutil_s2polygon_layer.o \ + s2/s2builderutil_s2polyline_layer.o \ + s2/s2builderutil_s2polyline_vector_layer.o \ + s2/s2builderutil_snap_functions.o \ + s2/s2builderutil_testing.o \ + s2/s2cap.o \ + s2/s2cell_id.o \ + s2/s2cell_index.o \ + s2/s2cell_union.o \ + s2/s2cell.o \ + s2/s2centroids.o \ + s2/s2closest_cell_query.o \ + s2/s2closest_edge_query.o \ + s2/s2closest_point_query.o \ + s2/s2contains_vertex_query.o \ + s2/s2convex_hull_query.o \ + s2/s2coords.o \ + s2/s2crossing_edge_query.o \ + s2/s2debug.o \ + s2/s2earth.o \ + s2/s2edge_clipping.o \ + s2/s2edge_crosser.o \ + s2/s2edge_crossings.o \ + s2/s2edge_distances.o \ + s2/s2edge_tessellator.o \ + s2/s2error.o \ + s2/s2furthest_edge_query.o \ + s2/s2latlng_rect_bounder.o \ + s2/s2latlng_rect.o \ + s2/s2latlng.o \ + s2/s2lax_loop_shape.o \ + s2/s2lax_polygon_shape.o \ + s2/s2lax_polyline_shape.o \ + s2/s2loop_measures.o \ + s2/s2loop.o \ + s2/s2max_distance_targets.o \ + s2/s2measures.o \ + s2/s2metrics.o \ + s2/s2min_distance_targets.o \ + s2/s2padded_cell.o \ + s2/s2point_compression.o \ + s2/s2point_region.o \ + s2/s2pointutil.o \ + s2/s2polygon.o \ + s2/s2polyline_alignment.o \ + s2/s2polyline_measures.o \ + s2/s2polyline_simplifier.o \ + s2/s2polyline.o \ + s2/s2predicates.o \ + s2/s2projections.o \ + s2/s2r2rect.o \ + s2/s2region_coverer.o \ + s2/s2region_intersection.o \ + s2/s2region_term_indexer.o \ + s2/s2region_union.o \ + s2/s2region.o \ + s2/s2shape_index_buffered_region.o \ + s2/s2shape_index_measures.o \ + s2/s2shape_index.o \ + s2/s2shape_measures.o \ + s2/s2shapeutil_build_polygon_boundaries.o \ + s2/s2shapeutil_coding.o \ + s2/s2shapeutil_contains_brute_force.o \ + s2/s2shapeutil_edge_iterator.o \ + s2/s2shapeutil_get_reference_point.o \ + s2/s2shapeutil_range_iterator.o \ + s2/s2shapeutil_visit_crossing_edge_pairs.o \ + s2/s2testing.o \ + s2/s2text_format.o \ + s2/s2wedge_relations.o \ + s2/strings/ostringstream.o \ + s2/strings/serialize.o \ + s2/util/bits/bit-interleave.o \ + s2/util/bits/bits.o \ + s2/util/coding/coder.o \ + s2/util/coding/varint.o \ + s2/util/math/exactfloat/exactfloat.o \ + s2/util/math/mathutil.o \ + s2/util/units/length-units.o diff --git a/src/Makevars.ucrt b/src/Makevars.ucrt new file mode 100644 index 0000000..1a2e101 --- /dev/null +++ b/src/Makevars.ucrt @@ -0,0 +1,2 @@ +CRT=-ucrt +include Makevars.win diff --git a/src/Makevars.win b/src/Makevars.win new file mode 100644 index 0000000..f4c8cfd --- /dev/null +++ b/src/Makevars.win @@ -0,0 +1,224 @@ +VERSION = 1.1.1k +PKG_CPPFLAGS = -DS2_USE_EXACTFLOAT -D_USE_MATH_DEFINES -DNDEBUG -DIS_LITTLE_ENDIAN -DOMIT_STRPTIME -I../windows/openssl-$(VERSION)/include -I../src +PKG_LIBS = -Ls2 -ls2static -L../windows/openssl-$(VERSION)/lib${R_ARCH}${CRT} -lssl -lcrypto -lws2_32 -lgdi32 -lcrypt32 + +CXX_STD = CXX11 + +STATLIB = s2/libs2static.a + +ABSL_LIBS = absl/base/internal/cycleclock.o \ + absl/base/internal/low_level_alloc.o \ + absl/base/internal/raw_logging.o \ + absl/base/internal/scoped_set_env.o \ + absl/base/internal/spinlock_wait.o \ + absl/base/internal/spinlock.o \ + absl/base/internal/strerror.o \ + absl/base/internal/sysinfo.o \ + absl/base/internal/thread_identity.o \ + absl/base/internal/throw_delegate.o \ + absl/base/internal/unscaledcycleclock.o \ + absl/base/log_severity.o \ + absl/container/internal/hashtablez_sampler_force_weak_definition.o \ + absl/container/internal/hashtablez_sampler.o \ + absl/container/internal/raw_hash_set.o \ + absl/debugging/failure_signal_handler.o \ + absl/debugging/internal/address_is_readable.o \ + absl/debugging/internal/demangle.o \ + absl/debugging/internal/elf_mem_image.o \ + absl/debugging/internal/examine_stack.o \ + absl/debugging/internal/stack_consumption.o \ + absl/debugging/internal/vdso_support.o \ + absl/debugging/leak_check.o \ + absl/debugging/stacktrace.o \ + absl/debugging/symbolize.o \ + absl/numeric/int128.o \ + absl/profiling/internal/exponential_biased.o \ + absl/profiling/internal/periodic_sampler.o \ + absl/strings/ascii.o \ + absl/strings/charconv.o \ + absl/strings/cord_analysis.o \ + absl/strings/cord_buffer.o \ + absl/strings/cord.o \ + absl/strings/escaping.o \ + absl/strings/internal/charconv_bigint.o \ + absl/strings/internal/charconv_parse.o \ + absl/strings/internal/cord_internal.o \ + absl/strings/internal/cord_rep_btree_navigator.o \ + absl/strings/internal/cord_rep_btree_reader.o \ + absl/strings/internal/cord_rep_btree.o \ + absl/strings/internal/cord_rep_consume.o \ + absl/strings/internal/cord_rep_crc.o \ + absl/strings/internal/cord_rep_ring.o \ + absl/strings/internal/cordz_functions.o \ + absl/strings/internal/cordz_handle.o \ + absl/strings/internal/cordz_info.o \ + absl/strings/internal/cordz_sample_token.o \ + absl/strings/internal/escaping.o \ + absl/strings/internal/memutil.o \ + absl/strings/internal/ostringstream.o \ + absl/strings/internal/pow10_helper.o \ + absl/strings/internal/str_format/arg.o \ + absl/strings/internal/str_format/bind.o \ + absl/strings/internal/str_format/extension.o \ + absl/strings/internal/str_format/float_conversion.o \ + absl/strings/internal/str_format/output.o \ + absl/strings/internal/str_format/parser.o \ + absl/strings/internal/utf8.o \ + absl/strings/match.o \ + absl/strings/numbers.o \ + absl/strings/str_cat.o \ + absl/strings/str_replace.o \ + absl/strings/str_split.o \ + absl/strings/string_view.o \ + absl/strings/substitute.o \ + absl/synchronization/barrier.o \ + absl/synchronization/blocking_counter.o \ + absl/synchronization/internal/create_thread_identity.o \ + absl/synchronization/internal/graphcycles.o \ + absl/synchronization/internal/per_thread_sem.o \ + absl/synchronization/internal/waiter.o \ + absl/synchronization/mutex.o \ + absl/synchronization/notification.o \ + absl/time/civil_time.o \ + absl/time/clock.o \ + absl/time/duration.o \ + absl/time/format.o \ + absl/time/internal/cctz/src/civil_time_detail.o \ + absl/time/internal/cctz/src/time_zone_fixed.o \ + absl/time/internal/cctz/src/time_zone_format.o \ + absl/time/internal/cctz/src/time_zone_if.o \ + absl/time/internal/cctz/src/time_zone_impl.o \ + absl/time/internal/cctz/src/time_zone_info.o \ + absl/time/internal/cctz/src/time_zone_libc.o \ + absl/time/internal/cctz/src/time_zone_lookup.o \ + absl/time/internal/cctz/src/time_zone_posix.o \ + absl/time/internal/cctz/src/zone_info_source.o \ + absl/time/time.o \ + absl/types/bad_any_cast.o \ + absl/types/bad_optional_access.o \ + absl/types/bad_variant_access.o + +S2LIBS = $(ABSL_LIBS) \ + s2geography/linear-referencing.o \ + s2geography/distance.o \ + s2geography/accessors.o \ + s2geography/accessors-geog.o \ + s2geography/build.o \ + s2geography/coverings.o \ + s2geography/geography.o \ + s2geography/predicates.o \ + s2/base/stringprintf.o \ + s2/base/strtoint.o \ + s2/encoded_s2cell_id_vector.o \ + s2/encoded_s2point_vector.o \ + s2/encoded_s2shape_index.o \ + s2/encoded_string_vector.o \ + s2/id_set_lexicon.o \ + s2/mutable_s2shape_index.o \ + s2/r2rect.o \ + s2/s1angle.o \ + s2/s1chord_angle.o \ + s2/s1interval.o \ + s2/s2boolean_operation.o \ + s2/s2builder_graph.o \ + s2/s2builder.o \ + s2/s2builderutil_closed_set_normalizer.o \ + s2/s2builderutil_find_polygon_degeneracies.o \ + s2/s2builderutil_lax_polygon_layer.o \ + s2/s2builderutil_s2point_vector_layer.o \ + s2/s2builderutil_s2polygon_layer.o \ + s2/s2builderutil_s2polyline_layer.o \ + s2/s2builderutil_s2polyline_vector_layer.o \ + s2/s2builderutil_snap_functions.o \ + s2/s2builderutil_testing.o \ + s2/s2cap.o \ + s2/s2cell_id.o \ + s2/s2cell_index.o \ + s2/s2cell_union.o \ + s2/s2cell.o \ + s2/s2centroids.o \ + s2/s2closest_cell_query.o \ + s2/s2closest_edge_query.o \ + s2/s2closest_point_query.o \ + s2/s2contains_vertex_query.o \ + s2/s2convex_hull_query.o \ + s2/s2coords.o \ + s2/s2crossing_edge_query.o \ + s2/s2debug.o \ + s2/s2earth.o \ + s2/s2edge_clipping.o \ + s2/s2edge_crosser.o \ + s2/s2edge_crossings.o \ + s2/s2edge_distances.o \ + s2/s2edge_tessellator.o \ + s2/s2error.o \ + s2/s2furthest_edge_query.o \ + s2/s2latlng_rect_bounder.o \ + s2/s2latlng_rect.o \ + s2/s2latlng.o \ + s2/s2lax_loop_shape.o \ + s2/s2lax_polygon_shape.o \ + s2/s2lax_polyline_shape.o \ + s2/s2loop_measures.o \ + s2/s2loop.o \ + s2/s2max_distance_targets.o \ + s2/s2measures.o \ + s2/s2metrics.o \ + s2/s2min_distance_targets.o \ + s2/s2padded_cell.o \ + s2/s2point_compression.o \ + s2/s2point_region.o \ + s2/s2pointutil.o \ + s2/s2polygon.o \ + s2/s2polyline_alignment.o \ + s2/s2polyline_measures.o \ + s2/s2polyline_simplifier.o \ + s2/s2polyline.o \ + s2/s2predicates.o \ + s2/s2projections.o \ + s2/s2r2rect.o \ + s2/s2region_coverer.o \ + s2/s2region_intersection.o \ + s2/s2region_term_indexer.o \ + s2/s2region_union.o \ + s2/s2region.o \ + s2/s2shape_index_buffered_region.o \ + s2/s2shape_index_measures.o \ + s2/s2shape_index.o \ + s2/s2shape_measures.o \ + s2/s2shapeutil_build_polygon_boundaries.o \ + s2/s2shapeutil_coding.o \ + s2/s2shapeutil_contains_brute_force.o \ + s2/s2shapeutil_edge_iterator.o \ + s2/s2shapeutil_get_reference_point.o \ + s2/s2shapeutil_range_iterator.o \ + s2/s2shapeutil_visit_crossing_edge_pairs.o \ + s2/s2testing.o \ + s2/s2text_format.o \ + s2/s2wedge_relations.o \ + s2/strings/ostringstream.o \ + s2/strings/serialize.o \ + s2/util/bits/bit-interleave.o \ + s2/util/bits/bits.o \ + s2/util/coding/coder.o \ + s2/util/coding/varint.o \ + s2/util/math/exactfloat/exactfloat.o \ + s2/util/math/mathutil.o \ + s2/util/units/length-units.o + +$(SHLIB): $(STATLIB) + +$(STATLIB): $(S2LIBS) + +$(S2LIBS): winlibs + +#all: clean + +winlibs: + mkdir -p ../inst + "${R_HOME}/bin${R_ARCH_BIN}/Rscript.exe" "../tools/winlibs.R" $(VERSION) + +clean: + rm -f $(SHLIB) $(STATLIB) $(OBJECTS) $(S2LIBS) + +.PHONY: all winlibs clean diff --git a/src/RcppExports.cpp b/src/RcppExports.cpp new file mode 100644 index 0000000..927a6bf --- /dev/null +++ b/src/RcppExports.cpp @@ -0,0 +1,1482 @@ +// Generated by using Rcpp::compileAttributes() -> do not edit by hand +// Generator token: 10BE3573-1514-4C36-9D1C-5A225CD40393 + +#include + +using namespace Rcpp; + +#ifdef RCPP_USE_GLOBAL_ROSTREAM +Rcpp::Rostream& Rcpp::Rcout = Rcpp::Rcpp_cout_get(); +Rcpp::Rostream& Rcpp::Rcerr = Rcpp::Rcpp_cerr_get(); +#endif + +// cpp_s2_init +void cpp_s2_init(); +RcppExport SEXP _s2_cpp_s2_init() { +BEGIN_RCPP + Rcpp::RNGScope rcpp_rngScope_gen; + cpp_s2_init(); + return R_NilValue; +END_RCPP +} +// cpp_s2_is_collection +LogicalVector cpp_s2_is_collection(List geog); +RcppExport SEXP _s2_cpp_s2_is_collection(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_is_collection(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_is_valid +LogicalVector cpp_s2_is_valid(List geog); +RcppExport SEXP _s2_cpp_s2_is_valid(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_is_valid(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_is_valid_reason +CharacterVector cpp_s2_is_valid_reason(List geog); +RcppExport SEXP _s2_cpp_s2_is_valid_reason(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_is_valid_reason(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_dimension +IntegerVector cpp_s2_dimension(List geog); +RcppExport SEXP _s2_cpp_s2_dimension(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_dimension(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_num_points +IntegerVector cpp_s2_num_points(List geog); +RcppExport SEXP _s2_cpp_s2_num_points(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_num_points(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_is_empty +LogicalVector cpp_s2_is_empty(List geog); +RcppExport SEXP _s2_cpp_s2_is_empty(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_is_empty(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_area +NumericVector cpp_s2_area(List geog); +RcppExport SEXP _s2_cpp_s2_area(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_area(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_length +NumericVector cpp_s2_length(List geog); +RcppExport SEXP _s2_cpp_s2_length(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_length(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_perimeter +NumericVector cpp_s2_perimeter(List geog); +RcppExport SEXP _s2_cpp_s2_perimeter(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_perimeter(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_x +NumericVector cpp_s2_x(List geog); +RcppExport SEXP _s2_cpp_s2_x(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_x(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_y +NumericVector cpp_s2_y(List geog); +RcppExport SEXP _s2_cpp_s2_y(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_y(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_project_normalized +NumericVector cpp_s2_project_normalized(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_project_normalized(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_project_normalized(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_distance +NumericVector cpp_s2_distance(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_distance(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_distance(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_max_distance +NumericVector cpp_s2_max_distance(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_max_distance(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_max_distance(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_bounds_cap +DataFrame cpp_s2_bounds_cap(List geog); +RcppExport SEXP _s2_cpp_s2_bounds_cap(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_bounds_cap(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_bounds_rect +DataFrame cpp_s2_bounds_rect(List geog); +RcppExport SEXP _s2_cpp_s2_bounds_rect(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_bounds_rect(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_normalize +List cpp_s2_cell_union_normalize(List cellUnionVector); +RcppExport SEXP _s2_cpp_s2_cell_union_normalize(SEXP cellUnionVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector(cellUnionVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_normalize(cellUnionVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_is_na +LogicalVector cpp_s2_cell_union_is_na(List cellUnionVector); +RcppExport SEXP _s2_cpp_s2_cell_union_is_na(SEXP cellUnionVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector(cellUnionVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_is_na(cellUnionVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_contains +LogicalVector cpp_s2_cell_union_contains(List cellUnionVector1, List cellUnionVector2); +RcppExport SEXP _s2_cpp_s2_cell_union_contains(SEXP cellUnionVector1SEXP, SEXP cellUnionVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector1(cellUnionVector1SEXP); + Rcpp::traits::input_parameter< List >::type cellUnionVector2(cellUnionVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_contains(cellUnionVector1, cellUnionVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_contains_cell +LogicalVector cpp_s2_cell_union_contains_cell(List cellUnionVector, NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_union_contains_cell(SEXP cellUnionVectorSEXP, SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector(cellUnionVectorSEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_contains_cell(cellUnionVector, cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_intersects +LogicalVector cpp_s2_cell_union_intersects(List cellUnionVector1, List cellUnionVector2); +RcppExport SEXP _s2_cpp_s2_cell_union_intersects(SEXP cellUnionVector1SEXP, SEXP cellUnionVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector1(cellUnionVector1SEXP); + Rcpp::traits::input_parameter< List >::type cellUnionVector2(cellUnionVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_intersects(cellUnionVector1, cellUnionVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_intersection +List cpp_s2_cell_union_intersection(List cellUnionVector1, List cellUnionVector2); +RcppExport SEXP _s2_cpp_s2_cell_union_intersection(SEXP cellUnionVector1SEXP, SEXP cellUnionVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector1(cellUnionVector1SEXP); + Rcpp::traits::input_parameter< List >::type cellUnionVector2(cellUnionVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_intersection(cellUnionVector1, cellUnionVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_union +List cpp_s2_cell_union_union(List cellUnionVector1, List cellUnionVector2); +RcppExport SEXP _s2_cpp_s2_cell_union_union(SEXP cellUnionVector1SEXP, SEXP cellUnionVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector1(cellUnionVector1SEXP); + Rcpp::traits::input_parameter< List >::type cellUnionVector2(cellUnionVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_union(cellUnionVector1, cellUnionVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_union_difference +List cpp_s2_cell_union_difference(List cellUnionVector1, List cellUnionVector2); +RcppExport SEXP _s2_cpp_s2_cell_union_difference(SEXP cellUnionVector1SEXP, SEXP cellUnionVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector1(cellUnionVector1SEXP); + Rcpp::traits::input_parameter< List >::type cellUnionVector2(cellUnionVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_union_difference(cellUnionVector1, cellUnionVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_geography_from_cell_union +List cpp_s2_geography_from_cell_union(List cellUnionVector); +RcppExport SEXP _s2_cpp_s2_geography_from_cell_union(SEXP cellUnionVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type cellUnionVector(cellUnionVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_geography_from_cell_union(cellUnionVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_covering_cell_ids +List cpp_s2_covering_cell_ids(List geog, int min_level, int max_level, int max_cells, NumericVector buffer, bool interior); +RcppExport SEXP _s2_cpp_s2_covering_cell_ids(SEXP geogSEXP, SEXP min_levelSEXP, SEXP max_levelSEXP, SEXP max_cellsSEXP, SEXP bufferSEXP, SEXP interiorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< int >::type min_level(min_levelSEXP); + Rcpp::traits::input_parameter< int >::type max_level(max_levelSEXP); + Rcpp::traits::input_parameter< int >::type max_cells(max_cellsSEXP); + Rcpp::traits::input_parameter< NumericVector >::type buffer(bufferSEXP); + Rcpp::traits::input_parameter< bool >::type interior(interiorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_covering_cell_ids(geog, min_level, max_level, max_cells, buffer, interior)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_covering_cell_ids_agg +List cpp_s2_covering_cell_ids_agg(List geog, int min_level, int max_level, int max_cells, double buffer, bool interior, bool naRm); +RcppExport SEXP _s2_cpp_s2_covering_cell_ids_agg(SEXP geogSEXP, SEXP min_levelSEXP, SEXP max_levelSEXP, SEXP max_cellsSEXP, SEXP bufferSEXP, SEXP interiorSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< int >::type min_level(min_levelSEXP); + Rcpp::traits::input_parameter< int >::type max_level(max_levelSEXP); + Rcpp::traits::input_parameter< int >::type max_cells(max_cellsSEXP); + Rcpp::traits::input_parameter< double >::type buffer(bufferSEXP); + Rcpp::traits::input_parameter< bool >::type interior(interiorSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_covering_cell_ids_agg(geog, min_level, max_level, max_cells, buffer, interior, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_sentinel +NumericVector cpp_s2_cell_sentinel(); +RcppExport SEXP _s2_cpp_s2_cell_sentinel() { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_sentinel()); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_from_string +NumericVector cpp_s2_cell_from_string(CharacterVector cellString); +RcppExport SEXP _s2_cpp_s2_cell_from_string(SEXP cellStringSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< CharacterVector >::type cellString(cellStringSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_from_string(cellString)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_from_lnglat +NumericVector cpp_s2_cell_from_lnglat(List lnglat); +RcppExport SEXP _s2_cpp_s2_cell_from_lnglat(SEXP lnglatSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type lnglat(lnglatSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_from_lnglat(lnglat)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_to_lnglat +List cpp_s2_cell_to_lnglat(NumericVector cellId); +RcppExport SEXP _s2_cpp_s2_cell_to_lnglat(SEXP cellIdSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellId(cellIdSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_to_lnglat(cellId)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_to_cell_union +List cpp_s2_cell_to_cell_union(NumericVector cellId); +RcppExport SEXP _s2_cpp_s2_cell_to_cell_union(SEXP cellIdSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellId(cellIdSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_to_cell_union(cellId)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_is_na +LogicalVector cpp_s2_cell_is_na(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_is_na(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_is_na(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_sort +NumericVector cpp_s2_cell_sort(NumericVector cellIdVector, bool decreasing); +RcppExport SEXP _s2_cpp_s2_cell_sort(SEXP cellIdVectorSEXP, SEXP decreasingSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< bool >::type decreasing(decreasingSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_sort(cellIdVector, decreasing)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_range +NumericVector cpp_s2_cell_range(NumericVector cellIdVector, bool naRm); +RcppExport SEXP _s2_cpp_s2_cell_range(SEXP cellIdVectorSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_range(cellIdVector, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_unique +NumericVector cpp_s2_cell_unique(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_unique(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_unique(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_to_string +CharacterVector cpp_s2_cell_to_string(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_to_string(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_to_string(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_debug_string +CharacterVector cpp_s2_cell_debug_string(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_debug_string(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_debug_string(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_is_valid +LogicalVector cpp_s2_cell_is_valid(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_is_valid(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_is_valid(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_center +List cpp_s2_cell_center(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_center(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_center(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_polygon +List cpp_s2_cell_polygon(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_polygon(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_polygon(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_vertex +List cpp_s2_cell_vertex(NumericVector cellIdVector, IntegerVector k); +RcppExport SEXP _s2_cpp_s2_cell_vertex(SEXP cellIdVectorSEXP, SEXP kSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< IntegerVector >::type k(kSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_vertex(cellIdVector, k)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_level +IntegerVector cpp_s2_cell_level(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_level(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_level(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_area +NumericVector cpp_s2_cell_area(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_area(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_area(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_area_approx +NumericVector cpp_s2_cell_area_approx(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_area_approx(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_area_approx(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_parent +NumericVector cpp_s2_cell_parent(NumericVector cellIdVector, IntegerVector level); +RcppExport SEXP _s2_cpp_s2_cell_parent(SEXP cellIdVectorSEXP, SEXP levelSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< IntegerVector >::type level(levelSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_parent(cellIdVector, level)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_child +NumericVector cpp_s2_cell_child(NumericVector cellIdVector, IntegerVector k); +RcppExport SEXP _s2_cpp_s2_cell_child(SEXP cellIdVectorSEXP, SEXP kSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< IntegerVector >::type k(kSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_child(cellIdVector, k)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_edge_neighbour +NumericVector cpp_s2_cell_edge_neighbour(NumericVector cellIdVector, IntegerVector k); +RcppExport SEXP _s2_cpp_s2_cell_edge_neighbour(SEXP cellIdVectorSEXP, SEXP kSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + Rcpp::traits::input_parameter< IntegerVector >::type k(kSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_edge_neighbour(cellIdVector, k)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_cummax +NumericVector cpp_s2_cell_cummax(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_cummax(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_cummax(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_cummin +NumericVector cpp_s2_cell_cummin(NumericVector cellIdVector); +RcppExport SEXP _s2_cpp_s2_cell_cummin(SEXP cellIdVectorSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector(cellIdVectorSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_cummin(cellIdVector)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_eq +LogicalVector cpp_s2_cell_eq(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_eq(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_eq(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_neq +LogicalVector cpp_s2_cell_neq(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_neq(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_neq(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_lt +LogicalVector cpp_s2_cell_lt(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_lt(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_lt(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_lte +LogicalVector cpp_s2_cell_lte(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_lte(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_lte(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_gte +LogicalVector cpp_s2_cell_gte(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_gte(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_gte(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_gt +LogicalVector cpp_s2_cell_gt(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_gt(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_gt(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_contains +LogicalVector cpp_s2_cell_contains(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_contains(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_contains(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_may_intersect +LogicalVector cpp_s2_cell_may_intersect(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_may_intersect(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_may_intersect(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_distance +NumericVector cpp_s2_cell_distance(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_distance(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_distance(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_max_distance +NumericVector cpp_s2_cell_max_distance(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_max_distance(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_max_distance(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_common_ancestor_level +IntegerVector cpp_s2_cell_common_ancestor_level(NumericVector cellIdVector1, NumericVector cellIdVector2); +RcppExport SEXP _s2_cpp_s2_cell_common_ancestor_level(SEXP cellIdVector1SEXP, SEXP cellIdVector2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector1(cellIdVector1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type cellIdVector2(cellIdVector2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_common_ancestor_level(cellIdVector1, cellIdVector2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_cell_common_ancestor_level_agg +int cpp_s2_cell_common_ancestor_level_agg(NumericVector cellId); +RcppExport SEXP _s2_cpp_s2_cell_common_ancestor_level_agg(SEXP cellIdSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< NumericVector >::type cellId(cellIdSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_cell_common_ancestor_level_agg(cellId)); + return rcpp_result_gen; +END_RCPP +} +// s2_geography_full +List s2_geography_full(LogicalVector x); +RcppExport SEXP _s2_s2_geography_full(SEXP xSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< LogicalVector >::type x(xSEXP); + rcpp_result_gen = Rcpp::wrap(s2_geography_full(x)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_geography_is_na +LogicalVector cpp_s2_geography_is_na(List geog); +RcppExport SEXP _s2_cpp_s2_geography_is_na(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_geography_is_na(geog)); + return rcpp_result_gen; +END_RCPP +} +// s2_lnglat_from_s2_point +List s2_lnglat_from_s2_point(List s2_point); +RcppExport SEXP _s2_s2_lnglat_from_s2_point(SEXP s2_pointSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type s2_point(s2_pointSEXP); + rcpp_result_gen = Rcpp::wrap(s2_lnglat_from_s2_point(s2_point)); + return rcpp_result_gen; +END_RCPP +} +// s2_point_from_s2_lnglat +List s2_point_from_s2_lnglat(List s2_lnglat); +RcppExport SEXP _s2_s2_point_from_s2_lnglat(SEXP s2_lnglatSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type s2_lnglat(s2_lnglatSEXP); + rcpp_result_gen = Rcpp::wrap(s2_point_from_s2_lnglat(s2_lnglat)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_closest_feature +IntegerVector cpp_s2_closest_feature(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_closest_feature(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_closest_feature(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_farthest_feature +IntegerVector cpp_s2_farthest_feature(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_farthest_feature(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_farthest_feature(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_closest_edges +List cpp_s2_closest_edges(List geog1, List geog2, int n, double min_distance, double max_distance); +RcppExport SEXP _s2_cpp_s2_closest_edges(SEXP geog1SEXP, SEXP geog2SEXP, SEXP nSEXP, SEXP min_distanceSEXP, SEXP max_distanceSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< int >::type n(nSEXP); + Rcpp::traits::input_parameter< double >::type min_distance(min_distanceSEXP); + Rcpp::traits::input_parameter< double >::type max_distance(max_distanceSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_closest_edges(geog1, geog2, n, min_distance, max_distance)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_may_intersect_matrix +List cpp_s2_may_intersect_matrix(List geog1, List geog2, int maxEdgesPerCell, int maxFeatureCells, List s2options); +RcppExport SEXP _s2_cpp_s2_may_intersect_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP maxEdgesPerCellSEXP, SEXP maxFeatureCellsSEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< int >::type maxEdgesPerCell(maxEdgesPerCellSEXP); + Rcpp::traits::input_parameter< int >::type maxFeatureCells(maxFeatureCellsSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_may_intersect_matrix(geog1, geog2, maxEdgesPerCell, maxFeatureCells, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_contains_matrix +List cpp_s2_contains_matrix(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_contains_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_contains_matrix(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_within_matrix +List cpp_s2_within_matrix(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_within_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_within_matrix(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_intersects_matrix +List cpp_s2_intersects_matrix(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_intersects_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_intersects_matrix(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_equals_matrix +List cpp_s2_equals_matrix(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_equals_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_equals_matrix(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_touches_matrix +List cpp_s2_touches_matrix(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_touches_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_touches_matrix(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_dwithin_matrix +List cpp_s2_dwithin_matrix(List geog1, List geog2, double distance); +RcppExport SEXP _s2_cpp_s2_dwithin_matrix(SEXP geog1SEXP, SEXP geog2SEXP, SEXP distanceSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< double >::type distance(distanceSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_dwithin_matrix(geog1, geog2, distance)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_distance_matrix +NumericMatrix cpp_s2_distance_matrix(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_distance_matrix(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_distance_matrix(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_max_distance_matrix +NumericMatrix cpp_s2_max_distance_matrix(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_max_distance_matrix(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_max_distance_matrix(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_contains_matrix_brute_force +List cpp_s2_contains_matrix_brute_force(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_contains_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_contains_matrix_brute_force(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_within_matrix_brute_force +List cpp_s2_within_matrix_brute_force(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_within_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_within_matrix_brute_force(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_intersects_matrix_brute_force +List cpp_s2_intersects_matrix_brute_force(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_intersects_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_intersects_matrix_brute_force(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_disjoint_matrix_brute_force +List cpp_s2_disjoint_matrix_brute_force(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_disjoint_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_disjoint_matrix_brute_force(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_equals_matrix_brute_force +List cpp_s2_equals_matrix_brute_force(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_equals_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_equals_matrix_brute_force(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_dwithin_matrix_brute_force +List cpp_s2_dwithin_matrix_brute_force(List geog1, List geog2, double distance); +RcppExport SEXP _s2_cpp_s2_dwithin_matrix_brute_force(SEXP geog1SEXP, SEXP geog2SEXP, SEXP distanceSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< double >::type distance(distanceSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_dwithin_matrix_brute_force(geog1, geog2, distance)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_intersects +LogicalVector cpp_s2_intersects(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_intersects(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_intersects(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_equals +LogicalVector cpp_s2_equals(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_equals(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_equals(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_contains +LogicalVector cpp_s2_contains(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_contains(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_contains(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_touches +LogicalVector cpp_s2_touches(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_touches(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_touches(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_dwithin +LogicalVector cpp_s2_dwithin(List geog1, List geog2, NumericVector distance); +RcppExport SEXP _s2_cpp_s2_dwithin(SEXP geog1SEXP, SEXP geog2SEXP, SEXP distanceSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< NumericVector >::type distance(distanceSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_dwithin(geog1, geog2, distance)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_prepared_dwithin +LogicalVector cpp_s2_prepared_dwithin(List geog1, List geog2, NumericVector distance); +RcppExport SEXP _s2_cpp_s2_prepared_dwithin(SEXP geog1SEXP, SEXP geog2SEXP, SEXP distanceSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< NumericVector >::type distance(distanceSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_prepared_dwithin(geog1, geog2, distance)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_intersects_box +LogicalVector cpp_s2_intersects_box(List geog, NumericVector lng1, NumericVector lat1, NumericVector lng2, NumericVector lat2, IntegerVector detail, List s2options); +RcppExport SEXP _s2_cpp_s2_intersects_box(SEXP geogSEXP, SEXP lng1SEXP, SEXP lat1SEXP, SEXP lng2SEXP, SEXP lat2SEXP, SEXP detailSEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< NumericVector >::type lng1(lng1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type lat1(lat1SEXP); + Rcpp::traits::input_parameter< NumericVector >::type lng2(lng2SEXP); + Rcpp::traits::input_parameter< NumericVector >::type lat2(lat2SEXP); + Rcpp::traits::input_parameter< IntegerVector >::type detail(detailSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_intersects_box(geog, lng1, lat1, lng2, lat2, detail, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_intersection +List cpp_s2_intersection(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_intersection(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_intersection(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_union +List cpp_s2_union(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_union(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_union(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_difference +List cpp_s2_difference(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_difference(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_difference(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_sym_difference +List cpp_s2_sym_difference(List geog1, List geog2, List s2options); +RcppExport SEXP _s2_cpp_s2_sym_difference(SEXP geog1SEXP, SEXP geog2SEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_sym_difference(geog1, geog2, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_coverage_union_agg +List cpp_s2_coverage_union_agg(List geog, List s2options, bool naRm); +RcppExport SEXP _s2_cpp_s2_coverage_union_agg(SEXP geogSEXP, SEXP s2optionsSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_coverage_union_agg(geog, s2options, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_union_agg +List cpp_s2_union_agg(List geog, List s2options, bool naRm); +RcppExport SEXP _s2_cpp_s2_union_agg(SEXP geogSEXP, SEXP s2optionsSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_union_agg(geog, s2options, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_centroid_agg +List cpp_s2_centroid_agg(List geog, bool naRm); +RcppExport SEXP _s2_cpp_s2_centroid_agg(SEXP geogSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_centroid_agg(geog, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_rebuild_agg +List cpp_s2_rebuild_agg(List geog, List s2options, bool naRm); +RcppExport SEXP _s2_cpp_s2_rebuild_agg(SEXP geogSEXP, SEXP s2optionsSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_rebuild_agg(geog, s2options, naRm)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_closest_point +List cpp_s2_closest_point(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_closest_point(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_closest_point(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_minimum_clearance_line_between +List cpp_s2_minimum_clearance_line_between(List geog1, List geog2); +RcppExport SEXP _s2_cpp_s2_minimum_clearance_line_between(SEXP geog1SEXP, SEXP geog2SEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog1(geog1SEXP); + Rcpp::traits::input_parameter< List >::type geog2(geog2SEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_minimum_clearance_line_between(geog1, geog2)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_centroid +List cpp_s2_centroid(List geog); +RcppExport SEXP _s2_cpp_s2_centroid(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_centroid(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_point_on_surface +List cpp_s2_point_on_surface(List geog); +RcppExport SEXP _s2_cpp_s2_point_on_surface(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_point_on_surface(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_boundary +List cpp_s2_boundary(List geog); +RcppExport SEXP _s2_cpp_s2_boundary(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_boundary(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_rebuild +List cpp_s2_rebuild(List geog, List s2options); +RcppExport SEXP _s2_cpp_s2_rebuild(SEXP geogSEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_rebuild(geog, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_unary_union +List cpp_s2_unary_union(List geog, List s2options); +RcppExport SEXP _s2_cpp_s2_unary_union(SEXP geogSEXP, SEXP s2optionsSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< List >::type s2options(s2optionsSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_unary_union(geog, s2options)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_interpolate_normalized +List cpp_s2_interpolate_normalized(List geog, NumericVector distanceNormalized); +RcppExport SEXP _s2_cpp_s2_interpolate_normalized(SEXP geogSEXP, SEXP distanceNormalizedSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< NumericVector >::type distanceNormalized(distanceNormalizedSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_interpolate_normalized(geog, distanceNormalized)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_buffer_cells +List cpp_s2_buffer_cells(List geog, NumericVector distance, int maxCells, int minLevel); +RcppExport SEXP _s2_cpp_s2_buffer_cells(SEXP geogSEXP, SEXP distanceSEXP, SEXP maxCellsSEXP, SEXP minLevelSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< NumericVector >::type distance(distanceSEXP); + Rcpp::traits::input_parameter< int >::type maxCells(maxCellsSEXP); + Rcpp::traits::input_parameter< int >::type minLevel(minLevelSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_buffer_cells(geog, distance, maxCells, minLevel)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_convex_hull +List cpp_s2_convex_hull(List geog); +RcppExport SEXP _s2_cpp_s2_convex_hull(SEXP geogSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_convex_hull(geog)); + return rcpp_result_gen; +END_RCPP +} +// cpp_s2_convex_hull_agg +List cpp_s2_convex_hull_agg(List geog, bool naRm); +RcppExport SEXP _s2_cpp_s2_convex_hull_agg(SEXP geogSEXP, SEXP naRmSEXP) { +BEGIN_RCPP + Rcpp::RObject rcpp_result_gen; + Rcpp::RNGScope rcpp_rngScope_gen; + Rcpp::traits::input_parameter< List >::type geog(geogSEXP); + Rcpp::traits::input_parameter< bool >::type naRm(naRmSEXP); + rcpp_result_gen = Rcpp::wrap(cpp_s2_convex_hull_agg(geog, naRm)); + return rcpp_result_gen; +END_RCPP +} + +RcppExport SEXP c_s2_geography_writer_new(SEXP, SEXP, SEXP, SEXP); +RcppExport SEXP c_s2_handle_geography(SEXP, SEXP); +RcppExport SEXP c_s2_handle_geography_tessellated(SEXP, SEXP); +RcppExport SEXP c_s2_projection_mercator(SEXP); +RcppExport SEXP c_s2_projection_orthographic(SEXP); +RcppExport SEXP c_s2_projection_plate_carree(SEXP); +RcppExport SEXP c_s2_trans_s2_lnglat_new(); +RcppExport SEXP c_s2_trans_s2_point_new(); + +static const R_CallMethodDef CallEntries[] = { + {"_s2_cpp_s2_init", (DL_FUNC) &_s2_cpp_s2_init, 0}, + {"_s2_cpp_s2_is_collection", (DL_FUNC) &_s2_cpp_s2_is_collection, 1}, + {"_s2_cpp_s2_is_valid", (DL_FUNC) &_s2_cpp_s2_is_valid, 1}, + {"_s2_cpp_s2_is_valid_reason", (DL_FUNC) &_s2_cpp_s2_is_valid_reason, 1}, + {"_s2_cpp_s2_dimension", (DL_FUNC) &_s2_cpp_s2_dimension, 1}, + {"_s2_cpp_s2_num_points", (DL_FUNC) &_s2_cpp_s2_num_points, 1}, + {"_s2_cpp_s2_is_empty", (DL_FUNC) &_s2_cpp_s2_is_empty, 1}, + {"_s2_cpp_s2_area", (DL_FUNC) &_s2_cpp_s2_area, 1}, + {"_s2_cpp_s2_length", (DL_FUNC) &_s2_cpp_s2_length, 1}, + {"_s2_cpp_s2_perimeter", (DL_FUNC) &_s2_cpp_s2_perimeter, 1}, + {"_s2_cpp_s2_x", (DL_FUNC) &_s2_cpp_s2_x, 1}, + {"_s2_cpp_s2_y", (DL_FUNC) &_s2_cpp_s2_y, 1}, + {"_s2_cpp_s2_project_normalized", (DL_FUNC) &_s2_cpp_s2_project_normalized, 2}, + {"_s2_cpp_s2_distance", (DL_FUNC) &_s2_cpp_s2_distance, 2}, + {"_s2_cpp_s2_max_distance", (DL_FUNC) &_s2_cpp_s2_max_distance, 2}, + {"_s2_cpp_s2_bounds_cap", (DL_FUNC) &_s2_cpp_s2_bounds_cap, 1}, + {"_s2_cpp_s2_bounds_rect", (DL_FUNC) &_s2_cpp_s2_bounds_rect, 1}, + {"_s2_cpp_s2_cell_union_normalize", (DL_FUNC) &_s2_cpp_s2_cell_union_normalize, 1}, + {"_s2_cpp_s2_cell_union_is_na", (DL_FUNC) &_s2_cpp_s2_cell_union_is_na, 1}, + {"_s2_cpp_s2_cell_union_contains", (DL_FUNC) &_s2_cpp_s2_cell_union_contains, 2}, + {"_s2_cpp_s2_cell_union_contains_cell", (DL_FUNC) &_s2_cpp_s2_cell_union_contains_cell, 2}, + {"_s2_cpp_s2_cell_union_intersects", (DL_FUNC) &_s2_cpp_s2_cell_union_intersects, 2}, + {"_s2_cpp_s2_cell_union_intersection", (DL_FUNC) &_s2_cpp_s2_cell_union_intersection, 2}, + {"_s2_cpp_s2_cell_union_union", (DL_FUNC) &_s2_cpp_s2_cell_union_union, 2}, + {"_s2_cpp_s2_cell_union_difference", (DL_FUNC) &_s2_cpp_s2_cell_union_difference, 2}, + {"_s2_cpp_s2_geography_from_cell_union", (DL_FUNC) &_s2_cpp_s2_geography_from_cell_union, 1}, + {"_s2_cpp_s2_covering_cell_ids", (DL_FUNC) &_s2_cpp_s2_covering_cell_ids, 6}, + {"_s2_cpp_s2_covering_cell_ids_agg", (DL_FUNC) &_s2_cpp_s2_covering_cell_ids_agg, 7}, + {"_s2_cpp_s2_cell_sentinel", (DL_FUNC) &_s2_cpp_s2_cell_sentinel, 0}, + {"_s2_cpp_s2_cell_from_string", (DL_FUNC) &_s2_cpp_s2_cell_from_string, 1}, + {"_s2_cpp_s2_cell_from_lnglat", (DL_FUNC) &_s2_cpp_s2_cell_from_lnglat, 1}, + {"_s2_cpp_s2_cell_to_lnglat", (DL_FUNC) &_s2_cpp_s2_cell_to_lnglat, 1}, + {"_s2_cpp_s2_cell_to_cell_union", (DL_FUNC) &_s2_cpp_s2_cell_to_cell_union, 1}, + {"_s2_cpp_s2_cell_is_na", (DL_FUNC) &_s2_cpp_s2_cell_is_na, 1}, + {"_s2_cpp_s2_cell_sort", (DL_FUNC) &_s2_cpp_s2_cell_sort, 2}, + {"_s2_cpp_s2_cell_range", (DL_FUNC) &_s2_cpp_s2_cell_range, 2}, + {"_s2_cpp_s2_cell_unique", (DL_FUNC) &_s2_cpp_s2_cell_unique, 1}, + {"_s2_cpp_s2_cell_to_string", (DL_FUNC) &_s2_cpp_s2_cell_to_string, 1}, + {"_s2_cpp_s2_cell_debug_string", (DL_FUNC) &_s2_cpp_s2_cell_debug_string, 1}, + {"_s2_cpp_s2_cell_is_valid", (DL_FUNC) &_s2_cpp_s2_cell_is_valid, 1}, + {"_s2_cpp_s2_cell_center", (DL_FUNC) &_s2_cpp_s2_cell_center, 1}, + {"_s2_cpp_s2_cell_polygon", (DL_FUNC) &_s2_cpp_s2_cell_polygon, 1}, + {"_s2_cpp_s2_cell_vertex", (DL_FUNC) &_s2_cpp_s2_cell_vertex, 2}, + {"_s2_cpp_s2_cell_level", (DL_FUNC) &_s2_cpp_s2_cell_level, 1}, + {"_s2_cpp_s2_cell_area", (DL_FUNC) &_s2_cpp_s2_cell_area, 1}, + {"_s2_cpp_s2_cell_area_approx", (DL_FUNC) &_s2_cpp_s2_cell_area_approx, 1}, + {"_s2_cpp_s2_cell_parent", (DL_FUNC) &_s2_cpp_s2_cell_parent, 2}, + {"_s2_cpp_s2_cell_child", (DL_FUNC) &_s2_cpp_s2_cell_child, 2}, + {"_s2_cpp_s2_cell_edge_neighbour", (DL_FUNC) &_s2_cpp_s2_cell_edge_neighbour, 2}, + {"_s2_cpp_s2_cell_cummax", (DL_FUNC) &_s2_cpp_s2_cell_cummax, 1}, + {"_s2_cpp_s2_cell_cummin", (DL_FUNC) &_s2_cpp_s2_cell_cummin, 1}, + {"_s2_cpp_s2_cell_eq", (DL_FUNC) &_s2_cpp_s2_cell_eq, 2}, + {"_s2_cpp_s2_cell_neq", (DL_FUNC) &_s2_cpp_s2_cell_neq, 2}, + {"_s2_cpp_s2_cell_lt", (DL_FUNC) &_s2_cpp_s2_cell_lt, 2}, + {"_s2_cpp_s2_cell_lte", (DL_FUNC) &_s2_cpp_s2_cell_lte, 2}, + {"_s2_cpp_s2_cell_gte", (DL_FUNC) &_s2_cpp_s2_cell_gte, 2}, + {"_s2_cpp_s2_cell_gt", (DL_FUNC) &_s2_cpp_s2_cell_gt, 2}, + {"_s2_cpp_s2_cell_contains", (DL_FUNC) &_s2_cpp_s2_cell_contains, 2}, + {"_s2_cpp_s2_cell_may_intersect", (DL_FUNC) &_s2_cpp_s2_cell_may_intersect, 2}, + {"_s2_cpp_s2_cell_distance", (DL_FUNC) &_s2_cpp_s2_cell_distance, 2}, + {"_s2_cpp_s2_cell_max_distance", (DL_FUNC) &_s2_cpp_s2_cell_max_distance, 2}, + {"_s2_cpp_s2_cell_common_ancestor_level", (DL_FUNC) &_s2_cpp_s2_cell_common_ancestor_level, 2}, + {"_s2_cpp_s2_cell_common_ancestor_level_agg", (DL_FUNC) &_s2_cpp_s2_cell_common_ancestor_level_agg, 1}, + {"_s2_s2_geography_full", (DL_FUNC) &_s2_s2_geography_full, 1}, + {"_s2_cpp_s2_geography_is_na", (DL_FUNC) &_s2_cpp_s2_geography_is_na, 1}, + {"_s2_s2_lnglat_from_s2_point", (DL_FUNC) &_s2_s2_lnglat_from_s2_point, 1}, + {"_s2_s2_point_from_s2_lnglat", (DL_FUNC) &_s2_s2_point_from_s2_lnglat, 1}, + {"_s2_cpp_s2_closest_feature", (DL_FUNC) &_s2_cpp_s2_closest_feature, 2}, + {"_s2_cpp_s2_farthest_feature", (DL_FUNC) &_s2_cpp_s2_farthest_feature, 2}, + {"_s2_cpp_s2_closest_edges", (DL_FUNC) &_s2_cpp_s2_closest_edges, 5}, + {"_s2_cpp_s2_may_intersect_matrix", (DL_FUNC) &_s2_cpp_s2_may_intersect_matrix, 5}, + {"_s2_cpp_s2_contains_matrix", (DL_FUNC) &_s2_cpp_s2_contains_matrix, 3}, + {"_s2_cpp_s2_within_matrix", (DL_FUNC) &_s2_cpp_s2_within_matrix, 3}, + {"_s2_cpp_s2_intersects_matrix", (DL_FUNC) &_s2_cpp_s2_intersects_matrix, 3}, + {"_s2_cpp_s2_equals_matrix", (DL_FUNC) &_s2_cpp_s2_equals_matrix, 3}, + {"_s2_cpp_s2_touches_matrix", (DL_FUNC) &_s2_cpp_s2_touches_matrix, 3}, + {"_s2_cpp_s2_dwithin_matrix", (DL_FUNC) &_s2_cpp_s2_dwithin_matrix, 3}, + {"_s2_cpp_s2_distance_matrix", (DL_FUNC) &_s2_cpp_s2_distance_matrix, 2}, + {"_s2_cpp_s2_max_distance_matrix", (DL_FUNC) &_s2_cpp_s2_max_distance_matrix, 2}, + {"_s2_cpp_s2_contains_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_contains_matrix_brute_force, 3}, + {"_s2_cpp_s2_within_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_within_matrix_brute_force, 3}, + {"_s2_cpp_s2_intersects_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_intersects_matrix_brute_force, 3}, + {"_s2_cpp_s2_disjoint_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_disjoint_matrix_brute_force, 3}, + {"_s2_cpp_s2_equals_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_equals_matrix_brute_force, 3}, + {"_s2_cpp_s2_dwithin_matrix_brute_force", (DL_FUNC) &_s2_cpp_s2_dwithin_matrix_brute_force, 3}, + {"_s2_cpp_s2_intersects", (DL_FUNC) &_s2_cpp_s2_intersects, 3}, + {"_s2_cpp_s2_equals", (DL_FUNC) &_s2_cpp_s2_equals, 3}, + {"_s2_cpp_s2_contains", (DL_FUNC) &_s2_cpp_s2_contains, 3}, + {"_s2_cpp_s2_touches", (DL_FUNC) &_s2_cpp_s2_touches, 3}, + {"_s2_cpp_s2_dwithin", (DL_FUNC) &_s2_cpp_s2_dwithin, 3}, + {"_s2_cpp_s2_prepared_dwithin", (DL_FUNC) &_s2_cpp_s2_prepared_dwithin, 3}, + {"_s2_cpp_s2_intersects_box", (DL_FUNC) &_s2_cpp_s2_intersects_box, 7}, + {"_s2_cpp_s2_intersection", (DL_FUNC) &_s2_cpp_s2_intersection, 3}, + {"_s2_cpp_s2_union", (DL_FUNC) &_s2_cpp_s2_union, 3}, + {"_s2_cpp_s2_difference", (DL_FUNC) &_s2_cpp_s2_difference, 3}, + {"_s2_cpp_s2_sym_difference", (DL_FUNC) &_s2_cpp_s2_sym_difference, 3}, + {"_s2_cpp_s2_coverage_union_agg", (DL_FUNC) &_s2_cpp_s2_coverage_union_agg, 3}, + {"_s2_cpp_s2_union_agg", (DL_FUNC) &_s2_cpp_s2_union_agg, 3}, + {"_s2_cpp_s2_centroid_agg", (DL_FUNC) &_s2_cpp_s2_centroid_agg, 2}, + {"_s2_cpp_s2_rebuild_agg", (DL_FUNC) &_s2_cpp_s2_rebuild_agg, 3}, + {"_s2_cpp_s2_closest_point", (DL_FUNC) &_s2_cpp_s2_closest_point, 2}, + {"_s2_cpp_s2_minimum_clearance_line_between", (DL_FUNC) &_s2_cpp_s2_minimum_clearance_line_between, 2}, + {"_s2_cpp_s2_centroid", (DL_FUNC) &_s2_cpp_s2_centroid, 1}, + {"_s2_cpp_s2_point_on_surface", (DL_FUNC) &_s2_cpp_s2_point_on_surface, 1}, + {"_s2_cpp_s2_boundary", (DL_FUNC) &_s2_cpp_s2_boundary, 1}, + {"_s2_cpp_s2_rebuild", (DL_FUNC) &_s2_cpp_s2_rebuild, 2}, + {"_s2_cpp_s2_unary_union", (DL_FUNC) &_s2_cpp_s2_unary_union, 2}, + {"_s2_cpp_s2_interpolate_normalized", (DL_FUNC) &_s2_cpp_s2_interpolate_normalized, 2}, + {"_s2_cpp_s2_buffer_cells", (DL_FUNC) &_s2_cpp_s2_buffer_cells, 4}, + {"_s2_cpp_s2_convex_hull", (DL_FUNC) &_s2_cpp_s2_convex_hull, 1}, + {"_s2_cpp_s2_convex_hull_agg", (DL_FUNC) &_s2_cpp_s2_convex_hull_agg, 2}, + {"c_s2_geography_writer_new", (DL_FUNC) &c_s2_geography_writer_new, 4}, + {"c_s2_handle_geography", (DL_FUNC) &c_s2_handle_geography, 2}, + {"c_s2_handle_geography_tessellated", (DL_FUNC) &c_s2_handle_geography_tessellated, 2}, + {"c_s2_projection_mercator", (DL_FUNC) &c_s2_projection_mercator, 1}, + {"c_s2_projection_orthographic", (DL_FUNC) &c_s2_projection_orthographic, 1}, + {"c_s2_projection_plate_carree", (DL_FUNC) &c_s2_projection_plate_carree, 1}, + {"c_s2_trans_s2_lnglat_new", (DL_FUNC) &c_s2_trans_s2_lnglat_new, 0}, + {"c_s2_trans_s2_point_new", (DL_FUNC) &c_s2_trans_s2_point_new, 0}, + {NULL, NULL, 0} +}; + +RcppExport void R_init_s2(DllInfo *dll) { + R_registerRoutines(dll, NULL, CallEntries, NULL, NULL); + R_useDynamicSymbols(dll, FALSE); +} diff --git a/src/absl/algorithm/algorithm.h b/src/absl/algorithm/algorithm.h new file mode 100644 index 0000000..e9b4733 --- /dev/null +++ b/src/absl/algorithm/algorithm.h @@ -0,0 +1,159 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: algorithm.h +// ----------------------------------------------------------------------------- +// +// This header file contains Google extensions to the standard C++ +// header. + +#ifndef ABSL_ALGORITHM_ALGORITHM_H_ +#define ABSL_ALGORITHM_ALGORITHM_H_ + +#include +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +namespace algorithm_internal { + +// Performs comparisons with operator==, similar to C++14's `std::equal_to<>`. +struct EqualTo { + template + bool operator()(const T& a, const U& b) const { + return a == b; + } +}; + +template +bool EqualImpl(InputIter1 first1, InputIter1 last1, InputIter2 first2, + InputIter2 last2, Pred pred, std::input_iterator_tag, + std::input_iterator_tag) { + while (true) { + if (first1 == last1) return first2 == last2; + if (first2 == last2) return false; + if (!pred(*first1, *first2)) return false; + ++first1; + ++first2; + } +} + +template +bool EqualImpl(InputIter1 first1, InputIter1 last1, InputIter2 first2, + InputIter2 last2, Pred&& pred, std::random_access_iterator_tag, + std::random_access_iterator_tag) { + return (last1 - first1 == last2 - first2) && + std::equal(first1, last1, first2, std::forward(pred)); +} + +// When we are using our own internal predicate that just applies operator==, we +// forward to the non-predicate form of std::equal. This enables an optimization +// in libstdc++ that can result in std::memcmp being used for integer types. +template +bool EqualImpl(InputIter1 first1, InputIter1 last1, InputIter2 first2, + InputIter2 last2, algorithm_internal::EqualTo /* unused */, + std::random_access_iterator_tag, + std::random_access_iterator_tag) { + return (last1 - first1 == last2 - first2) && + std::equal(first1, last1, first2); +} + +template +It RotateImpl(It first, It middle, It last, std::true_type) { + return std::rotate(first, middle, last); +} + +template +It RotateImpl(It first, It middle, It last, std::false_type) { + std::rotate(first, middle, last); + return std::next(first, std::distance(middle, last)); +} + +} // namespace algorithm_internal + +// equal() +// +// Compares the equality of two ranges specified by pairs of iterators, using +// the given predicate, returning true iff for each corresponding iterator i1 +// and i2 in the first and second range respectively, pred(*i1, *i2) == true +// +// This comparison takes at most min(`last1` - `first1`, `last2` - `first2`) +// invocations of the predicate. Additionally, if InputIter1 and InputIter2 are +// both random-access iterators, and `last1` - `first1` != `last2` - `first2`, +// then the predicate is never invoked and the function returns false. +// +// This is a C++11-compatible implementation of C++14 `std::equal`. See +// https://en.cppreference.com/w/cpp/algorithm/equal for more information. +template +bool equal(InputIter1 first1, InputIter1 last1, InputIter2 first2, + InputIter2 last2, Pred&& pred) { + return algorithm_internal::EqualImpl( + first1, last1, first2, last2, std::forward(pred), + typename std::iterator_traits::iterator_category{}, + typename std::iterator_traits::iterator_category{}); +} + +// Overload of equal() that performs comparison of two ranges specified by pairs +// of iterators using operator==. +template +bool equal(InputIter1 first1, InputIter1 last1, InputIter2 first2, + InputIter2 last2) { + return absl::equal(first1, last1, first2, last2, + algorithm_internal::EqualTo{}); +} + +// linear_search() +// +// Performs a linear search for `value` using the iterator `first` up to +// but not including `last`, returning true if [`first`, `last`) contains an +// element equal to `value`. +// +// A linear search is of O(n) complexity which is guaranteed to make at most +// n = (`last` - `first`) comparisons. A linear search over short containers +// may be faster than a binary search, even when the container is sorted. +template +bool linear_search(InputIterator first, InputIterator last, + const EqualityComparable& value) { + return std::find(first, last, value) != last; +} + +// rotate() +// +// Performs a left rotation on a range of elements (`first`, `last`) such that +// `middle` is now the first element. `rotate()` returns an iterator pointing to +// the first element before rotation. This function is exactly the same as +// `std::rotate`, but fixes a bug in gcc +// <= 4.9 where `std::rotate` returns `void` instead of an iterator. +// +// The complexity of this algorithm is the same as that of `std::rotate`, but if +// `ForwardIterator` is not a random-access iterator, then `absl::rotate` +// performs an additional pass over the range to construct the return value. +template +ForwardIterator rotate(ForwardIterator first, ForwardIterator middle, + ForwardIterator last) { + return algorithm_internal::RotateImpl( + first, middle, last, + std::is_same()); +} + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_ALGORITHM_ALGORITHM_H_ diff --git a/src/absl/algorithm/container.h b/src/absl/algorithm/container.h new file mode 100644 index 0000000..26b1952 --- /dev/null +++ b/src/absl/algorithm/container.h @@ -0,0 +1,1774 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: container.h +// ----------------------------------------------------------------------------- +// +// This header file provides Container-based versions of algorithmic functions +// within the C++ standard library. The following standard library sets of +// functions are covered within this file: +// +// * Algorithmic functions +// * Algorithmic functions +// * functions +// +// The standard library functions operate on iterator ranges; the functions +// within this API operate on containers, though many return iterator ranges. +// +// All functions within this API are named with a `c_` prefix. Calls such as +// `absl::c_xx(container, ...) are equivalent to std:: functions such as +// `std::xx(std::begin(cont), std::end(cont), ...)`. Functions that act on +// iterators but not conceptually on iterator ranges (e.g. `std::iter_swap`) +// have no equivalent here. +// +// For template parameter and variable naming, `C` indicates the container type +// to which the function is applied, `Pred` indicates the predicate object type +// to be used by the function and `T` indicates the applicable element type. + +#ifndef ABSL_ALGORITHM_CONTAINER_H_ +#define ABSL_ALGORITHM_CONTAINER_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/algorithm/algorithm.h" +#include "absl/base/macros.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_algorithm_internal { + +// NOTE: it is important to defer to ADL lookup for building with C++ modules, +// especially for headers like which are not visible from this file +// but specialize std::begin and std::end. +using std::begin; +using std::end; + +// The type of the iterator given by begin(c) (possibly std::begin(c)). +// ContainerIter> gives vector::const_iterator, +// while ContainerIter> gives vector::iterator. +template +using ContainerIter = decltype(begin(std::declval())); + +// An MSVC bug involving template parameter substitution requires us to use +// decltype() here instead of just std::pair. +template +using ContainerIterPairType = + decltype(std::make_pair(ContainerIter(), ContainerIter())); + +template +using ContainerDifferenceType = + decltype(std::distance(std::declval>(), + std::declval>())); + +template +using ContainerPointerType = + typename std::iterator_traits>::pointer; + +// container_algorithm_internal::c_begin and +// container_algorithm_internal::c_end are abbreviations for proper ADL +// lookup of std::begin and std::end, i.e. +// using std::begin; +// using std::end; +// std::foo(begin(c), end(c)); +// becomes +// std::foo(container_algorithm_internal::begin(c), +// container_algorithm_internal::end(c)); +// These are meant for internal use only. + +template +ContainerIter c_begin(C& c) { return begin(c); } + +template +ContainerIter c_end(C& c) { return end(c); } + +template +struct IsUnorderedContainer : std::false_type {}; + +template +struct IsUnorderedContainer< + std::unordered_map> : std::true_type {}; + +template +struct IsUnorderedContainer> + : std::true_type {}; + +// container_algorithm_internal::c_size. It is meant for internal use only. + +template +auto c_size(C& c) -> decltype(c.size()) { + return c.size(); +} + +template +constexpr std::size_t c_size(T (&)[N]) { + return N; +} + +} // namespace container_algorithm_internal + +// PUBLIC API + +//------------------------------------------------------------------------------ +// Abseil algorithm.h functions +//------------------------------------------------------------------------------ + +// c_linear_search() +// +// Container-based version of absl::linear_search() for performing a linear +// search within a container. +template +bool c_linear_search(const C& c, EqualityComparable&& value) { + return linear_search(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(value)); +} + +//------------------------------------------------------------------------------ +// algorithms +//------------------------------------------------------------------------------ + +// c_distance() +// +// Container-based version of the `std::distance()` function to +// return the number of elements within a container. +template +container_algorithm_internal::ContainerDifferenceType c_distance( + const C& c) { + return std::distance(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +//------------------------------------------------------------------------------ +// Non-modifying sequence operations +//------------------------------------------------------------------------------ + +// c_all_of() +// +// Container-based version of the `std::all_of()` function to +// test if all elements within a container satisfy a condition. +template +bool c_all_of(const C& c, Pred&& pred) { + return std::all_of(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_any_of() +// +// Container-based version of the `std::any_of()` function to +// test if any element in a container fulfills a condition. +template +bool c_any_of(const C& c, Pred&& pred) { + return std::any_of(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_none_of() +// +// Container-based version of the `std::none_of()` function to +// test if no elements in a container fulfill a condition. +template +bool c_none_of(const C& c, Pred&& pred) { + return std::none_of(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_for_each() +// +// Container-based version of the `std::for_each()` function to +// apply a function to a container's elements. +template +decay_t c_for_each(C&& c, Function&& f) { + return std::for_each(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(f)); +} + +// c_find() +// +// Container-based version of the `std::find()` function to find +// the first element containing the passed value within a container value. +template +container_algorithm_internal::ContainerIter c_find(C& c, T&& value) { + return std::find(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(value)); +} + +// c_find_if() +// +// Container-based version of the `std::find_if()` function to find +// the first element in a container matching the given condition. +template +container_algorithm_internal::ContainerIter c_find_if(C& c, Pred&& pred) { + return std::find_if(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_find_if_not() +// +// Container-based version of the `std::find_if_not()` function to +// find the first element in a container not matching the given condition. +template +container_algorithm_internal::ContainerIter c_find_if_not(C& c, + Pred&& pred) { + return std::find_if_not(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_find_end() +// +// Container-based version of the `std::find_end()` function to +// find the last subsequence within a container. +template +container_algorithm_internal::ContainerIter c_find_end( + Sequence1& sequence, Sequence2& subsequence) { + return std::find_end(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(subsequence), + container_algorithm_internal::c_end(subsequence)); +} + +// Overload of c_find_end() for using a predicate evaluation other than `==` as +// the function's test condition. +template +container_algorithm_internal::ContainerIter c_find_end( + Sequence1& sequence, Sequence2& subsequence, BinaryPredicate&& pred) { + return std::find_end(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(subsequence), + container_algorithm_internal::c_end(subsequence), + std::forward(pred)); +} + +// c_find_first_of() +// +// Container-based version of the `std::find_first_of()` function to +// find the first element within the container that is also within the options +// container. +template +container_algorithm_internal::ContainerIter c_find_first_of(C1& container, + C2& options) { + return std::find_first_of(container_algorithm_internal::c_begin(container), + container_algorithm_internal::c_end(container), + container_algorithm_internal::c_begin(options), + container_algorithm_internal::c_end(options)); +} + +// Overload of c_find_first_of() for using a predicate evaluation other than +// `==` as the function's test condition. +template +container_algorithm_internal::ContainerIter c_find_first_of( + C1& container, C2& options, BinaryPredicate&& pred) { + return std::find_first_of(container_algorithm_internal::c_begin(container), + container_algorithm_internal::c_end(container), + container_algorithm_internal::c_begin(options), + container_algorithm_internal::c_end(options), + std::forward(pred)); +} + +// c_adjacent_find() +// +// Container-based version of the `std::adjacent_find()` function to +// find equal adjacent elements within a container. +template +container_algorithm_internal::ContainerIter c_adjacent_find( + Sequence& sequence) { + return std::adjacent_find(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_adjacent_find() for using a predicate evaluation other than +// `==` as the function's test condition. +template +container_algorithm_internal::ContainerIter c_adjacent_find( + Sequence& sequence, BinaryPredicate&& pred) { + return std::adjacent_find(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(pred)); +} + +// c_count() +// +// Container-based version of the `std::count()` function to count +// values that match within a container. +template +container_algorithm_internal::ContainerDifferenceType c_count( + const C& c, T&& value) { + return std::count(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(value)); +} + +// c_count_if() +// +// Container-based version of the `std::count_if()` function to +// count values matching a condition within a container. +template +container_algorithm_internal::ContainerDifferenceType c_count_if( + const C& c, Pred&& pred) { + return std::count_if(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_mismatch() +// +// Container-based version of the `std::mismatch()` function to +// return the first element where two ordered containers differ. Applies `==` to +// the first N elements of `c1` and `c2`, where N = min(size(c1), size(c2)). +template +container_algorithm_internal::ContainerIterPairType +c_mismatch(C1& c1, C2& c2) { + auto first1 = container_algorithm_internal::c_begin(c1); + auto last1 = container_algorithm_internal::c_end(c1); + auto first2 = container_algorithm_internal::c_begin(c2); + auto last2 = container_algorithm_internal::c_end(c2); + + for (; first1 != last1 && first2 != last2; ++first1, (void)++first2) { + // Negates equality because Cpp17EqualityComparable doesn't require clients + // to overload both `operator==` and `operator!=`. + if (!(*first1 == *first2)) { + break; + } + } + + return std::make_pair(first1, first2); +} + +// Overload of c_mismatch() for using a predicate evaluation other than `==` as +// the function's test condition. Applies `pred`to the first N elements of `c1` +// and `c2`, where N = min(size(c1), size(c2)). +template +container_algorithm_internal::ContainerIterPairType +c_mismatch(C1& c1, C2& c2, BinaryPredicate pred) { + auto first1 = container_algorithm_internal::c_begin(c1); + auto last1 = container_algorithm_internal::c_end(c1); + auto first2 = container_algorithm_internal::c_begin(c2); + auto last2 = container_algorithm_internal::c_end(c2); + + for (; first1 != last1 && first2 != last2; ++first1, (void)++first2) { + if (!pred(*first1, *first2)) { + break; + } + } + + return std::make_pair(first1, first2); +} + +// c_equal() +// +// Container-based version of the `std::equal()` function to +// test whether two containers are equal. +// +// NOTE: the semantics of c_equal() are slightly different than those of +// equal(): while the latter iterates over the second container only up to the +// size of the first container, c_equal() also checks whether the container +// sizes are equal. This better matches expectations about c_equal() based on +// its signature. +// +// Example: +// vector v1 = <1, 2, 3>; +// vector v2 = <1, 2, 3, 4>; +// equal(std::begin(v1), std::end(v1), std::begin(v2)) returns true +// c_equal(v1, v2) returns false + +template +bool c_equal(const C1& c1, const C2& c2) { + return ((container_algorithm_internal::c_size(c1) == + container_algorithm_internal::c_size(c2)) && + std::equal(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2))); +} + +// Overload of c_equal() for using a predicate evaluation other than `==` as +// the function's test condition. +template +bool c_equal(const C1& c1, const C2& c2, BinaryPredicate&& pred) { + return ((container_algorithm_internal::c_size(c1) == + container_algorithm_internal::c_size(c2)) && + std::equal(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + std::forward(pred))); +} + +// c_is_permutation() +// +// Container-based version of the `std::is_permutation()` function +// to test whether a container is a permutation of another. +template +bool c_is_permutation(const C1& c1, const C2& c2) { + using std::begin; + using std::end; + return c1.size() == c2.size() && + std::is_permutation(begin(c1), end(c1), begin(c2)); +} + +// Overload of c_is_permutation() for using a predicate evaluation other than +// `==` as the function's test condition. +template +bool c_is_permutation(const C1& c1, const C2& c2, BinaryPredicate&& pred) { + using std::begin; + using std::end; + return c1.size() == c2.size() && + std::is_permutation(begin(c1), end(c1), begin(c2), + std::forward(pred)); +} + +// c_search() +// +// Container-based version of the `std::search()` function to search +// a container for a subsequence. +template +container_algorithm_internal::ContainerIter c_search( + Sequence1& sequence, Sequence2& subsequence) { + return std::search(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(subsequence), + container_algorithm_internal::c_end(subsequence)); +} + +// Overload of c_search() for using a predicate evaluation other than +// `==` as the function's test condition. +template +container_algorithm_internal::ContainerIter c_search( + Sequence1& sequence, Sequence2& subsequence, BinaryPredicate&& pred) { + return std::search(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(subsequence), + container_algorithm_internal::c_end(subsequence), + std::forward(pred)); +} + +// c_search_n() +// +// Container-based version of the `std::search_n()` function to +// search a container for the first sequence of N elements. +template +container_algorithm_internal::ContainerIter c_search_n( + Sequence& sequence, Size count, T&& value) { + return std::search_n(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), count, + std::forward(value)); +} + +// Overload of c_search_n() for using a predicate evaluation other than +// `==` as the function's test condition. +template +container_algorithm_internal::ContainerIter c_search_n( + Sequence& sequence, Size count, T&& value, BinaryPredicate&& pred) { + return std::search_n(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), count, + std::forward(value), + std::forward(pred)); +} + +//------------------------------------------------------------------------------ +// Modifying sequence operations +//------------------------------------------------------------------------------ + +// c_copy() +// +// Container-based version of the `std::copy()` function to copy a +// container's elements into an iterator. +template +OutputIterator c_copy(const InputSequence& input, OutputIterator output) { + return std::copy(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), output); +} + +// c_copy_n() +// +// Container-based version of the `std::copy_n()` function to copy a +// container's first N elements into an iterator. +template +OutputIterator c_copy_n(const C& input, Size n, OutputIterator output) { + return std::copy_n(container_algorithm_internal::c_begin(input), n, output); +} + +// c_copy_if() +// +// Container-based version of the `std::copy_if()` function to copy +// a container's elements satisfying some condition into an iterator. +template +OutputIterator c_copy_if(const InputSequence& input, OutputIterator output, + Pred&& pred) { + return std::copy_if(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), output, + std::forward(pred)); +} + +// c_copy_backward() +// +// Container-based version of the `std::copy_backward()` function to +// copy a container's elements in reverse order into an iterator. +template +BidirectionalIterator c_copy_backward(const C& src, + BidirectionalIterator dest) { + return std::copy_backward(container_algorithm_internal::c_begin(src), + container_algorithm_internal::c_end(src), dest); +} + +// c_move() +// +// Container-based version of the `std::move()` function to move +// a container's elements into an iterator. +template +OutputIterator c_move(C&& src, OutputIterator dest) { + return std::move(container_algorithm_internal::c_begin(src), + container_algorithm_internal::c_end(src), dest); +} + +// c_move_backward() +// +// Container-based version of the `std::move_backward()` function to +// move a container's elements into an iterator in reverse order. +template +BidirectionalIterator c_move_backward(C&& src, BidirectionalIterator dest) { + return std::move_backward(container_algorithm_internal::c_begin(src), + container_algorithm_internal::c_end(src), dest); +} + +// c_swap_ranges() +// +// Container-based version of the `std::swap_ranges()` function to +// swap a container's elements with another container's elements. Swaps the +// first N elements of `c1` and `c2`, where N = min(size(c1), size(c2)). +template +container_algorithm_internal::ContainerIter c_swap_ranges(C1& c1, C2& c2) { + auto first1 = container_algorithm_internal::c_begin(c1); + auto last1 = container_algorithm_internal::c_end(c1); + auto first2 = container_algorithm_internal::c_begin(c2); + auto last2 = container_algorithm_internal::c_end(c2); + + using std::swap; + for (; first1 != last1 && first2 != last2; ++first1, (void)++first2) { + swap(*first1, *first2); + } + return first2; +} + +// c_transform() +// +// Container-based version of the `std::transform()` function to +// transform a container's elements using the unary operation, storing the +// result in an iterator pointing to the last transformed element in the output +// range. +template +OutputIterator c_transform(const InputSequence& input, OutputIterator output, + UnaryOp&& unary_op) { + return std::transform(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), output, + std::forward(unary_op)); +} + +// Overload of c_transform() for performing a transformation using a binary +// predicate. Applies `binary_op` to the first N elements of `c1` and `c2`, +// where N = min(size(c1), size(c2)). +template +OutputIterator c_transform(const InputSequence1& input1, + const InputSequence2& input2, OutputIterator output, + BinaryOp&& binary_op) { + auto first1 = container_algorithm_internal::c_begin(input1); + auto last1 = container_algorithm_internal::c_end(input1); + auto first2 = container_algorithm_internal::c_begin(input2); + auto last2 = container_algorithm_internal::c_end(input2); + for (; first1 != last1 && first2 != last2; + ++first1, (void)++first2, ++output) { + *output = binary_op(*first1, *first2); + } + + return output; +} + +// c_replace() +// +// Container-based version of the `std::replace()` function to +// replace a container's elements of some value with a new value. The container +// is modified in place. +template +void c_replace(Sequence& sequence, const T& old_value, const T& new_value) { + std::replace(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), old_value, + new_value); +} + +// c_replace_if() +// +// Container-based version of the `std::replace_if()` function to +// replace a container's elements of some value with a new value based on some +// condition. The container is modified in place. +template +void c_replace_if(C& c, Pred&& pred, T&& new_value) { + std::replace_if(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred), std::forward(new_value)); +} + +// c_replace_copy() +// +// Container-based version of the `std::replace_copy()` function to +// replace a container's elements of some value with a new value and return the +// results within an iterator. +template +OutputIterator c_replace_copy(const C& c, OutputIterator result, T&& old_value, + T&& new_value) { + return std::replace_copy(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result, + std::forward(old_value), + std::forward(new_value)); +} + +// c_replace_copy_if() +// +// Container-based version of the `std::replace_copy_if()` function +// to replace a container's elements of some value with a new value based on +// some condition, and return the results within an iterator. +template +OutputIterator c_replace_copy_if(const C& c, OutputIterator result, Pred&& pred, + T&& new_value) { + return std::replace_copy_if(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result, + std::forward(pred), + std::forward(new_value)); +} + +// c_fill() +// +// Container-based version of the `std::fill()` function to fill a +// container with some value. +template +void c_fill(C& c, T&& value) { + std::fill(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), std::forward(value)); +} + +// c_fill_n() +// +// Container-based version of the `std::fill_n()` function to fill +// the first N elements in a container with some value. +template +void c_fill_n(C& c, Size n, T&& value) { + std::fill_n(container_algorithm_internal::c_begin(c), n, + std::forward(value)); +} + +// c_generate() +// +// Container-based version of the `std::generate()` function to +// assign a container's elements to the values provided by the given generator. +template +void c_generate(C& c, Generator&& gen) { + std::generate(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(gen)); +} + +// c_generate_n() +// +// Container-based version of the `std::generate_n()` function to +// assign a container's first N elements to the values provided by the given +// generator. +template +container_algorithm_internal::ContainerIter c_generate_n(C& c, Size n, + Generator&& gen) { + return std::generate_n(container_algorithm_internal::c_begin(c), n, + std::forward(gen)); +} + +// Note: `c_xx()` container versions for `remove()`, `remove_if()`, +// and `unique()` are omitted, because it's not clear whether or not such +// functions should call erase on their supplied sequences afterwards. Either +// behavior would be surprising for a different set of users. + +// c_remove_copy() +// +// Container-based version of the `std::remove_copy()` function to +// copy a container's elements while removing any elements matching the given +// `value`. +template +OutputIterator c_remove_copy(const C& c, OutputIterator result, T&& value) { + return std::remove_copy(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result, + std::forward(value)); +} + +// c_remove_copy_if() +// +// Container-based version of the `std::remove_copy_if()` function +// to copy a container's elements while removing any elements matching the given +// condition. +template +OutputIterator c_remove_copy_if(const C& c, OutputIterator result, + Pred&& pred) { + return std::remove_copy_if(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result, + std::forward(pred)); +} + +// c_unique_copy() +// +// Container-based version of the `std::unique_copy()` function to +// copy a container's elements while removing any elements containing duplicate +// values. +template +OutputIterator c_unique_copy(const C& c, OutputIterator result) { + return std::unique_copy(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result); +} + +// Overload of c_unique_copy() for using a predicate evaluation other than +// `==` for comparing uniqueness of the element values. +template +OutputIterator c_unique_copy(const C& c, OutputIterator result, + BinaryPredicate&& pred) { + return std::unique_copy(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), result, + std::forward(pred)); +} + +// c_reverse() +// +// Container-based version of the `std::reverse()` function to +// reverse a container's elements. +template +void c_reverse(Sequence& sequence) { + std::reverse(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// c_reverse_copy() +// +// Container-based version of the `std::reverse()` function to +// reverse a container's elements and write them to an iterator range. +template +OutputIterator c_reverse_copy(const C& sequence, OutputIterator result) { + return std::reverse_copy(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + result); +} + +// c_rotate() +// +// Container-based version of the `std::rotate()` function to +// shift a container's elements leftward such that the `middle` element becomes +// the first element in the container. +template > +Iterator c_rotate(C& sequence, Iterator middle) { + return absl::rotate(container_algorithm_internal::c_begin(sequence), middle, + container_algorithm_internal::c_end(sequence)); +} + +// c_rotate_copy() +// +// Container-based version of the `std::rotate_copy()` function to +// shift a container's elements leftward such that the `middle` element becomes +// the first element in a new iterator range. +template +OutputIterator c_rotate_copy( + const C& sequence, + container_algorithm_internal::ContainerIter middle, + OutputIterator result) { + return std::rotate_copy(container_algorithm_internal::c_begin(sequence), + middle, container_algorithm_internal::c_end(sequence), + result); +} + +// c_shuffle() +// +// Container-based version of the `std::shuffle()` function to +// randomly shuffle elements within the container using a `gen()` uniform random +// number generator. +template +void c_shuffle(RandomAccessContainer& c, UniformRandomBitGenerator&& gen) { + std::shuffle(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(gen)); +} + +//------------------------------------------------------------------------------ +// Partition functions +//------------------------------------------------------------------------------ + +// c_is_partitioned() +// +// Container-based version of the `std::is_partitioned()` function +// to test whether all elements in the container for which `pred` returns `true` +// precede those for which `pred` is `false`. +template +bool c_is_partitioned(const C& c, Pred&& pred) { + return std::is_partitioned(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_partition() +// +// Container-based version of the `std::partition()` function +// to rearrange all elements in a container in such a way that all elements for +// which `pred` returns `true` precede all those for which it returns `false`, +// returning an iterator to the first element of the second group. +template +container_algorithm_internal::ContainerIter c_partition(C& c, Pred&& pred) { + return std::partition(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_stable_partition() +// +// Container-based version of the `std::stable_partition()` function +// to rearrange all elements in a container in such a way that all elements for +// which `pred` returns `true` precede all those for which it returns `false`, +// preserving the relative ordering between the two groups. The function returns +// an iterator to the first element of the second group. +template +container_algorithm_internal::ContainerIter c_stable_partition(C& c, + Pred&& pred) { + return std::stable_partition(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +// c_partition_copy() +// +// Container-based version of the `std::partition_copy()` function +// to partition a container's elements and return them into two iterators: one +// for which `pred` returns `true`, and one for which `pred` returns `false.` + +template +std::pair c_partition_copy( + const C& c, OutputIterator1 out_true, OutputIterator2 out_false, + Pred&& pred) { + return std::partition_copy(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), out_true, + out_false, std::forward(pred)); +} + +// c_partition_point() +// +// Container-based version of the `std::partition_point()` function +// to return the first element of an already partitioned container for which +// the given `pred` is not `true`. +template +container_algorithm_internal::ContainerIter c_partition_point(C& c, + Pred&& pred) { + return std::partition_point(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(pred)); +} + +//------------------------------------------------------------------------------ +// Sorting functions +//------------------------------------------------------------------------------ + +// c_sort() +// +// Container-based version of the `std::sort()` function +// to sort elements in ascending order of their values. +template +void c_sort(C& c) { + std::sort(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_sort() for performing a `comp` comparison other than the +// default `operator<`. +template +void c_sort(C& c, LessThan&& comp) { + std::sort(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_stable_sort() +// +// Container-based version of the `std::stable_sort()` function +// to sort elements in ascending order of their values, preserving the order +// of equivalents. +template +void c_stable_sort(C& c) { + std::stable_sort(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_stable_sort() for performing a `comp` comparison other than the +// default `operator<`. +template +void c_stable_sort(C& c, LessThan&& comp) { + std::stable_sort(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_is_sorted() +// +// Container-based version of the `std::is_sorted()` function +// to evaluate whether the given container is sorted in ascending order. +template +bool c_is_sorted(const C& c) { + return std::is_sorted(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// c_is_sorted() overload for performing a `comp` comparison other than the +// default `operator<`. +template +bool c_is_sorted(const C& c, LessThan&& comp) { + return std::is_sorted(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_partial_sort() +// +// Container-based version of the `std::partial_sort()` function +// to rearrange elements within a container such that elements before `middle` +// are sorted in ascending order. +template +void c_partial_sort( + RandomAccessContainer& sequence, + container_algorithm_internal::ContainerIter middle) { + std::partial_sort(container_algorithm_internal::c_begin(sequence), middle, + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_partial_sort() for performing a `comp` comparison other than +// the default `operator<`. +template +void c_partial_sort( + RandomAccessContainer& sequence, + container_algorithm_internal::ContainerIter middle, + LessThan&& comp) { + std::partial_sort(container_algorithm_internal::c_begin(sequence), middle, + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_partial_sort_copy() +// +// Container-based version of the `std::partial_sort_copy()` +// function to sort the elements in the given range `result` within the larger +// `sequence` in ascending order (and using `result` as the output parameter). +// At most min(result.last - result.first, sequence.last - sequence.first) +// elements from the sequence will be stored in the result. +template +container_algorithm_internal::ContainerIter +c_partial_sort_copy(const C& sequence, RandomAccessContainer& result) { + return std::partial_sort_copy(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(result), + container_algorithm_internal::c_end(result)); +} + +// Overload of c_partial_sort_copy() for performing a `comp` comparison other +// than the default `operator<`. +template +container_algorithm_internal::ContainerIter +c_partial_sort_copy(const C& sequence, RandomAccessContainer& result, + LessThan&& comp) { + return std::partial_sort_copy(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + container_algorithm_internal::c_begin(result), + container_algorithm_internal::c_end(result), + std::forward(comp)); +} + +// c_is_sorted_until() +// +// Container-based version of the `std::is_sorted_until()` function +// to return the first element within a container that is not sorted in +// ascending order as an iterator. +template +container_algorithm_internal::ContainerIter c_is_sorted_until(C& c) { + return std::is_sorted_until(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_is_sorted_until() for performing a `comp` comparison other than +// the default `operator<`. +template +container_algorithm_internal::ContainerIter c_is_sorted_until( + C& c, LessThan&& comp) { + return std::is_sorted_until(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_nth_element() +// +// Container-based version of the `std::nth_element()` function +// to rearrange the elements within a container such that the `nth` element +// would be in that position in an ordered sequence; other elements may be in +// any order, except that all preceding `nth` will be less than that element, +// and all following `nth` will be greater than that element. +template +void c_nth_element( + RandomAccessContainer& sequence, + container_algorithm_internal::ContainerIter nth) { + std::nth_element(container_algorithm_internal::c_begin(sequence), nth, + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_nth_element() for performing a `comp` comparison other than +// the default `operator<`. +template +void c_nth_element( + RandomAccessContainer& sequence, + container_algorithm_internal::ContainerIter nth, + LessThan&& comp) { + std::nth_element(container_algorithm_internal::c_begin(sequence), nth, + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// Binary Search +//------------------------------------------------------------------------------ + +// c_lower_bound() +// +// Container-based version of the `std::lower_bound()` function +// to return an iterator pointing to the first element in a sorted container +// which does not compare less than `value`. +template +container_algorithm_internal::ContainerIter c_lower_bound( + Sequence& sequence, T&& value) { + return std::lower_bound(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value)); +} + +// Overload of c_lower_bound() for performing a `comp` comparison other than +// the default `operator<`. +template +container_algorithm_internal::ContainerIter c_lower_bound( + Sequence& sequence, T&& value, LessThan&& comp) { + return std::lower_bound(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value), std::forward(comp)); +} + +// c_upper_bound() +// +// Container-based version of the `std::upper_bound()` function +// to return an iterator pointing to the first element in a sorted container +// which is greater than `value`. +template +container_algorithm_internal::ContainerIter c_upper_bound( + Sequence& sequence, T&& value) { + return std::upper_bound(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value)); +} + +// Overload of c_upper_bound() for performing a `comp` comparison other than +// the default `operator<`. +template +container_algorithm_internal::ContainerIter c_upper_bound( + Sequence& sequence, T&& value, LessThan&& comp) { + return std::upper_bound(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value), std::forward(comp)); +} + +// c_equal_range() +// +// Container-based version of the `std::equal_range()` function +// to return an iterator pair pointing to the first and last elements in a +// sorted container which compare equal to `value`. +template +container_algorithm_internal::ContainerIterPairType +c_equal_range(Sequence& sequence, T&& value) { + return std::equal_range(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value)); +} + +// Overload of c_equal_range() for performing a `comp` comparison other than +// the default `operator<`. +template +container_algorithm_internal::ContainerIterPairType +c_equal_range(Sequence& sequence, T&& value, LessThan&& comp) { + return std::equal_range(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value), std::forward(comp)); +} + +// c_binary_search() +// +// Container-based version of the `std::binary_search()` function +// to test if any element in the sorted container contains a value equivalent to +// 'value'. +template +bool c_binary_search(Sequence&& sequence, T&& value) { + return std::binary_search(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value)); +} + +// Overload of c_binary_search() for performing a `comp` comparison other than +// the default `operator<`. +template +bool c_binary_search(Sequence&& sequence, T&& value, LessThan&& comp) { + return std::binary_search(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value), + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// Merge functions +//------------------------------------------------------------------------------ + +// c_merge() +// +// Container-based version of the `std::merge()` function +// to merge two sorted containers into a single sorted iterator. +template +OutputIterator c_merge(const C1& c1, const C2& c2, OutputIterator result) { + return std::merge(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), result); +} + +// Overload of c_merge() for performing a `comp` comparison other than +// the default `operator<`. +template +OutputIterator c_merge(const C1& c1, const C2& c2, OutputIterator result, + LessThan&& comp) { + return std::merge(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), result, + std::forward(comp)); +} + +// c_inplace_merge() +// +// Container-based version of the `std::inplace_merge()` function +// to merge a supplied iterator `middle` into a container. +template +void c_inplace_merge(C& c, + container_algorithm_internal::ContainerIter middle) { + std::inplace_merge(container_algorithm_internal::c_begin(c), middle, + container_algorithm_internal::c_end(c)); +} + +// Overload of c_inplace_merge() for performing a merge using a `comp` other +// than `operator<`. +template +void c_inplace_merge(C& c, + container_algorithm_internal::ContainerIter middle, + LessThan&& comp) { + std::inplace_merge(container_algorithm_internal::c_begin(c), middle, + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_includes() +// +// Container-based version of the `std::includes()` function +// to test whether a sorted container `c1` entirely contains another sorted +// container `c2`. +template +bool c_includes(const C1& c1, const C2& c2) { + return std::includes(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2)); +} + +// Overload of c_includes() for performing a merge using a `comp` other than +// `operator<`. +template +bool c_includes(const C1& c1, const C2& c2, LessThan&& comp) { + return std::includes(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), + std::forward(comp)); +} + +// c_set_union() +// +// Container-based version of the `std::set_union()` function +// to return an iterator containing the union of two containers; duplicate +// values are not copied into the output. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_union(const C1& c1, const C2& c2, OutputIterator output) { + return std::set_union(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output); +} + +// Overload of c_set_union() for performing a merge using a `comp` other than +// `operator<`. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_union(const C1& c1, const C2& c2, OutputIterator output, + LessThan&& comp) { + return std::set_union(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output, + std::forward(comp)); +} + +// c_set_intersection() +// +// Container-based version of the `std::set_intersection()` function +// to return an iterator containing the intersection of two sorted containers. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_intersection(const C1& c1, const C2& c2, + OutputIterator output) { + // In debug builds, ensure that both containers are sorted with respect to the + // default comparator. std::set_intersection requires the containers be sorted + // using operator<. + assert(absl::c_is_sorted(c1)); + assert(absl::c_is_sorted(c2)); + return std::set_intersection(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output); +} + +// Overload of c_set_intersection() for performing a merge using a `comp` other +// than `operator<`. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_intersection(const C1& c1, const C2& c2, + OutputIterator output, LessThan&& comp) { + // In debug builds, ensure that both containers are sorted with respect to the + // default comparator. std::set_intersection requires the containers be sorted + // using the same comparator. + assert(absl::c_is_sorted(c1, comp)); + assert(absl::c_is_sorted(c2, comp)); + return std::set_intersection(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output, + std::forward(comp)); +} + +// c_set_difference() +// +// Container-based version of the `std::set_difference()` function +// to return an iterator containing elements present in the first container but +// not in the second. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_difference(const C1& c1, const C2& c2, + OutputIterator output) { + return std::set_difference(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output); +} + +// Overload of c_set_difference() for performing a merge using a `comp` other +// than `operator<`. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_difference(const C1& c1, const C2& c2, + OutputIterator output, LessThan&& comp) { + return std::set_difference(container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output, + std::forward(comp)); +} + +// c_set_symmetric_difference() +// +// Container-based version of the `std::set_symmetric_difference()` +// function to return an iterator containing elements present in either one +// container or the other, but not both. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_symmetric_difference(const C1& c1, const C2& c2, + OutputIterator output) { + return std::set_symmetric_difference( + container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output); +} + +// Overload of c_set_symmetric_difference() for performing a merge using a +// `comp` other than `operator<`. +template ::value, + void>::type, + typename = typename std::enable_if< + !container_algorithm_internal::IsUnorderedContainer::value, + void>::type> +OutputIterator c_set_symmetric_difference(const C1& c1, const C2& c2, + OutputIterator output, + LessThan&& comp) { + return std::set_symmetric_difference( + container_algorithm_internal::c_begin(c1), + container_algorithm_internal::c_end(c1), + container_algorithm_internal::c_begin(c2), + container_algorithm_internal::c_end(c2), output, + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// Heap functions +//------------------------------------------------------------------------------ + +// c_push_heap() +// +// Container-based version of the `std::push_heap()` function +// to push a value onto a container heap. +template +void c_push_heap(RandomAccessContainer& sequence) { + std::push_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_push_heap() for performing a push operation on a heap using a +// `comp` other than `operator<`. +template +void c_push_heap(RandomAccessContainer& sequence, LessThan&& comp) { + std::push_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_pop_heap() +// +// Container-based version of the `std::pop_heap()` function +// to pop a value from a heap container. +template +void c_pop_heap(RandomAccessContainer& sequence) { + std::pop_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_pop_heap() for performing a pop operation on a heap using a +// `comp` other than `operator<`. +template +void c_pop_heap(RandomAccessContainer& sequence, LessThan&& comp) { + std::pop_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_make_heap() +// +// Container-based version of the `std::make_heap()` function +// to make a container a heap. +template +void c_make_heap(RandomAccessContainer& sequence) { + std::make_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_make_heap() for performing heap comparisons using a +// `comp` other than `operator<` +template +void c_make_heap(RandomAccessContainer& sequence, LessThan&& comp) { + std::make_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_sort_heap() +// +// Container-based version of the `std::sort_heap()` function +// to sort a heap into ascending order (after which it is no longer a heap). +template +void c_sort_heap(RandomAccessContainer& sequence) { + std::sort_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_sort_heap() for performing heap comparisons using a +// `comp` other than `operator<` +template +void c_sort_heap(RandomAccessContainer& sequence, LessThan&& comp) { + std::sort_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_is_heap() +// +// Container-based version of the `std::is_heap()` function +// to check whether the given container is a heap. +template +bool c_is_heap(const RandomAccessContainer& sequence) { + return std::is_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_is_heap() for performing heap comparisons using a +// `comp` other than `operator<` +template +bool c_is_heap(const RandomAccessContainer& sequence, LessThan&& comp) { + return std::is_heap(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_is_heap_until() +// +// Container-based version of the `std::is_heap_until()` function +// to find the first element in a given container which is not in heap order. +template +container_algorithm_internal::ContainerIter +c_is_heap_until(RandomAccessContainer& sequence) { + return std::is_heap_until(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_is_heap_until() for performing heap comparisons using a +// `comp` other than `operator<` +template +container_algorithm_internal::ContainerIter +c_is_heap_until(RandomAccessContainer& sequence, LessThan&& comp) { + return std::is_heap_until(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// Min/max +//------------------------------------------------------------------------------ + +// c_min_element() +// +// Container-based version of the `std::min_element()` function +// to return an iterator pointing to the element with the smallest value, using +// `operator<` to make the comparisons. +template +container_algorithm_internal::ContainerIter c_min_element( + Sequence& sequence) { + return std::min_element(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_min_element() for performing a `comp` comparison other than +// `operator<`. +template +container_algorithm_internal::ContainerIter c_min_element( + Sequence& sequence, LessThan&& comp) { + return std::min_element(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_max_element() +// +// Container-based version of the `std::max_element()` function +// to return an iterator pointing to the element with the largest value, using +// `operator<` to make the comparisons. +template +container_algorithm_internal::ContainerIter c_max_element( + Sequence& sequence) { + return std::max_element(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence)); +} + +// Overload of c_max_element() for performing a `comp` comparison other than +// `operator<`. +template +container_algorithm_internal::ContainerIter c_max_element( + Sequence& sequence, LessThan&& comp) { + return std::max_element(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(comp)); +} + +// c_minmax_element() +// +// Container-based version of the `std::minmax_element()` function +// to return a pair of iterators pointing to the elements containing the +// smallest and largest values, respectively, using `operator<` to make the +// comparisons. +template +container_algorithm_internal::ContainerIterPairType +c_minmax_element(C& c) { + return std::minmax_element(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_minmax_element() for performing `comp` comparisons other than +// `operator<`. +template +container_algorithm_internal::ContainerIterPairType +c_minmax_element(C& c, LessThan&& comp) { + return std::minmax_element(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// Lexicographical Comparisons +//------------------------------------------------------------------------------ + +// c_lexicographical_compare() +// +// Container-based version of the `std::lexicographical_compare()` +// function to lexicographically compare (e.g. sort words alphabetically) two +// container sequences. The comparison is performed using `operator<`. Note +// that capital letters ("A-Z") have ASCII values less than lowercase letters +// ("a-z"). +template +bool c_lexicographical_compare(Sequence1&& sequence1, Sequence2&& sequence2) { + return std::lexicographical_compare( + container_algorithm_internal::c_begin(sequence1), + container_algorithm_internal::c_end(sequence1), + container_algorithm_internal::c_begin(sequence2), + container_algorithm_internal::c_end(sequence2)); +} + +// Overload of c_lexicographical_compare() for performing a lexicographical +// comparison using a `comp` operator instead of `operator<`. +template +bool c_lexicographical_compare(Sequence1&& sequence1, Sequence2&& sequence2, + LessThan&& comp) { + return std::lexicographical_compare( + container_algorithm_internal::c_begin(sequence1), + container_algorithm_internal::c_end(sequence1), + container_algorithm_internal::c_begin(sequence2), + container_algorithm_internal::c_end(sequence2), + std::forward(comp)); +} + +// c_next_permutation() +// +// Container-based version of the `std::next_permutation()` function +// to rearrange a container's elements into the next lexicographically greater +// permutation. +template +bool c_next_permutation(C& c) { + return std::next_permutation(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_next_permutation() for performing a lexicographical +// comparison using a `comp` operator instead of `operator<`. +template +bool c_next_permutation(C& c, LessThan&& comp) { + return std::next_permutation(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +// c_prev_permutation() +// +// Container-based version of the `std::prev_permutation()` function +// to rearrange a container's elements into the next lexicographically lesser +// permutation. +template +bool c_prev_permutation(C& c) { + return std::prev_permutation(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c)); +} + +// Overload of c_prev_permutation() for performing a lexicographical +// comparison using a `comp` operator instead of `operator<`. +template +bool c_prev_permutation(C& c, LessThan&& comp) { + return std::prev_permutation(container_algorithm_internal::c_begin(c), + container_algorithm_internal::c_end(c), + std::forward(comp)); +} + +//------------------------------------------------------------------------------ +// algorithms +//------------------------------------------------------------------------------ + +// c_iota() +// +// Container-based version of the `std::iota()` function +// to compute successive values of `value`, as if incremented with `++value` +// after each element is written. and write them to the container. +template +void c_iota(Sequence& sequence, T&& value) { + std::iota(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(value)); +} +// c_accumulate() +// +// Container-based version of the `std::accumulate()` function +// to accumulate the element values of a container to `init` and return that +// accumulation by value. +// +// Note: Due to a language technicality this function has return type +// absl::decay_t. As a user of this function you can casually read +// this as "returns T by value" and assume it does the right thing. +template +decay_t c_accumulate(const Sequence& sequence, T&& init) { + return std::accumulate(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(init)); +} + +// Overload of c_accumulate() for using a binary operations other than +// addition for computing the accumulation. +template +decay_t c_accumulate(const Sequence& sequence, T&& init, + BinaryOp&& binary_op) { + return std::accumulate(container_algorithm_internal::c_begin(sequence), + container_algorithm_internal::c_end(sequence), + std::forward(init), + std::forward(binary_op)); +} + +// c_inner_product() +// +// Container-based version of the `std::inner_product()` function +// to compute the cumulative inner product of container element pairs. +// +// Note: Due to a language technicality this function has return type +// absl::decay_t. As a user of this function you can casually read +// this as "returns T by value" and assume it does the right thing. +template +decay_t c_inner_product(const Sequence1& factors1, const Sequence2& factors2, + T&& sum) { + return std::inner_product(container_algorithm_internal::c_begin(factors1), + container_algorithm_internal::c_end(factors1), + container_algorithm_internal::c_begin(factors2), + std::forward(sum)); +} + +// Overload of c_inner_product() for using binary operations other than +// `operator+` (for computing the accumulation) and `operator*` (for computing +// the product between the two container's element pair). +template +decay_t c_inner_product(const Sequence1& factors1, const Sequence2& factors2, + T&& sum, BinaryOp1&& op1, BinaryOp2&& op2) { + return std::inner_product(container_algorithm_internal::c_begin(factors1), + container_algorithm_internal::c_end(factors1), + container_algorithm_internal::c_begin(factors2), + std::forward(sum), std::forward(op1), + std::forward(op2)); +} + +// c_adjacent_difference() +// +// Container-based version of the `std::adjacent_difference()` +// function to compute the difference between each element and the one preceding +// it and write it to an iterator. +template +OutputIt c_adjacent_difference(const InputSequence& input, + OutputIt output_first) { + return std::adjacent_difference(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), + output_first); +} + +// Overload of c_adjacent_difference() for using a binary operation other than +// subtraction to compute the adjacent difference. +template +OutputIt c_adjacent_difference(const InputSequence& input, + OutputIt output_first, BinaryOp&& op) { + return std::adjacent_difference(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), + output_first, std::forward(op)); +} + +// c_partial_sum() +// +// Container-based version of the `std::partial_sum()` function +// to compute the partial sum of the elements in a sequence and write them +// to an iterator. The partial sum is the sum of all element values so far in +// the sequence. +template +OutputIt c_partial_sum(const InputSequence& input, OutputIt output_first) { + return std::partial_sum(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), + output_first); +} + +// Overload of c_partial_sum() for using a binary operation other than addition +// to compute the "partial sum". +template +OutputIt c_partial_sum(const InputSequence& input, OutputIt output_first, + BinaryOp&& op) { + return std::partial_sum(container_algorithm_internal::c_begin(input), + container_algorithm_internal::c_end(input), + output_first, std::forward(op)); +} + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_ALGORITHM_CONTAINER_H_ diff --git a/src/absl/base/attributes.h b/src/absl/base/attributes.h new file mode 100644 index 0000000..cf611c6 --- /dev/null +++ b/src/absl/base/attributes.h @@ -0,0 +1,764 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This header file defines macros for declaring attributes for functions, +// types, and variables. +// +// These macros are used within Abseil and allow the compiler to optimize, where +// applicable, certain function calls. +// +// Most macros here are exposing GCC or Clang features, and are stubbed out for +// other compilers. +// +// GCC attributes documentation: +// https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Function-Attributes.html +// https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Variable-Attributes.html +// https://gcc.gnu.org/onlinedocs/gcc-4.7.0/gcc/Type-Attributes.html +// +// Most attributes in this file are already supported by GCC 4.7. However, some +// of them are not supported in older version of Clang. Thus, we check +// `__has_attribute()` first. If the check fails, we check if we are on GCC and +// assume the attribute exists on GCC (which is verified on GCC 4.7). + +#ifndef ABSL_BASE_ATTRIBUTES_H_ +#define ABSL_BASE_ATTRIBUTES_H_ + +#include "absl/base/config.h" + +// ABSL_HAVE_ATTRIBUTE +// +// A function-like feature checking macro that is a wrapper around +// `__has_attribute`, which is defined by GCC 5+ and Clang and evaluates to a +// nonzero constant integer if the attribute is supported or 0 if not. +// +// It evaluates to zero if `__has_attribute` is not defined by the compiler. +// +// GCC: https://gcc.gnu.org/gcc-5/changes.html +// Clang: https://clang.llvm.org/docs/LanguageExtensions.html +#ifdef __has_attribute +#define ABSL_HAVE_ATTRIBUTE(x) __has_attribute(x) +#else +#define ABSL_HAVE_ATTRIBUTE(x) 0 +#endif + +// ABSL_HAVE_CPP_ATTRIBUTE +// +// A function-like feature checking macro that accepts C++11 style attributes. +// It's a wrapper around `__has_cpp_attribute`, defined by ISO C++ SD-6 +// (https://en.cppreference.com/w/cpp/experimental/feature_test). If we don't +// find `__has_cpp_attribute`, will evaluate to 0. +#if defined(__cplusplus) && defined(__has_cpp_attribute) +// NOTE: requiring __cplusplus above should not be necessary, but +// works around https://bugs.llvm.org/show_bug.cgi?id=23435. +#define ABSL_HAVE_CPP_ATTRIBUTE(x) __has_cpp_attribute(x) +#else +#define ABSL_HAVE_CPP_ATTRIBUTE(x) 0 +#endif + +// ----------------------------------------------------------------------------- +// Function Attributes +// ----------------------------------------------------------------------------- +// +// GCC: https://gcc.gnu.org/onlinedocs/gcc/Function-Attributes.html +// Clang: https://clang.llvm.org/docs/AttributeReference.html + +// ABSL_PRINTF_ATTRIBUTE +// ABSL_SCANF_ATTRIBUTE +// +// Tells the compiler to perform `printf` format string checking if the +// compiler supports it; see the 'format' attribute in +// . +// +// Note: As the GCC manual states, "[s]ince non-static C++ methods +// have an implicit 'this' argument, the arguments of such methods +// should be counted from two, not one." +#if ABSL_HAVE_ATTRIBUTE(format) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_PRINTF_ATTRIBUTE(string_index, first_to_check) \ + __attribute__((__format__(__printf__, string_index, first_to_check))) +#define ABSL_SCANF_ATTRIBUTE(string_index, first_to_check) \ + __attribute__((__format__(__scanf__, string_index, first_to_check))) +#else +#define ABSL_PRINTF_ATTRIBUTE(string_index, first_to_check) +#define ABSL_SCANF_ATTRIBUTE(string_index, first_to_check) +#endif + +// ABSL_ATTRIBUTE_ALWAYS_INLINE +// ABSL_ATTRIBUTE_NOINLINE +// +// Forces functions to either inline or not inline. Introduced in gcc 3.1. +#if ABSL_HAVE_ATTRIBUTE(always_inline) || \ + (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_ALWAYS_INLINE __attribute__((always_inline)) +#define ABSL_HAVE_ATTRIBUTE_ALWAYS_INLINE 1 +#else +#define ABSL_ATTRIBUTE_ALWAYS_INLINE +#endif + +#if ABSL_HAVE_ATTRIBUTE(noinline) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_NOINLINE __attribute__((noinline)) +#define ABSL_HAVE_ATTRIBUTE_NOINLINE 1 +#else +#define ABSL_ATTRIBUTE_NOINLINE +#endif + +// ABSL_ATTRIBUTE_NO_TAIL_CALL +// +// Prevents the compiler from optimizing away stack frames for functions which +// end in a call to another function. +#if ABSL_HAVE_ATTRIBUTE(disable_tail_calls) +#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 1 +#define ABSL_ATTRIBUTE_NO_TAIL_CALL __attribute__((disable_tail_calls)) +#elif defined(__GNUC__) && !defined(__clang__) && !defined(__e2k__) +#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 1 +#define ABSL_ATTRIBUTE_NO_TAIL_CALL \ + __attribute__((optimize("no-optimize-sibling-calls"))) +#else +#define ABSL_ATTRIBUTE_NO_TAIL_CALL +#define ABSL_HAVE_ATTRIBUTE_NO_TAIL_CALL 0 +#endif + +// ABSL_ATTRIBUTE_WEAK +// +// Tags a function as weak for the purposes of compilation and linking. +// Weak attributes did not work properly in LLVM's Windows backend before +// 9.0.0, so disable them there. See https://bugs.llvm.org/show_bug.cgi?id=37598 +// for further information. +// The MinGW compiler doesn't complain about the weak attribute until the link +// step, presumably because Windows doesn't use ELF binaries. +#if (ABSL_HAVE_ATTRIBUTE(weak) || \ + (defined(__GNUC__) && !defined(__clang__))) && \ + (!defined(_WIN32) || (defined(__clang__) && __clang_major__ >= 9)) && \ + !defined(__MINGW32__) +#undef ABSL_ATTRIBUTE_WEAK +#define ABSL_ATTRIBUTE_WEAK __attribute__((weak)) +#define ABSL_HAVE_ATTRIBUTE_WEAK 1 +#else +#define ABSL_ATTRIBUTE_WEAK +#define ABSL_HAVE_ATTRIBUTE_WEAK 0 +#endif + +// ABSL_ATTRIBUTE_NONNULL +// +// Tells the compiler either (a) that a particular function parameter +// should be a non-null pointer, or (b) that all pointer arguments should +// be non-null. +// +// Note: As the GCC manual states, "[s]ince non-static C++ methods +// have an implicit 'this' argument, the arguments of such methods +// should be counted from two, not one." +// +// Args are indexed starting at 1. +// +// For non-static class member functions, the implicit `this` argument +// is arg 1, and the first explicit argument is arg 2. For static class member +// functions, there is no implicit `this`, and the first explicit argument is +// arg 1. +// +// Example: +// +// /* arg_a cannot be null, but arg_b can */ +// void Function(void* arg_a, void* arg_b) ABSL_ATTRIBUTE_NONNULL(1); +// +// class C { +// /* arg_a cannot be null, but arg_b can */ +// void Method(void* arg_a, void* arg_b) ABSL_ATTRIBUTE_NONNULL(2); +// +// /* arg_a cannot be null, but arg_b can */ +// static void StaticMethod(void* arg_a, void* arg_b) +// ABSL_ATTRIBUTE_NONNULL(1); +// }; +// +// If no arguments are provided, then all pointer arguments should be non-null. +// +// /* No pointer arguments may be null. */ +// void Function(void* arg_a, void* arg_b, int arg_c) ABSL_ATTRIBUTE_NONNULL(); +// +// NOTE: The GCC nonnull attribute actually accepts a list of arguments, but +// ABSL_ATTRIBUTE_NONNULL does not. +#if ABSL_HAVE_ATTRIBUTE(nonnull) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_NONNULL(arg_index) __attribute__((nonnull(arg_index))) +#else +#define ABSL_ATTRIBUTE_NONNULL(...) +#endif + +// ABSL_ATTRIBUTE_NORETURN +// +// Tells the compiler that a given function never returns. +#if ABSL_HAVE_ATTRIBUTE(noreturn) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_NORETURN __attribute__((noreturn)) +#elif defined(_MSC_VER) +#define ABSL_ATTRIBUTE_NORETURN __declspec(noreturn) +#else +#define ABSL_ATTRIBUTE_NORETURN +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS +// +// Tells the AddressSanitizer (or other memory testing tools) to ignore a given +// function. Useful for cases when a function reads random locations on stack, +// calls _exit from a cloned subprocess, deliberately accesses buffer +// out of bounds or does other scary things with memory. +// NOTE: GCC supports AddressSanitizer(asan) since 4.8. +// https://gcc.gnu.org/gcc-4.8/changes.html +#if ABSL_HAVE_ATTRIBUTE(no_sanitize_address) +#define ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS __attribute__((no_sanitize_address)) +#elif defined(_MSC_VER) && _MSC_VER >= 1928 +// https://docs.microsoft.com/en-us/cpp/cpp/no-sanitize-address +#define ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS __declspec(no_sanitize_address) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY +// +// Tells the MemorySanitizer to relax the handling of a given function. All "Use +// of uninitialized value" warnings from such functions will be suppressed, and +// all values loaded from memory will be considered fully initialized. This +// attribute is similar to the ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS attribute +// above, but deals with initialized-ness rather than addressability issues. +// NOTE: MemorySanitizer(msan) is supported by Clang but not GCC. +#if ABSL_HAVE_ATTRIBUTE(no_sanitize_memory) +#define ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY __attribute__((no_sanitize_memory)) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_THREAD +// +// Tells the ThreadSanitizer to not instrument a given function. +// NOTE: GCC supports ThreadSanitizer(tsan) since 4.8. +// https://gcc.gnu.org/gcc-4.8/changes.html +#if ABSL_HAVE_ATTRIBUTE(no_sanitize_thread) +#define ABSL_ATTRIBUTE_NO_SANITIZE_THREAD __attribute__((no_sanitize_thread)) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_THREAD +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED +// +// Tells the UndefinedSanitizer to ignore a given function. Useful for cases +// where certain behavior (eg. division by zero) is being used intentionally. +// NOTE: GCC supports UndefinedBehaviorSanitizer(ubsan) since 4.9. +// https://gcc.gnu.org/gcc-4.9/changes.html +#if ABSL_HAVE_ATTRIBUTE(no_sanitize_undefined) +#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED \ + __attribute__((no_sanitize_undefined)) +#elif ABSL_HAVE_ATTRIBUTE(no_sanitize) +#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED \ + __attribute__((no_sanitize("undefined"))) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_UNDEFINED +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_CFI +// +// Tells the ControlFlowIntegrity sanitizer to not instrument a given function. +// See https://clang.llvm.org/docs/ControlFlowIntegrity.html for details. +#if ABSL_HAVE_ATTRIBUTE(no_sanitize) +#define ABSL_ATTRIBUTE_NO_SANITIZE_CFI __attribute__((no_sanitize("cfi"))) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_CFI +#endif + +// ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK +// +// Tells the SafeStack to not instrument a given function. +// See https://clang.llvm.org/docs/SafeStack.html for details. +#if ABSL_HAVE_ATTRIBUTE(no_sanitize) +#define ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK \ + __attribute__((no_sanitize("safe-stack"))) +#else +#define ABSL_ATTRIBUTE_NO_SANITIZE_SAFESTACK +#endif + +// ABSL_ATTRIBUTE_RETURNS_NONNULL +// +// Tells the compiler that a particular function never returns a null pointer. +#if ABSL_HAVE_ATTRIBUTE(returns_nonnull) +#define ABSL_ATTRIBUTE_RETURNS_NONNULL __attribute__((returns_nonnull)) +#else +#define ABSL_ATTRIBUTE_RETURNS_NONNULL +#endif + +// ABSL_HAVE_ATTRIBUTE_SECTION +// +// Indicates whether labeled sections are supported. Weak symbol support is +// a prerequisite. Labeled sections are not supported on Darwin/iOS. +#ifdef ABSL_HAVE_ATTRIBUTE_SECTION +#error ABSL_HAVE_ATTRIBUTE_SECTION cannot be directly set +#elif (ABSL_HAVE_ATTRIBUTE(section) || \ + (defined(__GNUC__) && !defined(__clang__))) && \ + !defined(__APPLE__) && ABSL_HAVE_ATTRIBUTE_WEAK +#define ABSL_HAVE_ATTRIBUTE_SECTION 1 + +// ABSL_ATTRIBUTE_SECTION +// +// Tells the compiler/linker to put a given function into a section and define +// `__start_ ## name` and `__stop_ ## name` symbols to bracket the section. +// This functionality is supported by GNU linker. Any function annotated with +// `ABSL_ATTRIBUTE_SECTION` must not be inlined, or it will be placed into +// whatever section its caller is placed into. +// +#ifndef ABSL_ATTRIBUTE_SECTION +#define ABSL_ATTRIBUTE_SECTION(name) \ + __attribute__((section(#name))) __attribute__((noinline)) +#endif + +// ABSL_ATTRIBUTE_SECTION_VARIABLE +// +// Tells the compiler/linker to put a given variable into a section and define +// `__start_ ## name` and `__stop_ ## name` symbols to bracket the section. +// This functionality is supported by GNU linker. +#ifndef ABSL_ATTRIBUTE_SECTION_VARIABLE +#ifdef _AIX +// __attribute__((section(#name))) on AIX is achived by using the `.csect` psudo +// op which includes an additional integer as part of its syntax indcating +// alignment. If data fall under different alignments then you might get a +// compilation error indicating a `Section type conflict`. +#define ABSL_ATTRIBUTE_SECTION_VARIABLE(name) +#else +#define ABSL_ATTRIBUTE_SECTION_VARIABLE(name) __attribute__((section(#name))) +#endif +#endif + +// ABSL_DECLARE_ATTRIBUTE_SECTION_VARS +// +// A weak section declaration to be used as a global declaration +// for ABSL_ATTRIBUTE_SECTION_START|STOP(name) to compile and link +// even without functions with ABSL_ATTRIBUTE_SECTION(name). +// ABSL_DEFINE_ATTRIBUTE_SECTION should be in the exactly one file; it's +// a no-op on ELF but not on Mach-O. +// +#ifndef ABSL_DECLARE_ATTRIBUTE_SECTION_VARS +#define ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name) \ + extern char __start_##name[] ABSL_ATTRIBUTE_WEAK; \ + extern char __stop_##name[] ABSL_ATTRIBUTE_WEAK +#endif +#ifndef ABSL_DEFINE_ATTRIBUTE_SECTION_VARS +#define ABSL_INIT_ATTRIBUTE_SECTION_VARS(name) +#define ABSL_DEFINE_ATTRIBUTE_SECTION_VARS(name) +#endif + +// ABSL_ATTRIBUTE_SECTION_START +// +// Returns `void*` pointers to start/end of a section of code with +// functions having ABSL_ATTRIBUTE_SECTION(name). +// Returns 0 if no such functions exist. +// One must ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name) for this to compile and +// link. +// +#define ABSL_ATTRIBUTE_SECTION_START(name) \ + (reinterpret_cast(__start_##name)) +#define ABSL_ATTRIBUTE_SECTION_STOP(name) \ + (reinterpret_cast(__stop_##name)) + +#else // !ABSL_HAVE_ATTRIBUTE_SECTION + +#define ABSL_HAVE_ATTRIBUTE_SECTION 0 + +// provide dummy definitions +#define ABSL_ATTRIBUTE_SECTION(name) +#define ABSL_ATTRIBUTE_SECTION_VARIABLE(name) +#define ABSL_INIT_ATTRIBUTE_SECTION_VARS(name) +#define ABSL_DEFINE_ATTRIBUTE_SECTION_VARS(name) +#define ABSL_DECLARE_ATTRIBUTE_SECTION_VARS(name) +#define ABSL_ATTRIBUTE_SECTION_START(name) (reinterpret_cast(0)) +#define ABSL_ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast(0)) + +#endif // ABSL_ATTRIBUTE_SECTION + +// ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC +// +// Support for aligning the stack on 32-bit x86. +#if ABSL_HAVE_ATTRIBUTE(force_align_arg_pointer) || \ + (defined(__GNUC__) && !defined(__clang__)) +#if defined(__i386__) +#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC \ + __attribute__((force_align_arg_pointer)) +#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0) +#elif defined(__x86_64__) +#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (1) +#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC +#else // !__i386__ && !__x86_64 +#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0) +#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC +#endif // __i386__ +#else +#define ABSL_ATTRIBUTE_STACK_ALIGN_FOR_OLD_LIBC +#define ABSL_REQUIRE_STACK_ALIGN_TRAMPOLINE (0) +#endif + +// ABSL_MUST_USE_RESULT +// +// Tells the compiler to warn about unused results. +// +// For code or headers that are assured to only build with C++17 and up, prefer +// just using the standard `[[nodiscard]]` directly over this macro. +// +// When annotating a function, it must appear as the first part of the +// declaration or definition. The compiler will warn if the return value from +// such a function is unused: +// +// ABSL_MUST_USE_RESULT Sprocket* AllocateSprocket(); +// AllocateSprocket(); // Triggers a warning. +// +// When annotating a class, it is equivalent to annotating every function which +// returns an instance. +// +// class ABSL_MUST_USE_RESULT Sprocket {}; +// Sprocket(); // Triggers a warning. +// +// Sprocket MakeSprocket(); +// MakeSprocket(); // Triggers a warning. +// +// Note that references and pointers are not instances: +// +// Sprocket* SprocketPointer(); +// SprocketPointer(); // Does *not* trigger a warning. +// +// ABSL_MUST_USE_RESULT allows using cast-to-void to suppress the unused result +// warning. For that, warn_unused_result is used only for clang but not for gcc. +// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66425 +// +// Note: past advice was to place the macro after the argument list. +// +// TODO(b/176172494): Use ABSL_HAVE_CPP_ATTRIBUTE(nodiscard) when all code is +// compliant with the stricter [[nodiscard]]. +#if defined(__clang__) && ABSL_HAVE_ATTRIBUTE(warn_unused_result) +#define ABSL_MUST_USE_RESULT __attribute__((warn_unused_result)) +#else +#define ABSL_MUST_USE_RESULT +#endif + +// ABSL_ATTRIBUTE_HOT, ABSL_ATTRIBUTE_COLD +// +// Tells GCC that a function is hot or cold. GCC can use this information to +// improve static analysis, i.e. a conditional branch to a cold function +// is likely to be not-taken. +// This annotation is used for function declarations. +// +// Example: +// +// int foo() ABSL_ATTRIBUTE_HOT; +#if ABSL_HAVE_ATTRIBUTE(hot) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_HOT __attribute__((hot)) +#else +#define ABSL_ATTRIBUTE_HOT +#endif + +#if ABSL_HAVE_ATTRIBUTE(cold) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_COLD __attribute__((cold)) +#else +#define ABSL_ATTRIBUTE_COLD +#endif + +// ABSL_XRAY_ALWAYS_INSTRUMENT, ABSL_XRAY_NEVER_INSTRUMENT, ABSL_XRAY_LOG_ARGS +// +// We define the ABSL_XRAY_ALWAYS_INSTRUMENT and ABSL_XRAY_NEVER_INSTRUMENT +// macro used as an attribute to mark functions that must always or never be +// instrumented by XRay. Currently, this is only supported in Clang/LLVM. +// +// For reference on the LLVM XRay instrumentation, see +// http://llvm.org/docs/XRay.html. +// +// A function with the XRAY_ALWAYS_INSTRUMENT macro attribute in its declaration +// will always get the XRay instrumentation sleds. These sleds may introduce +// some binary size and runtime overhead and must be used sparingly. +// +// These attributes only take effect when the following conditions are met: +// +// * The file/target is built in at least C++11 mode, with a Clang compiler +// that supports XRay attributes. +// * The file/target is built with the -fxray-instrument flag set for the +// Clang/LLVM compiler. +// * The function is defined in the translation unit (the compiler honors the +// attribute in either the definition or the declaration, and must match). +// +// There are cases when, even when building with XRay instrumentation, users +// might want to control specifically which functions are instrumented for a +// particular build using special-case lists provided to the compiler. These +// special case lists are provided to Clang via the +// -fxray-always-instrument=... and -fxray-never-instrument=... flags. The +// attributes in source take precedence over these special-case lists. +// +// To disable the XRay attributes at build-time, users may define +// ABSL_NO_XRAY_ATTRIBUTES. Do NOT define ABSL_NO_XRAY_ATTRIBUTES on specific +// packages/targets, as this may lead to conflicting definitions of functions at +// link-time. +// +// XRay isn't currently supported on Android: +// https://github.com/android/ndk/issues/368 +#if ABSL_HAVE_CPP_ATTRIBUTE(clang::xray_always_instrument) && \ + !defined(ABSL_NO_XRAY_ATTRIBUTES) && !defined(__ANDROID__) +#define ABSL_XRAY_ALWAYS_INSTRUMENT [[clang::xray_always_instrument]] +#define ABSL_XRAY_NEVER_INSTRUMENT [[clang::xray_never_instrument]] +#if ABSL_HAVE_CPP_ATTRIBUTE(clang::xray_log_args) +#define ABSL_XRAY_LOG_ARGS(N) \ + [[clang::xray_always_instrument, clang::xray_log_args(N)]] +#else +#define ABSL_XRAY_LOG_ARGS(N) [[clang::xray_always_instrument]] +#endif +#else +#define ABSL_XRAY_ALWAYS_INSTRUMENT +#define ABSL_XRAY_NEVER_INSTRUMENT +#define ABSL_XRAY_LOG_ARGS(N) +#endif + +// ABSL_ATTRIBUTE_REINITIALIZES +// +// Indicates that a member function reinitializes the entire object to a known +// state, independent of the previous state of the object. +// +// The clang-tidy check bugprone-use-after-move allows member functions marked +// with this attribute to be called on objects that have been moved from; +// without the attribute, this would result in a use-after-move warning. +#if ABSL_HAVE_CPP_ATTRIBUTE(clang::reinitializes) +#define ABSL_ATTRIBUTE_REINITIALIZES [[clang::reinitializes]] +#else +#define ABSL_ATTRIBUTE_REINITIALIZES +#endif + +// ----------------------------------------------------------------------------- +// Variable Attributes +// ----------------------------------------------------------------------------- + +// ABSL_ATTRIBUTE_UNUSED +// +// Prevents the compiler from complaining about variables that appear unused. +// +// For code or headers that are assured to only build with C++17 and up, prefer +// just using the standard '[[maybe_unused]]' directly over this macro. +// +// Due to differences in positioning requirements between the old, compiler +// specific __attribute__ syntax and the now standard [[maybe_unused]], this +// macro does not attempt to take advantage of '[[maybe_unused]]'. +#if ABSL_HAVE_ATTRIBUTE(unused) || (defined(__GNUC__) && !defined(__clang__)) +#undef ABSL_ATTRIBUTE_UNUSED +#define ABSL_ATTRIBUTE_UNUSED __attribute__((__unused__)) +#else +#define ABSL_ATTRIBUTE_UNUSED +#endif + +// ABSL_ATTRIBUTE_INITIAL_EXEC +// +// Tells the compiler to use "initial-exec" mode for a thread-local variable. +// See http://people.redhat.com/drepper/tls.pdf for the gory details. +#if ABSL_HAVE_ATTRIBUTE(tls_model) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_INITIAL_EXEC __attribute__((tls_model("initial-exec"))) +#else +#define ABSL_ATTRIBUTE_INITIAL_EXEC +#endif + +// ABSL_ATTRIBUTE_PACKED +// +// Instructs the compiler not to use natural alignment for a tagged data +// structure, but instead to reduce its alignment to 1. +// +// Therefore, DO NOT APPLY THIS ATTRIBUTE TO STRUCTS CONTAINING ATOMICS. Doing +// so can cause atomic variables to be mis-aligned and silently violate +// atomicity on x86. +// +// This attribute can either be applied to members of a structure or to a +// structure in its entirety. Applying this attribute (judiciously) to a +// structure in its entirety to optimize the memory footprint of very +// commonly-used structs is fine. Do not apply this attribute to a structure in +// its entirety if the purpose is to control the offsets of the members in the +// structure. Instead, apply this attribute only to structure members that need +// it. +// +// When applying ABSL_ATTRIBUTE_PACKED only to specific structure members the +// natural alignment of structure members not annotated is preserved. Aligned +// member accesses are faster than non-aligned member accesses even if the +// targeted microprocessor supports non-aligned accesses. +#if ABSL_HAVE_ATTRIBUTE(packed) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_PACKED __attribute__((__packed__)) +#else +#define ABSL_ATTRIBUTE_PACKED +#endif + +// ABSL_ATTRIBUTE_FUNC_ALIGN +// +// Tells the compiler to align the function start at least to certain +// alignment boundary +#if ABSL_HAVE_ATTRIBUTE(aligned) || (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_ATTRIBUTE_FUNC_ALIGN(bytes) __attribute__((aligned(bytes))) +#else +#define ABSL_ATTRIBUTE_FUNC_ALIGN(bytes) +#endif + +// ABSL_FALLTHROUGH_INTENDED +// +// Annotates implicit fall-through between switch labels, allowing a case to +// indicate intentional fallthrough and turn off warnings about any lack of a +// `break` statement. The ABSL_FALLTHROUGH_INTENDED macro should be followed by +// a semicolon and can be used in most places where `break` can, provided that +// no statements exist between it and the next switch label. +// +// Example: +// +// switch (x) { +// case 40: +// case 41: +// if (truth_is_out_there) { +// ++x; +// ABSL_FALLTHROUGH_INTENDED; // Use instead of/along with annotations +// // in comments +// } else { +// return x; +// } +// case 42: +// ... +// +// Notes: When supported, GCC and Clang can issue a warning on switch labels +// with unannotated fallthrough using the warning `-Wimplicit-fallthrough`. See +// clang documentation on language extensions for details: +// https://clang.llvm.org/docs/AttributeReference.html#fallthrough-clang-fallthrough +// +// When used with unsupported compilers, the ABSL_FALLTHROUGH_INTENDED macro has +// no effect on diagnostics. In any case this macro has no effect on runtime +// behavior and performance of code. + +// The use of ABSL_HAVE_CPP_ATTRIBUTE() here uses C++17 attributes +// even if -std=c++17 is not set, which does not work with -Wpedantic on R +// #ifdef ABSL_FALLTHROUGH_INTENDED +// #error "ABSL_FALLTHROUGH_INTENDED should not be defined." +// #elif ABSL_HAVE_CPP_ATTRIBUTE(fallthrough) +// #define ABSL_FALLTHROUGH_INTENDED [[fallthrough]] +// #elif ABSL_HAVE_CPP_ATTRIBUTE(clang::fallthrough) +// #define ABSL_FALLTHROUGH_INTENDED [[clang::fallthrough]] +// #elif ABSL_HAVE_CPP_ATTRIBUTE(gnu::fallthrough) +// #define ABSL_FALLTHROUGH_INTENDED [[gnu::fallthrough]] +// #else +#define ABSL_FALLTHROUGH_INTENDED \ + do { \ + } while (0) +// #endif + +// ABSL_DEPRECATED() +// +// Marks a deprecated class, struct, enum, function, method and variable +// declarations. The macro argument is used as a custom diagnostic message (e.g. +// suggestion of a better alternative). +// +// For code or headers that are assured to only build with C++14 and up, prefer +// just using the standard `[[deprecated("message")]]` directly over this macro. +// +// Examples: +// +// class ABSL_DEPRECATED("Use Bar instead") Foo {...}; +// +// ABSL_DEPRECATED("Use Baz() instead") void Bar() {...} +// +// template +// ABSL_DEPRECATED("Use DoThat() instead") +// void DoThis(); +// +// enum FooEnum { +// kBar ABSL_DEPRECATED("Use kBaz instead"), +// }; +// +// Every usage of a deprecated entity will trigger a warning when compiled with +// GCC/Clang's `-Wdeprecated-declarations` option. Google's production toolchain +// turns this warning off by default, instead relying on clang-tidy to report +// new uses of deprecated code. +#if ABSL_HAVE_ATTRIBUTE(deprecated) +#define ABSL_DEPRECATED(message) __attribute__((deprecated(message))) +#else +#define ABSL_DEPRECATED(message) +#endif + +// ABSL_CONST_INIT +// +// A variable declaration annotated with the `ABSL_CONST_INIT` attribute will +// not compile (on supported platforms) unless the variable has a constant +// initializer. This is useful for variables with static and thread storage +// duration, because it guarantees that they will not suffer from the so-called +// "static init order fiasco". +// +// This attribute must be placed on the initializing declaration of the +// variable. Some compilers will give a -Wmissing-constinit warning when this +// attribute is placed on some other declaration but missing from the +// initializing declaration. +// +// In some cases (notably with thread_local variables), `ABSL_CONST_INIT` can +// also be used in a non-initializing declaration to tell the compiler that a +// variable is already initialized, reducing overhead that would otherwise be +// incurred by a hidden guard variable. Thus annotating all declarations with +// this attribute is recommended to potentially enhance optimization. +// +// Example: +// +// class MyClass { +// public: +// ABSL_CONST_INIT static MyType my_var; +// }; +// +// ABSL_CONST_INIT MyType MyClass::my_var = MakeMyType(...); +// +// For code or headers that are assured to only build with C++20 and up, prefer +// just using the standard `constinit` keyword directly over this macro. +// +// Note that this attribute is redundant if the variable is declared constexpr. +#if defined(__cpp_constinit) && __cpp_constinit >= 201907L +#define ABSL_CONST_INIT constinit +#elif ABSL_HAVE_CPP_ATTRIBUTE(clang::require_constant_initialization) +#define ABSL_CONST_INIT [[clang::require_constant_initialization]] +#else +#define ABSL_CONST_INIT +#endif + +// ABSL_ATTRIBUTE_PURE_FUNCTION +// +// ABSL_ATTRIBUTE_PURE_FUNCTION is used to annotate declarations of "pure" +// functions. A function is pure if its return value is only a function of its +// arguments. The pure attribute prohibits a function from modifying the state +// of the program that is observable by means other than inspecting the +// function's return value. Declaring such functions with the pure attribute +// allows the compiler to avoid emitting some calls in repeated invocations of +// the function with the same argument values. +// +// Example: +// +// ABSL_ATTRIBUTE_PURE_FUNCTION int64_t ToInt64Milliseconds(Duration d); +#if ABSL_HAVE_CPP_ATTRIBUTE(gnu::pure) +#define ABSL_ATTRIBUTE_PURE_FUNCTION [[gnu::pure]] +#elif ABSL_HAVE_ATTRIBUTE(pure) +#define ABSL_ATTRIBUTE_PURE_FUNCTION __attribute__((pure)) +#else +#define ABSL_ATTRIBUTE_PURE_FUNCTION +#endif + +// ABSL_ATTRIBUTE_LIFETIME_BOUND indicates that a resource owned by a function +// parameter or implicit object parameter is retained by the return value of the +// annotated function (or, for a parameter of a constructor, in the value of the +// constructed object). This attribute causes warnings to be produced if a +// temporary object does not live long enough. +// +// When applied to a reference parameter, the referenced object is assumed to be +// retained by the return value of the function. When applied to a non-reference +// parameter (for example, a pointer or a class type), all temporaries +// referenced by the parameter are assumed to be retained by the return value of +// the function. +// +// See also the upstream documentation: +// https://clang.llvm.org/docs/AttributeReference.html#lifetimebound +#if ABSL_HAVE_CPP_ATTRIBUTE(clang::lifetimebound) +#define ABSL_ATTRIBUTE_LIFETIME_BOUND [[clang::lifetimebound]] +#elif ABSL_HAVE_ATTRIBUTE(lifetimebound) +#define ABSL_ATTRIBUTE_LIFETIME_BOUND __attribute__((lifetimebound)) +#else +#define ABSL_ATTRIBUTE_LIFETIME_BOUND +#endif + +#endif // ABSL_BASE_ATTRIBUTES_H_ diff --git a/src/absl/base/call_once.h b/src/absl/base/call_once.h new file mode 100644 index 0000000..96109f5 --- /dev/null +++ b/src/absl/base/call_once.h @@ -0,0 +1,219 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: call_once.h +// ----------------------------------------------------------------------------- +// +// This header file provides an Abseil version of `std::call_once` for invoking +// a given function at most once, across all threads. This Abseil version is +// faster than the C++11 version and incorporates the C++17 argument-passing +// fix, so that (for example) non-const references may be passed to the invoked +// function. + +#ifndef ABSL_BASE_CALL_ONCE_H_ +#define ABSL_BASE_CALL_ONCE_H_ + +#include +#include +#include +#include +#include + +#include "absl/base/internal/invoke.h" +#include "absl/base/internal/low_level_scheduling.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/scheduling_mode.h" +#include "absl/base/internal/spinlock_wait.h" +#include "absl/base/macros.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +class once_flag; + +namespace base_internal { +std::atomic* ControlWord(absl::once_flag* flag); +} // namespace base_internal + +// call_once() +// +// For all invocations using a given `once_flag`, invokes a given `fn` exactly +// once across all threads. The first call to `call_once()` with a particular +// `once_flag` argument (that does not throw an exception) will run the +// specified function with the provided `args`; other calls with the same +// `once_flag` argument will not run the function, but will wait +// for the provided function to finish running (if it is still running). +// +// This mechanism provides a safe, simple, and fast mechanism for one-time +// initialization in a multi-threaded process. +// +// Example: +// +// class MyInitClass { +// public: +// ... +// mutable absl::once_flag once_; +// +// MyInitClass* init() const { +// absl::call_once(once_, &MyInitClass::Init, this); +// return ptr_; +// } +// +template +void call_once(absl::once_flag& flag, Callable&& fn, Args&&... args); + +// once_flag +// +// Objects of this type are used to distinguish calls to `call_once()` and +// ensure the provided function is only invoked once across all threads. This +// type is not copyable or movable. However, it has a `constexpr` +// constructor, and is safe to use as a namespace-scoped global variable. +class once_flag { + public: + constexpr once_flag() : control_(0) {} + once_flag(const once_flag&) = delete; + once_flag& operator=(const once_flag&) = delete; + + private: + friend std::atomic* base_internal::ControlWord(once_flag* flag); + std::atomic control_; +}; + +//------------------------------------------------------------------------------ +// End of public interfaces. +// Implementation details follow. +//------------------------------------------------------------------------------ + +namespace base_internal { + +// Like call_once, but uses KERNEL_ONLY scheduling. Intended to be used to +// initialize entities used by the scheduler implementation. +template +void LowLevelCallOnce(absl::once_flag* flag, Callable&& fn, Args&&... args); + +// Disables scheduling while on stack when scheduling mode is non-cooperative. +// No effect for cooperative scheduling modes. +class SchedulingHelper { + public: + explicit SchedulingHelper(base_internal::SchedulingMode mode) : mode_(mode) { + if (mode_ == base_internal::SCHEDULE_KERNEL_ONLY) { + guard_result_ = base_internal::SchedulingGuard::DisableRescheduling(); + } + } + + ~SchedulingHelper() { + if (mode_ == base_internal::SCHEDULE_KERNEL_ONLY) { + base_internal::SchedulingGuard::EnableRescheduling(guard_result_); + } + } + + private: + base_internal::SchedulingMode mode_; + bool guard_result_; +}; + +// Bit patterns for call_once state machine values. Internal implementation +// detail, not for use by clients. +// +// The bit patterns are arbitrarily chosen from unlikely values, to aid in +// debugging. However, kOnceInit must be 0, so that a zero-initialized +// once_flag will be valid for immediate use. +enum { + kOnceInit = 0, + kOnceRunning = 0x65C2937B, + kOnceWaiter = 0x05A308D2, + // A very small constant is chosen for kOnceDone so that it fit in a single + // compare with immediate instruction for most common ISAs. This is verified + // for x86, POWER and ARM. + kOnceDone = 221, // Random Number +}; + +template +ABSL_ATTRIBUTE_NOINLINE +void CallOnceImpl(std::atomic* control, + base_internal::SchedulingMode scheduling_mode, Callable&& fn, + Args&&... args) { +#ifndef NDEBUG + { + uint32_t old_control = control->load(std::memory_order_relaxed); + if (old_control != kOnceInit && + old_control != kOnceRunning && + old_control != kOnceWaiter && + old_control != kOnceDone) { + ABSL_RAW_LOG(FATAL, "Unexpected value for control word: 0x%lx", + static_cast(old_control)); // NOLINT + } + } +#endif // NDEBUG + static const base_internal::SpinLockWaitTransition trans[] = { + {kOnceInit, kOnceRunning, true}, + {kOnceRunning, kOnceWaiter, false}, + {kOnceDone, kOnceDone, true}}; + + // Must do this before potentially modifying control word's state. + base_internal::SchedulingHelper maybe_disable_scheduling(scheduling_mode); + // Short circuit the simplest case to avoid procedure call overhead. + // The base_internal::SpinLockWait() call returns either kOnceInit or + // kOnceDone. If it returns kOnceDone, it must have loaded the control word + // with std::memory_order_acquire and seen a value of kOnceDone. + uint32_t old_control = kOnceInit; + if (control->compare_exchange_strong(old_control, kOnceRunning, + std::memory_order_relaxed) || + base_internal::SpinLockWait(control, ABSL_ARRAYSIZE(trans), trans, + scheduling_mode) == kOnceInit) { + base_internal::invoke(std::forward(fn), + std::forward(args)...); + old_control = + control->exchange(base_internal::kOnceDone, std::memory_order_release); + if (old_control == base_internal::kOnceWaiter) { + base_internal::SpinLockWake(control, true); + } + } // else *control is already kOnceDone +} + +inline std::atomic* ControlWord(once_flag* flag) { + return &flag->control_; +} + +template +void LowLevelCallOnce(absl::once_flag* flag, Callable&& fn, Args&&... args) { + std::atomic* once = base_internal::ControlWord(flag); + uint32_t s = once->load(std::memory_order_acquire); + if (ABSL_PREDICT_FALSE(s != base_internal::kOnceDone)) { + base_internal::CallOnceImpl(once, base_internal::SCHEDULE_KERNEL_ONLY, + std::forward(fn), + std::forward(args)...); + } +} + +} // namespace base_internal + +template +void call_once(absl::once_flag& flag, Callable&& fn, Args&&... args) { + std::atomic* once = base_internal::ControlWord(&flag); + uint32_t s = once->load(std::memory_order_acquire); + if (ABSL_PREDICT_FALSE(s != base_internal::kOnceDone)) { + base_internal::CallOnceImpl( + once, base_internal::SCHEDULE_COOPERATIVE_AND_KERNEL, + std::forward(fn), std::forward(args)...); + } +} + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_CALL_ONCE_H_ diff --git a/src/absl/base/casts.h b/src/absl/base/casts.h new file mode 100644 index 0000000..b99adb0 --- /dev/null +++ b/src/absl/base/casts.h @@ -0,0 +1,180 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: casts.h +// ----------------------------------------------------------------------------- +// +// This header file defines casting templates to fit use cases not covered by +// the standard casts provided in the C++ standard. As with all cast operations, +// use these with caution and only if alternatives do not exist. + +#ifndef ABSL_BASE_CASTS_H_ +#define ABSL_BASE_CASTS_H_ + +#include +#include +#include +#include + +#if defined(__cpp_lib_bit_cast) && __cpp_lib_bit_cast >= 201806L +#include // For std::bit_cast. +#endif // defined(__cpp_lib_bit_cast) && __cpp_lib_bit_cast >= 201806L + +#include "absl/base/internal/identity.h" +#include "absl/base/macros.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// implicit_cast() +// +// Performs an implicit conversion between types following the language +// rules for implicit conversion; if an implicit conversion is otherwise +// allowed by the language in the given context, this function performs such an +// implicit conversion. +// +// Example: +// +// // If the context allows implicit conversion: +// From from; +// To to = from; +// +// // Such code can be replaced by: +// implicit_cast(from); +// +// An `implicit_cast()` may also be used to annotate numeric type conversions +// that, although safe, may produce compiler warnings (such as `long` to `int`). +// Additionally, an `implicit_cast()` is also useful within return statements to +// indicate a specific implicit conversion is being undertaken. +// +// Example: +// +// return implicit_cast(size_in_bytes) / capacity_; +// +// Annotating code with `implicit_cast()` allows you to explicitly select +// particular overloads and template instantiations, while providing a safer +// cast than `reinterpret_cast()` or `static_cast()`. +// +// Additionally, an `implicit_cast()` can be used to allow upcasting within a +// type hierarchy where incorrect use of `static_cast()` could accidentally +// allow downcasting. +// +// Finally, an `implicit_cast()` can be used to perform implicit conversions +// from unrelated types that otherwise couldn't be implicitly cast directly; +// C++ will normally only implicitly cast "one step" in such conversions. +// +// That is, if C is a type which can be implicitly converted to B, with B being +// a type that can be implicitly converted to A, an `implicit_cast()` can be +// used to convert C to B (which the compiler can then implicitly convert to A +// using language rules). +// +// Example: +// +// // Assume an object C is convertible to B, which is implicitly convertible +// // to A +// A a = implicit_cast(C); +// +// Such implicit cast chaining may be useful within template logic. +template +constexpr To implicit_cast(typename absl::internal::identity_t to) { + return to; +} + +// bit_cast() +// +// Creates a value of the new type `Dest` whose representation is the same as +// that of the argument, which is of (deduced) type `Source` (a "bitwise cast"; +// every bit in the value representation of the result is equal to the +// corresponding bit in the object representation of the source). Source and +// destination types must be of the same size, and both types must be trivially +// copyable. +// +// As with most casts, use with caution. A `bit_cast()` might be needed when you +// need to treat a value as the value of some other type, for example, to access +// the individual bits of an object which are not normally accessible through +// the object's type, such as for working with the binary representation of a +// floating point value: +// +// float f = 3.14159265358979; +// int i = bit_cast(f); +// // i = 0x40490fdb +// +// Reinterpreting and accessing a value directly as a different type (as shown +// below) usually results in undefined behavior. +// +// Example: +// +// // WRONG +// float f = 3.14159265358979; +// int i = reinterpret_cast(f); // Wrong +// int j = *reinterpret_cast(&f); // Equally wrong +// int k = *bit_cast(&f); // Equally wrong +// +// Reinterpret-casting results in undefined behavior according to the ISO C++ +// specification, section [basic.lval]. Roughly, this section says: if an object +// in memory has one type, and a program accesses it with a different type, the +// result is undefined behavior for most "different type". +// +// Using bit_cast on a pointer and then dereferencing it is no better than using +// reinterpret_cast. You should only use bit_cast on the value itself. +// +// Such casting results in type punning: holding an object in memory of one type +// and reading its bits back using a different type. A `bit_cast()` avoids this +// issue by copying the object representation to a new value, which avoids +// introducing this undefined behavior (since the original value is never +// accessed in the wrong way). +// +// The requirements of `absl::bit_cast` are more strict than that of +// `std::bit_cast` unless compiler support is available. Specifically, without +// compiler support, this implementation also requires `Dest` to be +// default-constructible. In C++20, `absl::bit_cast` is replaced by +// `std::bit_cast`. +#if defined(__cpp_lib_bit_cast) && __cpp_lib_bit_cast >= 201806L + +using std::bit_cast; + +#else // defined(__cpp_lib_bit_cast) && __cpp_lib_bit_cast >= 201806L + +template ::value && + type_traits_internal::is_trivially_copyable::value +#if !ABSL_HAVE_BUILTIN(__builtin_bit_cast) + && std::is_default_constructible::value +#endif // !ABSL_HAVE_BUILTIN(__builtin_bit_cast) + , + int>::type = 0> +#if ABSL_HAVE_BUILTIN(__builtin_bit_cast) +inline constexpr Dest bit_cast(const Source& source) { + return __builtin_bit_cast(Dest, source); +} +#else // ABSL_HAVE_BUILTIN(__builtin_bit_cast) +inline Dest bit_cast(const Source& source) { + Dest dest; + memcpy(static_cast(std::addressof(dest)), + static_cast(std::addressof(source)), sizeof(dest)); + return dest; +} +#endif // ABSL_HAVE_BUILTIN(__builtin_bit_cast) + +#endif // defined(__cpp_lib_bit_cast) && __cpp_lib_bit_cast >= 201806L + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_CASTS_H_ diff --git a/src/absl/base/config.h b/src/absl/base/config.h new file mode 100644 index 0000000..eaf029c --- /dev/null +++ b/src/absl/base/config.h @@ -0,0 +1,915 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: config.h +// ----------------------------------------------------------------------------- +// +// This header file defines a set of macros for checking the presence of +// important compiler and platform features. Such macros can be used to +// produce portable code by parameterizing compilation based on the presence or +// lack of a given feature. +// +// We define a "feature" as some interface we wish to program to: for example, +// a library function or system call. A value of `1` indicates support for +// that feature; any other value indicates the feature support is undefined. +// +// Example: +// +// Suppose a programmer wants to write a program that uses the 'mmap()' system +// call. The Abseil macro for that feature (`ABSL_HAVE_MMAP`) allows you to +// selectively include the `mmap.h` header and bracket code using that feature +// in the macro: +// +// #include "absl/base/config.h" +// +// #ifdef ABSL_HAVE_MMAP +// #include "sys/mman.h" +// #endif //ABSL_HAVE_MMAP +// +// ... +// #ifdef ABSL_HAVE_MMAP +// void *ptr = mmap(...); +// ... +// #endif // ABSL_HAVE_MMAP + +#ifndef ABSL_BASE_CONFIG_H_ +#define ABSL_BASE_CONFIG_H_ + +// Included for the __GLIBC__ macro (or similar macros on other systems). +#include + +#ifdef __cplusplus +// Included for __GLIBCXX__, _LIBCPP_VERSION +#include +#endif // __cplusplus + +// ABSL_INTERNAL_CPLUSPLUS_LANG +// +// MSVC does not set the value of __cplusplus correctly, but instead uses +// _MSVC_LANG as a stand-in. +// https://docs.microsoft.com/en-us/cpp/preprocessor/predefined-macros +// +// However, there are reports that MSVC even sets _MSVC_LANG incorrectly at +// times, for example: +// https://github.com/microsoft/vscode-cpptools/issues/1770 +// https://reviews.llvm.org/D70996 +// +// For this reason, this symbol is considered INTERNAL and code outside of +// Abseil must not use it. +#if defined(_MSVC_LANG) +#define ABSL_INTERNAL_CPLUSPLUS_LANG _MSVC_LANG +#elif defined(__cplusplus) +#define ABSL_INTERNAL_CPLUSPLUS_LANG __cplusplus +#endif + +#if defined(__APPLE__) +// Included for TARGET_OS_IPHONE, __IPHONE_OS_VERSION_MIN_REQUIRED, +// __IPHONE_8_0. +#include +#include +#endif + +#include "absl/base/options.h" +#include "absl/base/policy_checks.h" + +// Abseil long-term support (LTS) releases will define +// `ABSL_LTS_RELEASE_VERSION` to the integer representing the date string of the +// LTS release version, and will define `ABSL_LTS_RELEASE_PATCH_LEVEL` to the +// integer representing the patch-level for that release. +// +// For example, for LTS release version "20300401.2", this would give us +// ABSL_LTS_RELEASE_VERSION == 20300401 && ABSL_LTS_RELEASE_PATCH_LEVEL == 2 +// +// These symbols will not be defined in non-LTS code. +// +// Abseil recommends that clients live-at-head. Therefore, if you are using +// these symbols to assert a minimum version requirement, we recommend you do it +// as +// +// #if defined(ABSL_LTS_RELEASE_VERSION) && ABSL_LTS_RELEASE_VERSION < 20300401 +// #error Project foo requires Abseil LTS version >= 20300401 +// #endif +// +// The `defined(ABSL_LTS_RELEASE_VERSION)` part of the check excludes +// live-at-head clients from the minimum version assertion. +// +// See https://abseil.io/about/releases for more information on Abseil release +// management. +// +// LTS releases can be obtained from +// https://github.com/abseil/abseil-cpp/releases. +#define ABSL_LTS_RELEASE_VERSION 20220623 +#define ABSL_LTS_RELEASE_PATCH_LEVEL 1 + +// Helper macro to convert a CPP variable to a string literal. +#define ABSL_INTERNAL_DO_TOKEN_STR(x) #x +#define ABSL_INTERNAL_TOKEN_STR(x) ABSL_INTERNAL_DO_TOKEN_STR(x) + +// ----------------------------------------------------------------------------- +// Abseil namespace annotations +// ----------------------------------------------------------------------------- + +// ABSL_NAMESPACE_BEGIN/ABSL_NAMESPACE_END +// +// An annotation placed at the beginning/end of each `namespace absl` scope. +// This is used to inject an inline namespace. +// +// The proper way to write Abseil code in the `absl` namespace is: +// +// namespace absl { +// ABSL_NAMESPACE_BEGIN +// +// void Foo(); // absl::Foo(). +// +// ABSL_NAMESPACE_END +// } // namespace absl +// +// Users of Abseil should not use these macros, because users of Abseil should +// not write `namespace absl {` in their own code for any reason. (Abseil does +// not support forward declarations of its own types, nor does it support +// user-provided specialization of Abseil templates. Code that violates these +// rules may be broken without warning.) +#if !defined(ABSL_OPTION_USE_INLINE_NAMESPACE) || \ + !defined(ABSL_OPTION_INLINE_NAMESPACE_NAME) +#error options.h is misconfigured. +#endif + +// Check that ABSL_OPTION_INLINE_NAMESPACE_NAME is neither "head" nor "" +#if defined(__cplusplus) && ABSL_OPTION_USE_INLINE_NAMESPACE == 1 + +#define ABSL_INTERNAL_INLINE_NAMESPACE_STR \ + ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME) + +static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != '\0', + "options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must " + "not be empty."); +static_assert(ABSL_INTERNAL_INLINE_NAMESPACE_STR[0] != 'h' || + ABSL_INTERNAL_INLINE_NAMESPACE_STR[1] != 'e' || + ABSL_INTERNAL_INLINE_NAMESPACE_STR[2] != 'a' || + ABSL_INTERNAL_INLINE_NAMESPACE_STR[3] != 'd' || + ABSL_INTERNAL_INLINE_NAMESPACE_STR[4] != '\0', + "options.h misconfigured: ABSL_OPTION_INLINE_NAMESPACE_NAME must " + "be changed to a new, unique identifier name."); + +#endif + +#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0 +#define ABSL_NAMESPACE_BEGIN +#define ABSL_NAMESPACE_END +#define ABSL_INTERNAL_C_SYMBOL(x) x +#elif ABSL_OPTION_USE_INLINE_NAMESPACE == 1 +#define ABSL_NAMESPACE_BEGIN \ + inline namespace ABSL_OPTION_INLINE_NAMESPACE_NAME { +#define ABSL_NAMESPACE_END } +#define ABSL_INTERNAL_C_SYMBOL_HELPER_2(x, v) x##_##v +#define ABSL_INTERNAL_C_SYMBOL_HELPER_1(x, v) \ + ABSL_INTERNAL_C_SYMBOL_HELPER_2(x, v) +#define ABSL_INTERNAL_C_SYMBOL(x) \ + ABSL_INTERNAL_C_SYMBOL_HELPER_1(x, ABSL_OPTION_INLINE_NAMESPACE_NAME) +#else +#error options.h is misconfigured. +#endif + +// ----------------------------------------------------------------------------- +// Compiler Feature Checks +// ----------------------------------------------------------------------------- + +// ABSL_HAVE_BUILTIN() +// +// Checks whether the compiler supports a Clang Feature Checking Macro, and if +// so, checks whether it supports the provided builtin function "x" where x +// is one of the functions noted in +// https://clang.llvm.org/docs/LanguageExtensions.html +// +// Note: Use this macro to avoid an extra level of #ifdef __has_builtin check. +// http://releases.llvm.org/3.3/tools/clang/docs/LanguageExtensions.html +#ifdef __has_builtin +#define ABSL_HAVE_BUILTIN(x) __has_builtin(x) +#else +#define ABSL_HAVE_BUILTIN(x) 0 +#endif + +#ifdef __has_feature +#define ABSL_HAVE_FEATURE(f) __has_feature(f) +#else +#define ABSL_HAVE_FEATURE(f) 0 +#endif + +// Portable check for GCC minimum version: +// https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html +#if defined(__GNUC__) && defined(__GNUC_MINOR__) +#define ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(x, y) \ + (__GNUC__ > (x) || __GNUC__ == (x) && __GNUC_MINOR__ >= (y)) +#else +#define ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(x, y) 0 +#endif + +#if defined(__clang__) && defined(__clang_major__) && defined(__clang_minor__) +#define ABSL_INTERNAL_HAVE_MIN_CLANG_VERSION(x, y) \ + (__clang_major__ > (x) || __clang_major__ == (x) && __clang_minor__ >= (y)) +#else +#define ABSL_INTERNAL_HAVE_MIN_CLANG_VERSION(x, y) 0 +#endif + +// ABSL_HAVE_TLS is defined to 1 when __thread should be supported. +// We assume __thread is supported on Linux or Asylo when compiled with Clang or +// compiled against libstdc++ with _GLIBCXX_HAVE_TLS defined. +#ifdef ABSL_HAVE_TLS +#error ABSL_HAVE_TLS cannot be directly set +#elif (defined(__linux__) || defined(__ASYLO__)) && \ + (defined(__clang__) || defined(_GLIBCXX_HAVE_TLS)) +#define ABSL_HAVE_TLS 1 +#endif + +// ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE +// +// Checks whether `std::is_trivially_destructible` is supported. +// +// Notes: All supported compilers using libc++ support this feature, as does +// gcc >= 4.8.1 using libstdc++, and Visual Studio. +#ifdef ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE +#error ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE cannot be directly set +#elif defined(_LIBCPP_VERSION) || defined(_MSC_VER) || \ + (!defined(__clang__) && defined(__GLIBCXX__) && \ + ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(4, 8)) +#define ABSL_HAVE_STD_IS_TRIVIALLY_DESTRUCTIBLE 1 +#endif + +// ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE +// +// Checks whether `std::is_trivially_default_constructible` and +// `std::is_trivially_copy_constructible` are supported. + +// ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE +// +// Checks whether `std::is_trivially_copy_assignable` is supported. + +// Notes: Clang with libc++ supports these features, as does gcc >= 7.4 with +// libstdc++, or gcc >= 8.2 with libc++, and Visual Studio (but not NVCC). +#if defined(ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE) +#error ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE cannot be directly set +#elif defined(ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE) +#error ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE cannot directly set +#elif (defined(__clang__) && defined(_LIBCPP_VERSION)) || \ + (!defined(__clang__) && \ + ((ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(7, 4) && defined(__GLIBCXX__)) || \ + (ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(8, 2) && \ + defined(_LIBCPP_VERSION)))) || \ + (defined(_MSC_VER) && !defined(__NVCC__)) +#define ABSL_HAVE_STD_IS_TRIVIALLY_CONSTRUCTIBLE 1 +#define ABSL_HAVE_STD_IS_TRIVIALLY_ASSIGNABLE 1 +#endif + +// ABSL_HAVE_THREAD_LOCAL +// +// Checks whether C++11's `thread_local` storage duration specifier is +// supported. +#ifdef ABSL_HAVE_THREAD_LOCAL +#error ABSL_HAVE_THREAD_LOCAL cannot be directly set +#elif defined(__APPLE__) +// Notes: +// * Xcode's clang did not support `thread_local` until version 8, and +// even then not for all iOS < 9.0. +// * Xcode 9.3 started disallowing `thread_local` for 32-bit iOS simulator +// targeting iOS 9.x. +// * Xcode 10 moves the deployment target check for iOS < 9.0 to link time +// making ABSL_HAVE_FEATURE unreliable there. +// +#if ABSL_HAVE_FEATURE(cxx_thread_local) && \ + !(TARGET_OS_IPHONE && __IPHONE_OS_VERSION_MIN_REQUIRED < __IPHONE_9_0) +#define ABSL_HAVE_THREAD_LOCAL 1 +#endif +#else // !defined(__APPLE__) +#define ABSL_HAVE_THREAD_LOCAL 1 +#endif + +// There are platforms for which TLS should not be used even though the compiler +// makes it seem like it's supported (Android NDK < r12b for example). +// This is primarily because of linker problems and toolchain misconfiguration: +// Abseil does not intend to support this indefinitely. Currently, the newest +// toolchain that we intend to support that requires this behavior is the +// r11 NDK - allowing for a 5 year support window on that means this option +// is likely to be removed around June of 2021. +// TLS isn't supported until NDK r12b per +// https://developer.android.com/ndk/downloads/revision_history.html +// Since NDK r16, `__NDK_MAJOR__` and `__NDK_MINOR__` are defined in +// . For NDK < r16, users should define these macros, +// e.g. `-D__NDK_MAJOR__=11 -D__NKD_MINOR__=0` for NDK r11. +#if defined(__ANDROID__) && defined(__clang__) +#if __has_include() +#include +#endif // __has_include() +#if defined(__ANDROID__) && defined(__clang__) && defined(__NDK_MAJOR__) && \ + defined(__NDK_MINOR__) && \ + ((__NDK_MAJOR__ < 12) || ((__NDK_MAJOR__ == 12) && (__NDK_MINOR__ < 1))) +#undef ABSL_HAVE_TLS +#undef ABSL_HAVE_THREAD_LOCAL +#endif +#endif // defined(__ANDROID__) && defined(__clang__) + +// ABSL_HAVE_INTRINSIC_INT128 +// +// Checks whether the __int128 compiler extension for a 128-bit integral type is +// supported. +// +// Note: __SIZEOF_INT128__ is defined by Clang and GCC when __int128 is +// supported, but we avoid using it in certain cases: +// * On Clang: +// * Building using Clang for Windows, where the Clang runtime library has +// 128-bit support only on LP64 architectures, but Windows is LLP64. +// * On Nvidia's nvcc: +// * nvcc also defines __GNUC__ and __SIZEOF_INT128__, but not all versions +// actually support __int128. +// R CMD check on CRAN uses -Wpedantic, which forces us not to use the __int128 +// intrinsic as implemented here. +// #ifdef ABSL_HAVE_INTRINSIC_INT128 +// #error ABSL_HAVE_INTRINSIC_INT128 cannot be directly set +// #elif defined(__SIZEOF_INT128__) +// #if (defined(__clang__) && !defined(_WIN32)) || \ +// (defined(__CUDACC__) && __CUDACC_VER_MAJOR__ >= 9) || \ +// (defined(__GNUC__) && !defined(__clang__) && !defined(__CUDACC__)) +// #define ABSL_HAVE_INTRINSIC_INT128 1 +// #elif defined(__CUDACC__) +// // __CUDACC_VER__ is a full version number before CUDA 9, and is defined to a +// // string explaining that it has been removed starting with CUDA 9. We use +// // nested #ifs because there is no short-circuiting in the preprocessor. +// // NOTE: `__CUDACC__` could be undefined while `__CUDACC_VER__` is defined. +// #if __CUDACC_VER__ >= 70000 +// #define ABSL_HAVE_INTRINSIC_INT128 1 +// #endif // __CUDACC_VER__ >= 70000 +// #endif // defined(__CUDACC__) +// #endif // ABSL_HAVE_INTRINSIC_INT128 + +// ABSL_HAVE_EXCEPTIONS +// +// Checks whether the compiler both supports and enables exceptions. Many +// compilers support a "no exceptions" mode that disables exceptions. +// +// Generally, when ABSL_HAVE_EXCEPTIONS is not defined: +// +// * Code using `throw` and `try` may not compile. +// * The `noexcept` specifier will still compile and behave as normal. +// * The `noexcept` operator may still return `false`. +// +// For further details, consult the compiler's documentation. +#ifdef ABSL_HAVE_EXCEPTIONS +#error ABSL_HAVE_EXCEPTIONS cannot be directly set. +#elif ABSL_INTERNAL_HAVE_MIN_CLANG_VERSION(3, 6) +// Clang >= 3.6 +#if ABSL_HAVE_FEATURE(cxx_exceptions) +#define ABSL_HAVE_EXCEPTIONS 1 +#endif // ABSL_HAVE_FEATURE(cxx_exceptions) +#elif defined(__clang__) +// Clang < 3.6 +// http://releases.llvm.org/3.6.0/tools/clang/docs/ReleaseNotes.html#the-exceptions-macro +#if defined(__EXCEPTIONS) && ABSL_HAVE_FEATURE(cxx_exceptions) +#define ABSL_HAVE_EXCEPTIONS 1 +#endif // defined(__EXCEPTIONS) && ABSL_HAVE_FEATURE(cxx_exceptions) +// Handle remaining special cases and default to exceptions being supported. +#elif !(defined(__GNUC__) && (__GNUC__ < 5) && !defined(__EXCEPTIONS)) && \ + !(ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(5, 0) && \ + !defined(__cpp_exceptions)) && \ + !(defined(_MSC_VER) && !defined(_CPPUNWIND)) +#define ABSL_HAVE_EXCEPTIONS 1 +#endif + +// ----------------------------------------------------------------------------- +// Platform Feature Checks +// ----------------------------------------------------------------------------- + +// Currently supported operating systems and associated preprocessor +// symbols: +// +// Linux and Linux-derived __linux__ +// Android __ANDROID__ (implies __linux__) +// Linux (non-Android) __linux__ && !__ANDROID__ +// Darwin (macOS and iOS) __APPLE__ +// Akaros (http://akaros.org) __ros__ +// Windows _WIN32 +// NaCL __native_client__ +// AsmJS __asmjs__ +// WebAssembly __wasm__ +// Fuchsia __Fuchsia__ +// +// Note that since Android defines both __ANDROID__ and __linux__, one +// may probe for either Linux or Android by simply testing for __linux__. + +// ABSL_HAVE_MMAP +// +// Checks whether the platform has an mmap(2) implementation as defined in +// POSIX.1-2001. +#ifdef ABSL_HAVE_MMAP +#error ABSL_HAVE_MMAP cannot be directly set +#elif defined(__linux__) || defined(__APPLE__) || defined(__FreeBSD__) || \ + defined(_AIX) || defined(__ros__) || defined(__native_client__) || \ + defined(__asmjs__) || defined(__wasm__) || defined(__Fuchsia__) || \ + defined(__sun) || defined(__ASYLO__) || defined(__myriad2__) || \ + defined(__HAIKU__) || defined(__OpenBSD__) || defined(__NetBSD__) || \ + defined(__QNX__) +#define ABSL_HAVE_MMAP 1 +#endif + +// ABSL_HAVE_PTHREAD_GETSCHEDPARAM +// +// Checks whether the platform implements the pthread_(get|set)schedparam(3) +// functions as defined in POSIX.1-2001. +#ifdef ABSL_HAVE_PTHREAD_GETSCHEDPARAM +#error ABSL_HAVE_PTHREAD_GETSCHEDPARAM cannot be directly set +#elif defined(__linux__) || defined(__APPLE__) || defined(__FreeBSD__) || \ + defined(_AIX) || defined(__ros__) || defined(__OpenBSD__) || \ + defined(__NetBSD__) +#define ABSL_HAVE_PTHREAD_GETSCHEDPARAM 1 +#endif + +// ABSL_HAVE_SCHED_GETCPU +// +// Checks whether sched_getcpu is available. +#ifdef ABSL_HAVE_SCHED_GETCPU +#error ABSL_HAVE_SCHED_GETCPU cannot be directly set +#elif defined(__linux__) +#define ABSL_HAVE_SCHED_GETCPU 1 +#endif + +// ABSL_HAVE_SCHED_YIELD +// +// Checks whether the platform implements sched_yield(2) as defined in +// POSIX.1-2001. +#ifdef ABSL_HAVE_SCHED_YIELD +#error ABSL_HAVE_SCHED_YIELD cannot be directly set +#elif defined(__linux__) || defined(__ros__) || defined(__native_client__) +#define ABSL_HAVE_SCHED_YIELD 1 +#endif + +// ABSL_HAVE_SEMAPHORE_H +// +// Checks whether the platform supports the header and sem_init(3) +// family of functions as standardized in POSIX.1-2001. +// +// Note: While Apple provides for both iOS and macOS, it is +// explicitly deprecated and will cause build failures if enabled for those +// platforms. We side-step the issue by not defining it here for Apple +// platforms. +#ifdef ABSL_HAVE_SEMAPHORE_H +#error ABSL_HAVE_SEMAPHORE_H cannot be directly set +#elif defined(__linux__) || defined(__ros__) +#define ABSL_HAVE_SEMAPHORE_H 1 +#endif + +// ABSL_HAVE_ALARM +// +// Checks whether the platform supports the header and alarm(2) +// function as standardized in POSIX.1-2001. +#ifdef ABSL_HAVE_ALARM +#error ABSL_HAVE_ALARM cannot be directly set +#elif defined(__GOOGLE_GRTE_VERSION__) +// feature tests for Google's GRTE +#define ABSL_HAVE_ALARM 1 +#elif defined(__GLIBC__) +// feature test for glibc +#define ABSL_HAVE_ALARM 1 +#elif defined(_MSC_VER) +// feature tests for Microsoft's library +#elif defined(__MINGW32__) +// mingw32 doesn't provide alarm(2): +// https://osdn.net/projects/mingw/scm/git/mingw-org-wsl/blobs/5.2-trunk/mingwrt/include/unistd.h +// mingw-w64 provides a no-op implementation: +// https://sourceforge.net/p/mingw-w64/mingw-w64/ci/master/tree/mingw-w64-crt/misc/alarm.c +#elif defined(__EMSCRIPTEN__) +// emscripten doesn't support signals +#elif defined(__Fuchsia__) +// Signals don't exist on fuchsia. +#elif defined(__native_client__) +#else +// other standard libraries +#define ABSL_HAVE_ALARM 1 +#endif + +// ABSL_IS_LITTLE_ENDIAN +// ABSL_IS_BIG_ENDIAN +// +// Checks the endianness of the platform. +// +// Notes: uses the built in endian macros provided by GCC (since 4.6) and +// Clang (since 3.2); see +// https://gcc.gnu.org/onlinedocs/cpp/Common-Predefined-Macros.html. +// Otherwise, if _WIN32, assume little endian. Otherwise, bail with an error. +#if defined(ABSL_IS_BIG_ENDIAN) +#error "ABSL_IS_BIG_ENDIAN cannot be directly set." +#endif +#if defined(ABSL_IS_LITTLE_ENDIAN) +#error "ABSL_IS_LITTLE_ENDIAN cannot be directly set." +#endif + +#if (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && \ + __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) +#define ABSL_IS_LITTLE_ENDIAN 1 +#elif defined(__BYTE_ORDER__) && defined(__ORDER_BIG_ENDIAN__) && \ + __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__ +#define ABSL_IS_BIG_ENDIAN 1 +#elif defined(_WIN32) +#define ABSL_IS_LITTLE_ENDIAN 1 +#else +#error "absl endian detection needs to be set up for your compiler" +#endif + +// macOS < 10.13 and iOS < 11 don't let you use , , or +// even though the headers exist and are publicly noted to work, because the +// libc++ shared library shipped on the system doesn't have the requisite +// exported symbols. See https://github.com/abseil/abseil-cpp/issues/207 and +// https://developer.apple.com/documentation/xcode_release_notes/xcode_10_release_notes +// +// libc++ spells out the availability requirements in the file +// llvm-project/libcxx/include/__config via the #define +// _LIBCPP_AVAILABILITY_BAD_OPTIONAL_ACCESS. +// +// Unfortunately, Apple initially mis-stated the requirements as macOS < 10.14 +// and iOS < 12 in the libc++ headers. This was corrected by +// https://github.com/llvm/llvm-project/commit/7fb40e1569dd66292b647f4501b85517e9247953 +// which subsequently made it into the XCode 12.5 release. We need to match the +// old (incorrect) conditions when built with old XCode, but can use the +// corrected earlier versions with new XCode. +#if defined(__APPLE__) && defined(_LIBCPP_VERSION) && \ + ((_LIBCPP_VERSION >= 11000 && /* XCode 12.5 or later: */ \ + ((defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 101300) || \ + (defined(__ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__ < 110000) || \ + (defined(__ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__ < 40000) || \ + (defined(__ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__ < 110000))) || \ + (_LIBCPP_VERSION < 11000 && /* Pre-XCode 12.5: */ \ + ((defined(__ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_MAC_OS_X_VERSION_MIN_REQUIRED__ < 101400) || \ + (defined(__ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_IPHONE_OS_VERSION_MIN_REQUIRED__ < 120000) || \ + (defined(__ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_WATCH_OS_VERSION_MIN_REQUIRED__ < 50000) || \ + (defined(__ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__) && \ + __ENVIRONMENT_TV_OS_VERSION_MIN_REQUIRED__ < 120000)))) +#define ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE 1 +#else +#define ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE 0 +#endif + +// ABSL_HAVE_STD_ANY +// +// Checks whether C++17 std::any is available by checking whether exists. +#ifdef ABSL_HAVE_STD_ANY +#error "ABSL_HAVE_STD_ANY cannot be directly set." +#endif + +#ifdef __has_include +#if __has_include() && defined(__cplusplus) && __cplusplus >= 201703L && \ + !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE +#define ABSL_HAVE_STD_ANY 1 +#endif +#endif + +// ABSL_HAVE_STD_OPTIONAL +// +// Checks whether C++17 std::optional is available. +#ifdef ABSL_HAVE_STD_OPTIONAL +#error "ABSL_HAVE_STD_OPTIONAL cannot be directly set." +#endif + +#ifdef __has_include +#if __has_include() && defined(__cplusplus) && \ + __cplusplus >= 201703L && !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE +#define ABSL_HAVE_STD_OPTIONAL 1 +#endif +#endif + +// ABSL_HAVE_STD_VARIANT +// +// Checks whether C++17 std::variant is available. +#ifdef ABSL_HAVE_STD_VARIANT +#error "ABSL_HAVE_STD_VARIANT cannot be directly set." +#endif + +#ifdef __has_include +#if __has_include() && defined(__cplusplus) && \ + __cplusplus >= 201703L && !ABSL_INTERNAL_APPLE_CXX17_TYPES_UNAVAILABLE +#define ABSL_HAVE_STD_VARIANT 1 +#endif +#endif + +// ABSL_HAVE_STD_STRING_VIEW +// +// Checks whether C++17 std::string_view is available. +#ifdef ABSL_HAVE_STD_STRING_VIEW +#error "ABSL_HAVE_STD_STRING_VIEW cannot be directly set." +#endif + +#ifdef __has_include +#if __has_include() && defined(__cplusplus) && \ + __cplusplus >= 201703L +#define ABSL_HAVE_STD_STRING_VIEW 1 +#endif +#endif + +// For MSVC, `__has_include` is supported in VS 2017 15.3, which is later than +// the support for , , , . So we use +// _MSC_VER to check whether we have VS 2017 RTM (when , , +// , is implemented) or higher. Also, `__cplusplus` is +// not correctly set by MSVC, so we use `_MSVC_LANG` to check the language +// version. +// TODO(zhangxy): fix tests before enabling aliasing for `std::any`. +#if defined(_MSC_VER) && _MSC_VER >= 1910 && \ + ((defined(_MSVC_LANG) && _MSVC_LANG > 201402) || \ + (defined(__cplusplus) && __cplusplus > 201402)) +// #define ABSL_HAVE_STD_ANY 1 +#define ABSL_HAVE_STD_OPTIONAL 1 +#define ABSL_HAVE_STD_VARIANT 1 +#define ABSL_HAVE_STD_STRING_VIEW 1 +#endif + +// ABSL_USES_STD_ANY +// +// Indicates whether absl::any is an alias for std::any. +#if !defined(ABSL_OPTION_USE_STD_ANY) +#error options.h is misconfigured. +#elif ABSL_OPTION_USE_STD_ANY == 0 || \ + (ABSL_OPTION_USE_STD_ANY == 2 && !defined(ABSL_HAVE_STD_ANY)) +#undef ABSL_USES_STD_ANY +#elif ABSL_OPTION_USE_STD_ANY == 1 || \ + (ABSL_OPTION_USE_STD_ANY == 2 && defined(ABSL_HAVE_STD_ANY)) +#define ABSL_USES_STD_ANY 1 +#else +#error options.h is misconfigured. +#endif + +// ABSL_USES_STD_OPTIONAL +// +// Indicates whether absl::optional is an alias for std::optional. +#if !defined(ABSL_OPTION_USE_STD_OPTIONAL) +#error options.h is misconfigured. +#elif ABSL_OPTION_USE_STD_OPTIONAL == 0 || \ + (ABSL_OPTION_USE_STD_OPTIONAL == 2 && !defined(ABSL_HAVE_STD_OPTIONAL)) +#undef ABSL_USES_STD_OPTIONAL +#elif ABSL_OPTION_USE_STD_OPTIONAL == 1 || \ + (ABSL_OPTION_USE_STD_OPTIONAL == 2 && defined(ABSL_HAVE_STD_OPTIONAL)) +#define ABSL_USES_STD_OPTIONAL 1 +#else +#error options.h is misconfigured. +#endif + +// ABSL_USES_STD_VARIANT +// +// Indicates whether absl::variant is an alias for std::variant. +#if !defined(ABSL_OPTION_USE_STD_VARIANT) +#error options.h is misconfigured. +#elif ABSL_OPTION_USE_STD_VARIANT == 0 || \ + (ABSL_OPTION_USE_STD_VARIANT == 2 && !defined(ABSL_HAVE_STD_VARIANT)) +#undef ABSL_USES_STD_VARIANT +#elif ABSL_OPTION_USE_STD_VARIANT == 1 || \ + (ABSL_OPTION_USE_STD_VARIANT == 2 && defined(ABSL_HAVE_STD_VARIANT)) +#define ABSL_USES_STD_VARIANT 1 +#else +#error options.h is misconfigured. +#endif + +// ABSL_USES_STD_STRING_VIEW +// +// Indicates whether absl::string_view is an alias for std::string_view. +#if !defined(ABSL_OPTION_USE_STD_STRING_VIEW) +#error options.h is misconfigured. +#elif ABSL_OPTION_USE_STD_STRING_VIEW == 0 || \ + (ABSL_OPTION_USE_STD_STRING_VIEW == 2 && \ + !defined(ABSL_HAVE_STD_STRING_VIEW)) +#undef ABSL_USES_STD_STRING_VIEW +#elif ABSL_OPTION_USE_STD_STRING_VIEW == 1 || \ + (ABSL_OPTION_USE_STD_STRING_VIEW == 2 && \ + defined(ABSL_HAVE_STD_STRING_VIEW)) +#define ABSL_USES_STD_STRING_VIEW 1 +#else +#error options.h is misconfigured. +#endif + +// In debug mode, MSVC 2017's std::variant throws a EXCEPTION_ACCESS_VIOLATION +// SEH exception from emplace for variant when constructing the +// struct can throw. This defeats some of variant_test and +// variant_exception_safety_test. +#if defined(_MSC_VER) && _MSC_VER >= 1700 && defined(_DEBUG) +#define ABSL_INTERNAL_MSVC_2017_DBG_MODE +#endif + +// ABSL_INTERNAL_MANGLED_NS +// ABSL_INTERNAL_MANGLED_BACKREFERENCE +// +// Internal macros for building up mangled names in our internal fork of CCTZ. +// This implementation detail is only needed and provided for the MSVC build. +// +// These macros both expand to string literals. ABSL_INTERNAL_MANGLED_NS is +// the mangled spelling of the `absl` namespace, and +// ABSL_INTERNAL_MANGLED_BACKREFERENCE is a back-reference integer representing +// the proper count to skip past the CCTZ fork namespace names. (This number +// is one larger when there is an inline namespace name to skip.) +#if defined(_MSC_VER) +#if ABSL_OPTION_USE_INLINE_NAMESPACE == 0 +#define ABSL_INTERNAL_MANGLED_NS "absl" +#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "5" +#else +#define ABSL_INTERNAL_MANGLED_NS \ + ABSL_INTERNAL_TOKEN_STR(ABSL_OPTION_INLINE_NAMESPACE_NAME) "@absl" +#define ABSL_INTERNAL_MANGLED_BACKREFERENCE "6" +#endif +#endif + +// ABSL_DLL +// +// When building Abseil as a DLL, this macro expands to `__declspec(dllexport)` +// so we can annotate symbols appropriately as being exported. When used in +// headers consuming a DLL, this macro expands to `__declspec(dllimport)` so +// that consumers know the symbol is defined inside the DLL. In all other cases, +// the macro expands to nothing. +#if defined(_MSC_VER) +#if defined(ABSL_BUILD_DLL) +#define ABSL_DLL __declspec(dllexport) +#elif defined(ABSL_CONSUME_DLL) +#define ABSL_DLL __declspec(dllimport) +#else +#define ABSL_DLL +#endif +#else +#define ABSL_DLL +#endif // defined(_MSC_VER) + +// ABSL_HAVE_MEMORY_SANITIZER +// +// MemorySanitizer (MSan) is a detector of uninitialized reads. It consists of +// a compiler instrumentation module and a run-time library. +#ifdef ABSL_HAVE_MEMORY_SANITIZER +#error "ABSL_HAVE_MEMORY_SANITIZER cannot be directly set." +#elif !defined(__native_client__) && ABSL_HAVE_FEATURE(memory_sanitizer) +#define ABSL_HAVE_MEMORY_SANITIZER 1 +#endif + +// ABSL_HAVE_THREAD_SANITIZER +// +// ThreadSanitizer (TSan) is a fast data race detector. +#ifdef ABSL_HAVE_THREAD_SANITIZER +#error "ABSL_HAVE_THREAD_SANITIZER cannot be directly set." +#elif defined(__SANITIZE_THREAD__) +#define ABSL_HAVE_THREAD_SANITIZER 1 +#elif ABSL_HAVE_FEATURE(thread_sanitizer) +#define ABSL_HAVE_THREAD_SANITIZER 1 +#endif + +// ABSL_HAVE_ADDRESS_SANITIZER +// +// AddressSanitizer (ASan) is a fast memory error detector. +#ifdef ABSL_HAVE_ADDRESS_SANITIZER +#error "ABSL_HAVE_ADDRESS_SANITIZER cannot be directly set." +#elif defined(__SANITIZE_ADDRESS__) +#define ABSL_HAVE_ADDRESS_SANITIZER 1 +#elif ABSL_HAVE_FEATURE(address_sanitizer) +#define ABSL_HAVE_ADDRESS_SANITIZER 1 +#endif + +// ABSL_HAVE_HWADDRESS_SANITIZER +// +// Hardware-Assisted AddressSanitizer (or HWASAN) is even faster than asan +// memory error detector which can use CPU features like ARM TBI, Intel LAM or +// AMD UAI. +#ifdef ABSL_HAVE_HWADDRESS_SANITIZER +#error "ABSL_HAVE_HWADDRESS_SANITIZER cannot be directly set." +#elif defined(__SANITIZE_HWADDRESS__) +#define ABSL_HAVE_HWADDRESS_SANITIZER 1 +#elif ABSL_HAVE_FEATURE(hwaddress_sanitizer) +#define ABSL_HAVE_HWADDRESS_SANITIZER 1 +#endif + +// ABSL_HAVE_LEAK_SANITIZER +// +// LeakSanitizer (or lsan) is a detector of memory leaks. +// https://clang.llvm.org/docs/LeakSanitizer.html +// https://github.com/google/sanitizers/wiki/AddressSanitizerLeakSanitizer +// +// The macro ABSL_HAVE_LEAK_SANITIZER can be used to detect at compile-time +// whether the LeakSanitizer is potentially available. However, just because the +// LeakSanitizer is available does not mean it is active. Use the +// always-available run-time interface in //absl/debugging/leak_check.h for +// interacting with LeakSanitizer. +#ifdef ABSL_HAVE_LEAK_SANITIZER +#error "ABSL_HAVE_LEAK_SANITIZER cannot be directly set." +#elif defined(LEAK_SANITIZER) +// GCC provides no method for detecting the presense of the standalone +// LeakSanitizer (-fsanitize=leak), so GCC users of -fsanitize=leak should also +// use -DLEAK_SANITIZER. +#define ABSL_HAVE_LEAK_SANITIZER 1 +// Clang standalone LeakSanitizer (-fsanitize=leak) +#elif ABSL_HAVE_FEATURE(leak_sanitizer) +#define ABSL_HAVE_LEAK_SANITIZER 1 +#elif defined(ABSL_HAVE_ADDRESS_SANITIZER) +// GCC or Clang using the LeakSanitizer integrated into AddressSanitizer. +#define ABSL_HAVE_LEAK_SANITIZER 1 +#endif + +// ABSL_HAVE_CLASS_TEMPLATE_ARGUMENT_DEDUCTION +// +// Class template argument deduction is a language feature added in C++17. +#ifdef ABSL_HAVE_CLASS_TEMPLATE_ARGUMENT_DEDUCTION +#error "ABSL_HAVE_CLASS_TEMPLATE_ARGUMENT_DEDUCTION cannot be directly set." +#elif defined(__cpp_deduction_guides) +#define ABSL_HAVE_CLASS_TEMPLATE_ARGUMENT_DEDUCTION 1 +#endif + +// ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +// +// Prior to C++17, static constexpr variables defined in classes required a +// separate definition outside of the class body, for example: +// +// class Foo { +// static constexpr int kBar = 0; +// }; +// constexpr int Foo::kBar; +// +// In C++17, these variables defined in classes are considered inline variables, +// and the extra declaration is redundant. Since some compilers warn on the +// extra declarations, ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL can be used +// conditionally ignore them: +// +// #ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +// constexpr int Foo::kBar; +// #endif +#if defined(ABSL_INTERNAL_CPLUSPLUS_LANG) && \ + ABSL_INTERNAL_CPLUSPLUS_LANG < 201703L +#define ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL 1 +#endif + +// `ABSL_INTERNAL_HAS_RTTI` determines whether abseil is being compiled with +// RTTI support. +#ifdef ABSL_INTERNAL_HAS_RTTI +#error ABSL_INTERNAL_HAS_RTTI cannot be directly set +#elif !defined(__GNUC__) || defined(__GXX_RTTI) +#define ABSL_INTERNAL_HAS_RTTI 1 +#endif // !defined(__GNUC__) || defined(__GXX_RTTI) + +// ABSL_INTERNAL_HAVE_SSE is used for compile-time detection of SSE support. +// See https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html for an overview of +// which architectures support the various x86 instruction sets. +#ifdef ABSL_INTERNAL_HAVE_SSE +#error ABSL_INTERNAL_HAVE_SSE cannot be directly set +#elif defined(__SSE__) +#define ABSL_INTERNAL_HAVE_SSE 1 +#elif defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1) +// MSVC only defines _M_IX86_FP for x86 32-bit code, and _M_IX86_FP >= 1 +// indicates that at least SSE was targeted with the /arch:SSE option. +// All x86-64 processors support SSE, so support can be assumed. +// https://docs.microsoft.com/en-us/cpp/preprocessor/predefined-macros +#define ABSL_INTERNAL_HAVE_SSE 1 +#endif + +// ABSL_INTERNAL_HAVE_SSE2 is used for compile-time detection of SSE2 support. +// See https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html for an overview of +// which architectures support the various x86 instruction sets. +#ifdef ABSL_INTERNAL_HAVE_SSE2 +#error ABSL_INTERNAL_HAVE_SSE2 cannot be directly set +#elif defined(__SSE2__) +#define ABSL_INTERNAL_HAVE_SSE2 1 +#elif defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 2) +// MSVC only defines _M_IX86_FP for x86 32-bit code, and _M_IX86_FP >= 2 +// indicates that at least SSE2 was targeted with the /arch:SSE2 option. +// All x86-64 processors support SSE2, so support can be assumed. +// https://docs.microsoft.com/en-us/cpp/preprocessor/predefined-macros +#define ABSL_INTERNAL_HAVE_SSE2 1 +#endif + +// ABSL_INTERNAL_HAVE_SSSE3 is used for compile-time detection of SSSE3 support. +// See https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html for an overview of +// which architectures support the various x86 instruction sets. +// +// MSVC does not have a mode that targets SSSE3 at compile-time. To use SSSE3 +// with MSVC requires either assuming that the code will only every run on CPUs +// that support SSSE3, otherwise __cpuid() can be used to detect support at +// runtime and fallback to a non-SSSE3 implementation when SSSE3 is unsupported +// by the CPU. +#ifdef ABSL_INTERNAL_HAVE_SSSE3 +#error ABSL_INTERNAL_HAVE_SSSE3 cannot be directly set +#elif defined(__SSSE3__) +#define ABSL_INTERNAL_HAVE_SSSE3 1 +#endif + +// ABSL_INTERNAL_HAVE_ARM_NEON is used for compile-time detection of NEON (ARM +// SIMD). +#ifdef ABSL_INTERNAL_HAVE_ARM_NEON +#error ABSL_INTERNAL_HAVE_ARM_NEON cannot be directly set +#elif defined(__ARM_NEON) +#define ABSL_INTERNAL_HAVE_ARM_NEON 1 +#endif + +#endif // ABSL_BASE_CONFIG_H_ diff --git a/src/absl/base/const_init.h b/src/absl/base/const_init.h new file mode 100644 index 0000000..16520b6 --- /dev/null +++ b/src/absl/base/const_init.h @@ -0,0 +1,76 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// kConstInit +// ----------------------------------------------------------------------------- +// +// A constructor tag used to mark an object as safe for use as a global +// variable, avoiding the usual lifetime issues that can affect globals. + +#ifndef ABSL_BASE_CONST_INIT_H_ +#define ABSL_BASE_CONST_INIT_H_ + +#include "absl/base/config.h" + +// In general, objects with static storage duration (such as global variables) +// can trigger tricky object lifetime situations. Attempting to access them +// from the constructors or destructors of other global objects can result in +// undefined behavior, unless their constructors and destructors are designed +// with this issue in mind. +// +// The normal way to deal with this issue in C++11 is to use constant +// initialization and trivial destructors. +// +// Constant initialization is guaranteed to occur before any other code +// executes. Constructors that are declared 'constexpr' are eligible for +// constant initialization. You can annotate a variable declaration with the +// ABSL_CONST_INIT macro to express this intent. For compilers that support +// it, this annotation will cause a compilation error for declarations that +// aren't subject to constant initialization (perhaps because a runtime value +// was passed as a constructor argument). +// +// On program shutdown, lifetime issues can be avoided on global objects by +// ensuring that they contain trivial destructors. A class has a trivial +// destructor unless it has a user-defined destructor, a virtual method or base +// class, or a data member or base class with a non-trivial destructor of its +// own. Objects with static storage duration and a trivial destructor are not +// cleaned up on program shutdown, and are thus safe to access from other code +// running during shutdown. +// +// For a few core Abseil classes, we make a best effort to allow for safe global +// instances, even though these classes have non-trivial destructors. These +// objects can be created with the absl::kConstInit tag. For example: +// ABSL_CONST_INIT absl::Mutex global_mutex(absl::kConstInit); +// +// The line above declares a global variable of type absl::Mutex which can be +// accessed at any point during startup or shutdown. global_mutex's destructor +// will still run, but will not invalidate the object. Note that C++ specifies +// that accessing an object after its destructor has run results in undefined +// behavior, but this pattern works on the toolchains we support. +// +// The absl::kConstInit tag should only be used to define objects with static +// or thread_local storage duration. + +namespace absl { +ABSL_NAMESPACE_BEGIN + +enum ConstInitType { + kConstInit, +}; + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_CONST_INIT_H_ diff --git a/src/absl/base/dynamic_annotations.h b/src/absl/base/dynamic_annotations.h new file mode 100644 index 0000000..3ea7c15 --- /dev/null +++ b/src/absl/base/dynamic_annotations.h @@ -0,0 +1,471 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This file defines dynamic annotations for use with dynamic analysis tool +// such as valgrind, PIN, etc. +// +// Dynamic annotation is a source code annotation that affects the generated +// code (that is, the annotation is not a comment). Each such annotation is +// attached to a particular instruction and/or to a particular object (address) +// in the program. +// +// The annotations that should be used by users are macros in all upper-case +// (e.g., ABSL_ANNOTATE_THREAD_NAME). +// +// Actual implementation of these macros may differ depending on the dynamic +// analysis tool being used. +// +// This file supports the following configurations: +// - Dynamic Annotations enabled (with static thread-safety warnings disabled). +// In this case, macros expand to functions implemented by Thread Sanitizer, +// when building with TSan. When not provided an external implementation, +// dynamic_annotations.cc provides no-op implementations. +// +// - Static Clang thread-safety warnings enabled. +// When building with a Clang compiler that supports thread-safety warnings, +// a subset of annotations can be statically-checked at compile-time. We +// expand these macros to static-inline functions that can be analyzed for +// thread-safety, but afterwards elided when building the final binary. +// +// - All annotations are disabled. +// If neither Dynamic Annotations nor Clang thread-safety warnings are +// enabled, then all annotation-macros expand to empty. + +#ifndef ABSL_BASE_DYNAMIC_ANNOTATIONS_H_ +#define ABSL_BASE_DYNAMIC_ANNOTATIONS_H_ + +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#ifdef __cplusplus +#include "absl/base/macros.h" +#endif + +// TODO(rogeeff): Remove after the backward compatibility period. +#include "absl/base/internal/dynamic_annotations.h" // IWYU pragma: export + +// ------------------------------------------------------------------------- +// Decide which features are enabled. + +#ifdef ABSL_HAVE_THREAD_SANITIZER + +#define ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_ANNOTALYSIS_ENABLED 0 +#define ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED 1 + +#else + +#define ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED 0 +#define ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED 0 +#define ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED 0 + +// Clang provides limited support for static thread-safety analysis through a +// feature called Annotalysis. We configure macro-definitions according to +// whether Annotalysis support is available. When running in opt-mode, GCC +// will issue a warning, if these attributes are compiled. Only include them +// when compiling using Clang. + +#if defined(__clang__) +#define ABSL_INTERNAL_ANNOTALYSIS_ENABLED 1 +#if !defined(SWIG) +#define ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED 1 +#endif +#else +#define ABSL_INTERNAL_ANNOTALYSIS_ENABLED 0 +#endif + +// Read/write annotations are enabled in Annotalysis mode; disabled otherwise. +#define ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED \ + ABSL_INTERNAL_ANNOTALYSIS_ENABLED + +#endif // ABSL_HAVE_THREAD_SANITIZER + +#ifdef __cplusplus +#define ABSL_INTERNAL_BEGIN_EXTERN_C extern "C" { +#define ABSL_INTERNAL_END_EXTERN_C } // extern "C" +#define ABSL_INTERNAL_GLOBAL_SCOPED(F) ::F +#define ABSL_INTERNAL_STATIC_INLINE inline +#else +#define ABSL_INTERNAL_BEGIN_EXTERN_C // empty +#define ABSL_INTERNAL_END_EXTERN_C // empty +#define ABSL_INTERNAL_GLOBAL_SCOPED(F) F +#define ABSL_INTERNAL_STATIC_INLINE static inline +#endif + +// ------------------------------------------------------------------------- +// Define race annotations. + +#if ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED == 1 +// Some of the symbols used in this section (e.g. AnnotateBenignRaceSized) are +// defined by the compiler-based santizer implementation, not by the Abseil +// library. Therefore they do not use ABSL_INTERNAL_C_SYMBOL. + +// ------------------------------------------------------------- +// Annotations that suppress errors. It is usually better to express the +// program's synchronization using the other annotations, but these can be used +// when all else fails. + +// Report that we may have a benign race at `pointer`, with size +// "sizeof(*(pointer))". `pointer` must be a non-void* pointer. Insert at the +// point where `pointer` has been allocated, preferably close to the point +// where the race happens. See also ABSL_ANNOTATE_BENIGN_RACE_STATIC. +#define ABSL_ANNOTATE_BENIGN_RACE(pointer, description) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateBenignRaceSized) \ + (__FILE__, __LINE__, pointer, sizeof(*(pointer)), description) + +// Same as ABSL_ANNOTATE_BENIGN_RACE(`address`, `description`), but applies to +// the memory range [`address`, `address`+`size`). +#define ABSL_ANNOTATE_BENIGN_RACE_SIZED(address, size, description) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateBenignRaceSized) \ + (__FILE__, __LINE__, address, size, description) + +// Enable (`enable`!=0) or disable (`enable`==0) race detection for all threads. +// This annotation could be useful if you want to skip expensive race analysis +// during some period of program execution, e.g. during initialization. +#define ABSL_ANNOTATE_ENABLE_RACE_DETECTION(enable) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateEnableRaceDetection) \ + (__FILE__, __LINE__, enable) + +// ------------------------------------------------------------- +// Annotations useful for debugging. + +// Report the current thread `name` to a race detector. +#define ABSL_ANNOTATE_THREAD_NAME(name) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateThreadName)(__FILE__, __LINE__, name) + +// ------------------------------------------------------------- +// Annotations useful when implementing locks. They are not normally needed by +// modules that merely use locks. The `lock` argument is a pointer to the lock +// object. + +// Report that a lock has been created at address `lock`. +#define ABSL_ANNOTATE_RWLOCK_CREATE(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockCreate)(__FILE__, __LINE__, lock) + +// Report that a linker initialized lock has been created at address `lock`. +#ifdef ABSL_HAVE_THREAD_SANITIZER +#define ABSL_ANNOTATE_RWLOCK_CREATE_STATIC(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockCreateStatic) \ + (__FILE__, __LINE__, lock) +#else +#define ABSL_ANNOTATE_RWLOCK_CREATE_STATIC(lock) \ + ABSL_ANNOTATE_RWLOCK_CREATE(lock) +#endif + +// Report that the lock at address `lock` is about to be destroyed. +#define ABSL_ANNOTATE_RWLOCK_DESTROY(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockDestroy)(__FILE__, __LINE__, lock) + +// Report that the lock at address `lock` has been acquired. +// `is_w`=1 for writer lock, `is_w`=0 for reader lock. +#define ABSL_ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockAcquired) \ + (__FILE__, __LINE__, lock, is_w) + +// Report that the lock at address `lock` is about to be released. +// `is_w`=1 for writer lock, `is_w`=0 for reader lock. +#define ABSL_ANNOTATE_RWLOCK_RELEASED(lock, is_w) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockReleased) \ + (__FILE__, __LINE__, lock, is_w) + +// Apply ABSL_ANNOTATE_BENIGN_RACE_SIZED to a static variable `static_var`. +#define ABSL_ANNOTATE_BENIGN_RACE_STATIC(static_var, description) \ + namespace { \ + class static_var##_annotator { \ + public: \ + static_var##_annotator() { \ + ABSL_ANNOTATE_BENIGN_RACE_SIZED(&static_var, sizeof(static_var), \ + #static_var ": " description); \ + } \ + }; \ + static static_var##_annotator the##static_var##_annotator; \ + } // namespace + +// Function prototypes of annotations provided by the compiler-based sanitizer +// implementation. +ABSL_INTERNAL_BEGIN_EXTERN_C +void AnnotateRWLockCreate(const char* file, int line, + const volatile void* lock); +void AnnotateRWLockCreateStatic(const char* file, int line, + const volatile void* lock); +void AnnotateRWLockDestroy(const char* file, int line, + const volatile void* lock); +void AnnotateRWLockAcquired(const char* file, int line, + const volatile void* lock, long is_w); // NOLINT +void AnnotateRWLockReleased(const char* file, int line, + const volatile void* lock, long is_w); // NOLINT +void AnnotateBenignRace(const char* file, int line, + const volatile void* address, const char* description); +void AnnotateBenignRaceSized(const char* file, int line, + const volatile void* address, size_t size, + const char* description); +void AnnotateThreadName(const char* file, int line, const char* name); +void AnnotateEnableRaceDetection(const char* file, int line, int enable); +ABSL_INTERNAL_END_EXTERN_C + +#else // ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED == 0 + +#define ABSL_ANNOTATE_RWLOCK_CREATE(lock) // empty +#define ABSL_ANNOTATE_RWLOCK_CREATE_STATIC(lock) // empty +#define ABSL_ANNOTATE_RWLOCK_DESTROY(lock) // empty +#define ABSL_ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) // empty +#define ABSL_ANNOTATE_RWLOCK_RELEASED(lock, is_w) // empty +#define ABSL_ANNOTATE_BENIGN_RACE(address, description) // empty +#define ABSL_ANNOTATE_BENIGN_RACE_SIZED(address, size, description) // empty +#define ABSL_ANNOTATE_THREAD_NAME(name) // empty +#define ABSL_ANNOTATE_ENABLE_RACE_DETECTION(enable) // empty +#define ABSL_ANNOTATE_BENIGN_RACE_STATIC(static_var, description) // empty + +#endif // ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED + +// ------------------------------------------------------------------------- +// Define memory annotations. + +#ifdef ABSL_HAVE_MEMORY_SANITIZER + +#include + +#define ABSL_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \ + __msan_unpoison(address, size) + +#define ABSL_ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) \ + __msan_allocated_memory(address, size) + +#else // !defined(ABSL_HAVE_MEMORY_SANITIZER) + +// TODO(rogeeff): remove this branch +#ifdef ABSL_HAVE_THREAD_SANITIZER +#define ABSL_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \ + do { \ + (void)(address); \ + (void)(size); \ + } while (0) +#define ABSL_ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) \ + do { \ + (void)(address); \ + (void)(size); \ + } while (0) +#else + +#define ABSL_ANNOTATE_MEMORY_IS_INITIALIZED(address, size) // empty +#define ABSL_ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) // empty + +#endif + +#endif // ABSL_HAVE_MEMORY_SANITIZER + +// ------------------------------------------------------------------------- +// Define IGNORE_READS_BEGIN/_END attributes. + +#if defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +#define ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE \ + __attribute((exclusive_lock_function("*"))) +#define ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE \ + __attribute((unlock_function("*"))) + +#else // !defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +#define ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE // empty +#define ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE // empty + +#endif // defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +// ------------------------------------------------------------------------- +// Define IGNORE_READS_BEGIN/_END annotations. + +#if ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED == 1 +// Some of the symbols used in this section (e.g. AnnotateIgnoreReadsBegin) are +// defined by the compiler-based implementation, not by the Abseil +// library. Therefore they do not use ABSL_INTERNAL_C_SYMBOL. + +// Request the analysis tool to ignore all reads in the current thread until +// ABSL_ANNOTATE_IGNORE_READS_END is called. Useful to ignore intentional racey +// reads, while still checking other reads and all writes. +// See also ABSL_ANNOTATE_UNPROTECTED_READ. +#define ABSL_ANNOTATE_IGNORE_READS_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreReadsBegin) \ + (__FILE__, __LINE__) + +// Stop ignoring reads. +#define ABSL_ANNOTATE_IGNORE_READS_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreReadsEnd) \ + (__FILE__, __LINE__) + +// Function prototypes of annotations provided by the compiler-based sanitizer +// implementation. +ABSL_INTERNAL_BEGIN_EXTERN_C +void AnnotateIgnoreReadsBegin(const char* file, int line) + ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE; +void AnnotateIgnoreReadsEnd(const char* file, + int line) ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE; +ABSL_INTERNAL_END_EXTERN_C + +#elif defined(ABSL_INTERNAL_ANNOTALYSIS_ENABLED) + +// When Annotalysis is enabled without Dynamic Annotations, the use of +// static-inline functions allows the annotations to be read at compile-time, +// while still letting the compiler elide the functions from the final build. +// +// TODO(delesley) -- The exclusive lock here ignores writes as well, but +// allows IGNORE_READS_AND_WRITES to work properly. + +#define ABSL_ANNOTATE_IGNORE_READS_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED( \ + ABSL_INTERNAL_C_SYMBOL(AbslInternalAnnotateIgnoreReadsBegin)) \ + () + +#define ABSL_ANNOTATE_IGNORE_READS_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED( \ + ABSL_INTERNAL_C_SYMBOL(AbslInternalAnnotateIgnoreReadsEnd)) \ + () + +ABSL_INTERNAL_STATIC_INLINE void ABSL_INTERNAL_C_SYMBOL( + AbslInternalAnnotateIgnoreReadsBegin)() + ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE {} + +ABSL_INTERNAL_STATIC_INLINE void ABSL_INTERNAL_C_SYMBOL( + AbslInternalAnnotateIgnoreReadsEnd)() + ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE {} + +#else + +#define ABSL_ANNOTATE_IGNORE_READS_BEGIN() // empty +#define ABSL_ANNOTATE_IGNORE_READS_END() // empty + +#endif + +// ------------------------------------------------------------------------- +// Define IGNORE_WRITES_BEGIN/_END annotations. + +#if ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED == 1 + +// Similar to ABSL_ANNOTATE_IGNORE_READS_BEGIN, but ignore writes instead. +#define ABSL_ANNOTATE_IGNORE_WRITES_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreWritesBegin)(__FILE__, __LINE__) + +// Stop ignoring writes. +#define ABSL_ANNOTATE_IGNORE_WRITES_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreWritesEnd)(__FILE__, __LINE__) + +// Function prototypes of annotations provided by the compiler-based sanitizer +// implementation. +ABSL_INTERNAL_BEGIN_EXTERN_C +void AnnotateIgnoreWritesBegin(const char* file, int line); +void AnnotateIgnoreWritesEnd(const char* file, int line); +ABSL_INTERNAL_END_EXTERN_C + +#else + +#define ABSL_ANNOTATE_IGNORE_WRITES_BEGIN() // empty +#define ABSL_ANNOTATE_IGNORE_WRITES_END() // empty + +#endif + +// ------------------------------------------------------------------------- +// Define the ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_* annotations using the more +// primitive annotations defined above. +// +// Instead of doing +// ABSL_ANNOTATE_IGNORE_READS_BEGIN(); +// ... = x; +// ABSL_ANNOTATE_IGNORE_READS_END(); +// one can use +// ... = ABSL_ANNOTATE_UNPROTECTED_READ(x); + +#if defined(ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED) + +// Start ignoring all memory accesses (both reads and writes). +#define ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() \ + do { \ + ABSL_ANNOTATE_IGNORE_READS_BEGIN(); \ + ABSL_ANNOTATE_IGNORE_WRITES_BEGIN(); \ + } while (0) + +// Stop ignoring both reads and writes. +#define ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_END() \ + do { \ + ABSL_ANNOTATE_IGNORE_WRITES_END(); \ + ABSL_ANNOTATE_IGNORE_READS_END(); \ + } while (0) + +#ifdef __cplusplus +// ABSL_ANNOTATE_UNPROTECTED_READ is the preferred way to annotate racey reads. +#define ABSL_ANNOTATE_UNPROTECTED_READ(x) \ + absl::base_internal::AnnotateUnprotectedRead(x) + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +template +inline T AnnotateUnprotectedRead(const volatile T& x) { // NOLINT + ABSL_ANNOTATE_IGNORE_READS_BEGIN(); + T res = x; + ABSL_ANNOTATE_IGNORE_READS_END(); + return res; +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl +#endif + +#else + +#define ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() // empty +#define ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_END() // empty +#define ABSL_ANNOTATE_UNPROTECTED_READ(x) (x) + +#endif + +// ------------------------------------------------------------------------- +// Address sanitizer annotations + +#ifdef ABSL_HAVE_ADDRESS_SANITIZER +// Describe the current state of a contiguous container such as e.g. +// std::vector or std::string. For more details see +// sanitizer/common_interface_defs.h, which is provided by the compiler. +#include + +#define ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(beg, end, old_mid, new_mid) \ + __sanitizer_annotate_contiguous_container(beg, end, old_mid, new_mid) +#define ABSL_ADDRESS_SANITIZER_REDZONE(name) \ + struct { \ + alignas(8) char x[8]; \ + } name + +#else + +#define ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(beg, end, old_mid, new_mid) // empty +#define ABSL_ADDRESS_SANITIZER_REDZONE(name) static_assert(true, "") + +#endif // ABSL_HAVE_ADDRESS_SANITIZER + +// ------------------------------------------------------------------------- +// Undefine the macros intended only for this file. + +#undef ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_ANNOTALYSIS_ENABLED +#undef ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_BEGIN_EXTERN_C +#undef ABSL_INTERNAL_END_EXTERN_C +#undef ABSL_INTERNAL_STATIC_INLINE + +#endif // ABSL_BASE_DYNAMIC_ANNOTATIONS_H_ diff --git a/src/absl/base/internal/atomic_hook.h b/src/absl/base/internal/atomic_hook.h new file mode 100644 index 0000000..ae21cd7 --- /dev/null +++ b/src/absl/base/internal/atomic_hook.h @@ -0,0 +1,200 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_ATOMIC_HOOK_H_ +#define ABSL_BASE_INTERNAL_ATOMIC_HOOK_H_ + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" + +#if defined(_MSC_VER) && !defined(__clang__) +#define ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT 0 +#else +#define ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT 1 +#endif + +#if defined(_MSC_VER) +#define ABSL_HAVE_WORKING_ATOMIC_POINTER 0 +#else +#define ABSL_HAVE_WORKING_ATOMIC_POINTER 1 +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +template +class AtomicHook; + +// To workaround AtomicHook not being constant-initializable on some platforms, +// prefer to annotate instances with `ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES` +// instead of `ABSL_CONST_INIT`. +#if ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT +#define ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES ABSL_CONST_INIT +#else +#define ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES +#endif + +// `AtomicHook` is a helper class, templatized on a raw function pointer type, +// for implementing Abseil customization hooks. It is a callable object that +// dispatches to the registered hook. Objects of type `AtomicHook` must have +// static or thread storage duration. +// +// A default constructed object performs a no-op (and returns a default +// constructed object) if no hook has been registered. +// +// Hooks can be pre-registered via constant initialization, for example: +// +// ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES static AtomicHook +// my_hook(DefaultAction); +// +// and then changed at runtime via a call to `Store()`. +// +// Reads and writes guarantee memory_order_acquire/memory_order_release +// semantics. +template +class AtomicHook { + public: + using FnPtr = ReturnType (*)(Args...); + + // Constructs an object that by default performs a no-op (and + // returns a default constructed object) when no hook as been registered. + constexpr AtomicHook() : AtomicHook(DummyFunction) {} + + // Constructs an object that by default dispatches to/returns the + // pre-registered default_fn when no hook has been registered at runtime. +#if ABSL_HAVE_WORKING_ATOMIC_POINTER && ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT + explicit constexpr AtomicHook(FnPtr default_fn) + : hook_(default_fn), default_fn_(default_fn) {} +#elif ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT + explicit constexpr AtomicHook(FnPtr default_fn) + : hook_(kUninitialized), default_fn_(default_fn) {} +#else + // As of January 2020, on all known versions of MSVC this constructor runs in + // the global constructor sequence. If `Store()` is called by a dynamic + // initializer, we want to preserve the value, even if this constructor runs + // after the call to `Store()`. If not, `hook_` will be + // zero-initialized by the linker and we have no need to set it. + // https://developercommunity.visualstudio.com/content/problem/336946/class-with-constexpr-constructor-not-using-static.html + explicit constexpr AtomicHook(FnPtr default_fn) + : /* hook_(deliberately omitted), */ default_fn_(default_fn) { + static_assert(kUninitialized == 0, "here we rely on zero-initialization"); + } +#endif + + // Stores the provided function pointer as the value for this hook. + // + // This is intended to be called once. Multiple calls are legal only if the + // same function pointer is provided for each call. The store is implemented + // as a memory_order_release operation, and read accesses are implemented as + // memory_order_acquire. + void Store(FnPtr fn) { + bool success = DoStore(fn); + static_cast(success); + assert(success); + } + + // Invokes the registered callback. If no callback has yet been registered, a + // default-constructed object of the appropriate type is returned instead. + template + ReturnType operator()(CallArgs&&... args) const { + return DoLoad()(std::forward(args)...); + } + + // Returns the registered callback, or nullptr if none has been registered. + // Useful if client code needs to conditionalize behavior based on whether a + // callback was registered. + // + // Note that atomic_hook.Load()() and atomic_hook() have different semantics: + // operator()() will perform a no-op if no callback was registered, while + // Load()() will dereference a null function pointer. Prefer operator()() to + // Load()() unless you must conditionalize behavior on whether a hook was + // registered. + FnPtr Load() const { + FnPtr ptr = DoLoad(); + return (ptr == DummyFunction) ? nullptr : ptr; + } + + private: + static ReturnType DummyFunction(Args...) { + return ReturnType(); + } + + // Current versions of MSVC (as of September 2017) have a broken + // implementation of std::atomic: Its constructor attempts to do the + // equivalent of a reinterpret_cast in a constexpr context, which is not + // allowed. + // + // This causes an issue when building with LLVM under Windows. To avoid this, + // we use a less-efficient, intptr_t-based implementation on Windows. +#if ABSL_HAVE_WORKING_ATOMIC_POINTER + // Return the stored value, or DummyFunction if no value has been stored. + FnPtr DoLoad() const { return hook_.load(std::memory_order_acquire); } + + // Store the given value. Returns false if a different value was already + // stored to this object. + bool DoStore(FnPtr fn) { + assert(fn); + FnPtr expected = default_fn_; + const bool store_succeeded = hook_.compare_exchange_strong( + expected, fn, std::memory_order_acq_rel, std::memory_order_acquire); + const bool same_value_already_stored = (expected == fn); + return store_succeeded || same_value_already_stored; + } + + std::atomic hook_; +#else // !ABSL_HAVE_WORKING_ATOMIC_POINTER + // Use a sentinel value unlikely to be the address of an actual function. + static constexpr intptr_t kUninitialized = 0; + + static_assert(sizeof(intptr_t) >= sizeof(FnPtr), + "intptr_t can't contain a function pointer"); + + FnPtr DoLoad() const { + const intptr_t value = hook_.load(std::memory_order_acquire); + if (value == kUninitialized) { + return default_fn_; + } + return reinterpret_cast(value); + } + + bool DoStore(FnPtr fn) { + assert(fn); + const auto value = reinterpret_cast(fn); + intptr_t expected = kUninitialized; + const bool store_succeeded = hook_.compare_exchange_strong( + expected, value, std::memory_order_acq_rel, std::memory_order_acquire); + const bool same_value_already_stored = (expected == value); + return store_succeeded || same_value_already_stored; + } + + std::atomic hook_; +#endif + + const FnPtr default_fn_; +}; + +#undef ABSL_HAVE_WORKING_ATOMIC_POINTER +#undef ABSL_HAVE_WORKING_CONSTEXPR_STATIC_INIT + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_ATOMIC_HOOK_H_ diff --git a/src/absl/base/internal/cycleclock.cc b/src/absl/base/internal/cycleclock.cc new file mode 100644 index 0000000..902e3f5 --- /dev/null +++ b/src/absl/base/internal/cycleclock.cc @@ -0,0 +1,77 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// The implementation of CycleClock::Frequency. +// +// NOTE: only i386 and x86_64 have been well tested. +// PPC, sparc, alpha, and ia64 are based on +// http://peter.kuscsik.com/wordpress/?p=14 +// with modifications by m3b. See also +// https://setisvn.ssl.berkeley.edu/svn/lib/fftw-3.0.1/kernel/cycle.h + +#include "absl/base/internal/cycleclock.h" + +#include +#include // NOLINT(build/c++11) + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/internal/unscaledcycleclock.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +#if ABSL_USE_UNSCALED_CYCLECLOCK + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +constexpr int32_t CycleClock::kShift; +constexpr double CycleClock::kFrequencyScale; +#endif + +ABSL_CONST_INIT std::atomic + CycleClock::cycle_clock_source_{nullptr}; + +void CycleClockSource::Register(CycleClockSourceFunc source) { + // Corresponds to the load(std::memory_order_acquire) in LoadCycleClockSource. + CycleClock::cycle_clock_source_.store(source, std::memory_order_release); +} + +#ifdef _WIN32 +int64_t CycleClock::Now() { + auto fn = LoadCycleClockSource(); + if (fn == nullptr) { + return base_internal::UnscaledCycleClock::Now() >> kShift; + } + return fn() >> kShift; +} +#endif + +#else + +int64_t CycleClock::Now() { + return std::chrono::duration_cast( + std::chrono::steady_clock::now().time_since_epoch()) + .count(); +} + +double CycleClock::Frequency() { + return 1e9; +} + +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/cycleclock.h b/src/absl/base/internal/cycleclock.h new file mode 100644 index 0000000..9704e38 --- /dev/null +++ b/src/absl/base/internal/cycleclock.h @@ -0,0 +1,159 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +// ----------------------------------------------------------------------------- +// File: cycleclock.h +// ----------------------------------------------------------------------------- +// +// This header file defines a `CycleClock`, which yields the value and frequency +// of a cycle counter that increments at a rate that is approximately constant. +// +// NOTE: +// +// The cycle counter frequency is not necessarily related to the core clock +// frequency and should not be treated as such. That is, `CycleClock` cycles are +// not necessarily "CPU cycles" and code should not rely on that behavior, even +// if experimentally observed. +// +// An arbitrary offset may have been added to the counter at power on. +// +// On some platforms, the rate and offset of the counter may differ +// slightly when read from different CPUs of a multiprocessor. Usually, +// we try to ensure that the operating system adjusts values periodically +// so that values agree approximately. If you need stronger guarantees, +// consider using alternate interfaces. +// +// The CPU is not required to maintain the ordering of a cycle counter read +// with respect to surrounding instructions. + +#ifndef ABSL_BASE_INTERNAL_CYCLECLOCK_H_ +#define ABSL_BASE_INTERNAL_CYCLECLOCK_H_ + +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/internal/unscaledcycleclock.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +using CycleClockSourceFunc = int64_t (*)(); + +// ----------------------------------------------------------------------------- +// CycleClock +// ----------------------------------------------------------------------------- +class CycleClock { + public: + // CycleClock::Now() + // + // Returns the value of a cycle counter that counts at a rate that is + // approximately constant. + static int64_t Now(); + + // CycleClock::Frequency() + // + // Returns the amount by which `CycleClock::Now()` increases per second. Note + // that this value may not necessarily match the core CPU clock frequency. + static double Frequency(); + + private: +#if ABSL_USE_UNSCALED_CYCLECLOCK + static CycleClockSourceFunc LoadCycleClockSource(); + +#ifdef NDEBUG +#ifdef ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY + // Not debug mode and the UnscaledCycleClock frequency is the CPU + // frequency. Scale the CycleClock to prevent overflow if someone + // tries to represent the time as cycles since the Unix epoch. + static constexpr int32_t kShift = 1; +#else + // Not debug mode and the UnscaledCycleClock isn't operating at the + // raw CPU frequency. There is no need to do any scaling, so don't + // needlessly sacrifice precision. + static constexpr int32_t kShift = 0; +#endif +#else // NDEBUG + // In debug mode use a different shift to discourage depending on a + // particular shift value. + static constexpr int32_t kShift = 2; +#endif // NDEBUG + + static constexpr double kFrequencyScale = 1.0 / (1 << kShift); + ABSL_CONST_INIT static std::atomic cycle_clock_source_; +#endif // ABSL_USE_UNSCALED_CYCLECLOC + + CycleClock() = delete; // no instances + CycleClock(const CycleClock&) = delete; + CycleClock& operator=(const CycleClock&) = delete; + + friend class CycleClockSource; +}; + +class CycleClockSource { + private: + // CycleClockSource::Register() + // + // Register a function that provides an alternate source for the unscaled CPU + // cycle count value. The source function must be async signal safe, must not + // call CycleClock::Now(), and must have a frequency that matches that of the + // unscaled clock used by CycleClock. A nullptr value resets CycleClock to use + // the default source. + static void Register(CycleClockSourceFunc source); +}; + +#if ABSL_USE_UNSCALED_CYCLECLOCK + +inline CycleClockSourceFunc CycleClock::LoadCycleClockSource() { +#if !defined(__x86_64__) + // Optimize for the common case (no callback) by first doing a relaxed load; + // this is significantly faster on non-x86 platforms. + if (cycle_clock_source_.load(std::memory_order_relaxed) == nullptr) { + return nullptr; + } +#endif // !defined(__x86_64__) + + // This corresponds to the store(std::memory_order_release) in + // CycleClockSource::Register, and makes sure that any updates made prior to + // registering the callback are visible to this thread before the callback + // is invoked. + return cycle_clock_source_.load(std::memory_order_acquire); +} + +// Accessing globals in inlined code in Window DLLs is problematic. +#ifndef _WIN32 +inline int64_t CycleClock::Now() { + auto fn = LoadCycleClockSource(); + if (fn == nullptr) { + return base_internal::UnscaledCycleClock::Now() >> kShift; + } + return fn() >> kShift; +} +#endif + +inline double CycleClock::Frequency() { + return kFrequencyScale * base_internal::UnscaledCycleClock::Frequency(); +} + +#endif // ABSL_USE_UNSCALED_CYCLECLOCK + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_CYCLECLOCK_H_ diff --git a/src/absl/base/internal/direct_mmap.h b/src/absl/base/internal/direct_mmap.h new file mode 100644 index 0000000..e492bb0 --- /dev/null +++ b/src/absl/base/internal/direct_mmap.h @@ -0,0 +1,169 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Functions for directly invoking mmap() via syscall, avoiding the case where +// mmap() has been locally overridden. + +#ifndef ABSL_BASE_INTERNAL_DIRECT_MMAP_H_ +#define ABSL_BASE_INTERNAL_DIRECT_MMAP_H_ + +#include "absl/base/config.h" + +#ifdef ABSL_HAVE_MMAP + +#include + +#ifdef __linux__ + +#include +#ifdef __BIONIC__ +#include +#else +#include +#endif + +#include +#include +#include +#include +#include + +#ifdef __mips__ +// Include definitions of the ABI currently in use. +#if defined(__BIONIC__) || !defined(__GLIBC__) +// Android doesn't have sgidefs.h, but does have asm/sgidefs.h, which has the +// definitions we need. +#include +#else +#include +#endif // __BIONIC__ || !__GLIBC__ +#endif // __mips__ + +// SYS_mmap and SYS_munmap are not defined in Android. +#ifdef __BIONIC__ +extern "C" void* __mmap2(void*, size_t, int, int, int, size_t); +#if defined(__NR_mmap) && !defined(SYS_mmap) +#define SYS_mmap __NR_mmap +#endif +#ifndef SYS_munmap +#define SYS_munmap __NR_munmap +#endif +#endif // __BIONIC__ + +#if defined(__NR_mmap2) && !defined(SYS_mmap2) +#define SYS_mmap2 __NR_mmap2 +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Platform specific logic extracted from +// https://chromium.googlesource.com/linux-syscall-support/+/master/linux_syscall_support.h +inline void* DirectMmap(void* start, size_t length, int prot, int flags, int fd, + off64_t offset) noexcept { +#if defined(__i386__) || defined(__ARM_ARCH_3__) || defined(__ARM_EABI__) || \ + defined(__m68k__) || defined(__sh__) || \ + (defined(__hppa__) && !defined(__LP64__)) || \ + (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) || \ + (defined(__PPC__) && !defined(__PPC64__)) || \ + (defined(__riscv) && __riscv_xlen == 32) || \ + (defined(__s390__) && !defined(__s390x__)) || \ + (defined(__sparc__) && !defined(__arch64__)) + // On these architectures, implement mmap with mmap2. + static int pagesize = 0; + if (pagesize == 0) { +#if defined(__wasm__) || defined(__asmjs__) + pagesize = getpagesize(); +#else + pagesize = sysconf(_SC_PAGESIZE); +#endif + } + if (offset < 0 || offset % pagesize != 0) { + errno = EINVAL; + return MAP_FAILED; + } +#ifdef __BIONIC__ + // SYS_mmap2 has problems on Android API level <= 16. + // Workaround by invoking __mmap2() instead. + return __mmap2(start, length, prot, flags, fd, offset / pagesize); +#else + return reinterpret_cast( + syscall(SYS_mmap2, start, length, prot, flags, fd, + static_cast(offset / pagesize))); +#endif +#elif defined(__s390x__) + // On s390x, mmap() arguments are passed in memory. + unsigned long buf[6] = {reinterpret_cast(start), // NOLINT + static_cast(length), // NOLINT + static_cast(prot), // NOLINT + static_cast(flags), // NOLINT + static_cast(fd), // NOLINT + static_cast(offset)}; // NOLINT + return reinterpret_cast(syscall(SYS_mmap, buf)); +#elif defined(__x86_64__) +// The x32 ABI has 32 bit longs, but the syscall interface is 64 bit. +// We need to explicitly cast to an unsigned 64 bit type to avoid implicit +// sign extension. We can't cast pointers directly because those are +// 32 bits, and gcc will dump ugly warnings about casting from a pointer +// to an integer of a different size. We also need to make sure __off64_t +// isn't truncated to 32-bits under x32. +#define MMAP_SYSCALL_ARG(x) ((uint64_t)(uintptr_t)(x)) + return reinterpret_cast( + syscall(SYS_mmap, MMAP_SYSCALL_ARG(start), MMAP_SYSCALL_ARG(length), + MMAP_SYSCALL_ARG(prot), MMAP_SYSCALL_ARG(flags), + MMAP_SYSCALL_ARG(fd), static_cast(offset))); +#undef MMAP_SYSCALL_ARG +#else // Remaining 64-bit aritectures. + static_assert(sizeof(unsigned long) == 8, "Platform is not 64-bit"); + return reinterpret_cast( + syscall(SYS_mmap, start, length, prot, flags, fd, offset)); +#endif +} + +inline int DirectMunmap(void* start, size_t length) { + return static_cast(syscall(SYS_munmap, start, length)); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#else // !__linux__ + +// For non-linux platforms where we have mmap, just dispatch directly to the +// actual mmap()/munmap() methods. + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +inline void* DirectMmap(void* start, size_t length, int prot, int flags, int fd, + off_t offset) { + return mmap(start, length, prot, flags, fd, offset); +} + +inline int DirectMunmap(void* start, size_t length) { + return munmap(start, length); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // __linux__ + +#endif // ABSL_HAVE_MMAP + +#endif // ABSL_BASE_INTERNAL_DIRECT_MMAP_H_ diff --git a/src/absl/base/internal/dynamic_annotations.h b/src/absl/base/internal/dynamic_annotations.h new file mode 100644 index 0000000..b23c5ec --- /dev/null +++ b/src/absl/base/internal/dynamic_annotations.h @@ -0,0 +1,398 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This file defines dynamic annotations for use with dynamic analysis tool +// such as valgrind, PIN, etc. +// +// Dynamic annotation is a source code annotation that affects the generated +// code (that is, the annotation is not a comment). Each such annotation is +// attached to a particular instruction and/or to a particular object (address) +// in the program. +// +// The annotations that should be used by users are macros in all upper-case +// (e.g., ANNOTATE_THREAD_NAME). +// +// Actual implementation of these macros may differ depending on the dynamic +// analysis tool being used. +// +// This file supports the following configurations: +// - Dynamic Annotations enabled (with static thread-safety warnings disabled). +// In this case, macros expand to functions implemented by Thread Sanitizer, +// when building with TSan. When not provided an external implementation, +// dynamic_annotations.cc provides no-op implementations. +// +// - Static Clang thread-safety warnings enabled. +// When building with a Clang compiler that supports thread-safety warnings, +// a subset of annotations can be statically-checked at compile-time. We +// expand these macros to static-inline functions that can be analyzed for +// thread-safety, but afterwards elided when building the final binary. +// +// - All annotations are disabled. +// If neither Dynamic Annotations nor Clang thread-safety warnings are +// enabled, then all annotation-macros expand to empty. + +#ifndef ABSL_BASE_INTERNAL_DYNAMIC_ANNOTATIONS_H_ +#define ABSL_BASE_INTERNAL_DYNAMIC_ANNOTATIONS_H_ + +#include + +#include "absl/base/config.h" + +// ------------------------------------------------------------------------- +// Decide which features are enabled + +#ifndef DYNAMIC_ANNOTATIONS_ENABLED +#define DYNAMIC_ANNOTATIONS_ENABLED 0 +#endif + +#if defined(__clang__) && !defined(SWIG) +#define ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED 1 +#endif + +#if DYNAMIC_ANNOTATIONS_ENABLED != 0 + +#define ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED 1 +#define ABSL_INTERNAL_ANNOTALYSIS_ENABLED 0 +#define ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED 1 + +#else + +#define ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED 0 +#define ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED 0 +#define ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED 0 + +// Clang provides limited support for static thread-safety analysis through a +// feature called Annotalysis. We configure macro-definitions according to +// whether Annotalysis support is available. When running in opt-mode, GCC +// will issue a warning, if these attributes are compiled. Only include them +// when compiling using Clang. + +// ANNOTALYSIS_ENABLED == 1 when IGNORE_READ_ATTRIBUTE_ENABLED == 1 +#define ABSL_INTERNAL_ANNOTALYSIS_ENABLED \ + defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) +// Read/write annotations are enabled in Annotalysis mode; disabled otherwise. +#define ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED \ + ABSL_INTERNAL_ANNOTALYSIS_ENABLED +#endif + +// Memory annotations are also made available to LLVM's Memory Sanitizer +#if defined(ABSL_HAVE_MEMORY_SANITIZER) && !defined(__native_client__) +#define ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED 1 +#endif + +#ifndef ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED +#define ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED 0 +#endif + +#ifdef __cplusplus +#define ABSL_INTERNAL_BEGIN_EXTERN_C extern "C" { +#define ABSL_INTERNAL_END_EXTERN_C } // extern "C" +#define ABSL_INTERNAL_GLOBAL_SCOPED(F) ::F +#define ABSL_INTERNAL_STATIC_INLINE inline +#else +#define ABSL_INTERNAL_BEGIN_EXTERN_C // empty +#define ABSL_INTERNAL_END_EXTERN_C // empty +#define ABSL_INTERNAL_GLOBAL_SCOPED(F) F +#define ABSL_INTERNAL_STATIC_INLINE static inline +#endif + +// ------------------------------------------------------------------------- +// Define race annotations. + +#if ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED == 1 + +// ------------------------------------------------------------- +// Annotations that suppress errors. It is usually better to express the +// program's synchronization using the other annotations, but these can be used +// when all else fails. + +// Report that we may have a benign race at `pointer`, with size +// "sizeof(*(pointer))". `pointer` must be a non-void* pointer. Insert at the +// point where `pointer` has been allocated, preferably close to the point +// where the race happens. See also ANNOTATE_BENIGN_RACE_STATIC. +#define ANNOTATE_BENIGN_RACE(pointer, description) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateBenignRaceSized) \ + (__FILE__, __LINE__, pointer, sizeof(*(pointer)), description) + +// Same as ANNOTATE_BENIGN_RACE(`address`, `description`), but applies to +// the memory range [`address`, `address`+`size`). +#define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateBenignRaceSized) \ + (__FILE__, __LINE__, address, size, description) + +// Enable (`enable`!=0) or disable (`enable`==0) race detection for all threads. +// This annotation could be useful if you want to skip expensive race analysis +// during some period of program execution, e.g. during initialization. +#define ANNOTATE_ENABLE_RACE_DETECTION(enable) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateEnableRaceDetection) \ + (__FILE__, __LINE__, enable) + +// ------------------------------------------------------------- +// Annotations useful for debugging. + +// Report the current thread `name` to a race detector. +#define ANNOTATE_THREAD_NAME(name) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateThreadName)(__FILE__, __LINE__, name) + +// ------------------------------------------------------------- +// Annotations useful when implementing locks. They are not normally needed by +// modules that merely use locks. The `lock` argument is a pointer to the lock +// object. + +// Report that a lock has been created at address `lock`. +#define ANNOTATE_RWLOCK_CREATE(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockCreate)(__FILE__, __LINE__, lock) + +// Report that a linker initialized lock has been created at address `lock`. +#ifdef ABSL_HAVE_THREAD_SANITIZER +#define ANNOTATE_RWLOCK_CREATE_STATIC(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockCreateStatic) \ + (__FILE__, __LINE__, lock) +#else +#define ANNOTATE_RWLOCK_CREATE_STATIC(lock) ANNOTATE_RWLOCK_CREATE(lock) +#endif + +// Report that the lock at address `lock` is about to be destroyed. +#define ANNOTATE_RWLOCK_DESTROY(lock) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockDestroy)(__FILE__, __LINE__, lock) + +// Report that the lock at address `lock` has been acquired. +// `is_w`=1 for writer lock, `is_w`=0 for reader lock. +#define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockAcquired) \ + (__FILE__, __LINE__, lock, is_w) + +// Report that the lock at address `lock` is about to be released. +// `is_w`=1 for writer lock, `is_w`=0 for reader lock. +#define ANNOTATE_RWLOCK_RELEASED(lock, is_w) \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateRWLockReleased) \ + (__FILE__, __LINE__, lock, is_w) + +// Apply ANNOTATE_BENIGN_RACE_SIZED to a static variable `static_var`. +#define ANNOTATE_BENIGN_RACE_STATIC(static_var, description) \ + namespace { \ + class static_var##_annotator { \ + public: \ + static_var##_annotator() { \ + ANNOTATE_BENIGN_RACE_SIZED(&static_var, sizeof(static_var), \ + #static_var ": " description); \ + } \ + }; \ + static static_var##_annotator the##static_var##_annotator; \ + } // namespace + +#else // ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED == 0 + +#define ANNOTATE_RWLOCK_CREATE(lock) // empty +#define ANNOTATE_RWLOCK_CREATE_STATIC(lock) // empty +#define ANNOTATE_RWLOCK_DESTROY(lock) // empty +#define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) // empty +#define ANNOTATE_RWLOCK_RELEASED(lock, is_w) // empty +#define ANNOTATE_BENIGN_RACE(address, description) // empty +#define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) // empty +#define ANNOTATE_THREAD_NAME(name) // empty +#define ANNOTATE_ENABLE_RACE_DETECTION(enable) // empty +#define ANNOTATE_BENIGN_RACE_STATIC(static_var, description) // empty + +#endif // ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED + +// ------------------------------------------------------------------------- +// Define memory annotations. + +#if ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED == 1 + +#include + +#define ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \ + __msan_unpoison(address, size) + +#define ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) \ + __msan_allocated_memory(address, size) + +#else // ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED == 0 + +#if DYNAMIC_ANNOTATIONS_ENABLED == 1 +#define ANNOTATE_MEMORY_IS_INITIALIZED(address, size) \ + do { \ + (void)(address); \ + (void)(size); \ + } while (0) +#define ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) \ + do { \ + (void)(address); \ + (void)(size); \ + } while (0) +#else +#define ANNOTATE_MEMORY_IS_INITIALIZED(address, size) // empty +#define ANNOTATE_MEMORY_IS_UNINITIALIZED(address, size) // empty +#endif + +#endif // ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED + +// ------------------------------------------------------------------------- +// Define IGNORE_READS_BEGIN/_END attributes. + +#if defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +#define ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE \ + __attribute((exclusive_lock_function("*"))) +#define ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE \ + __attribute((unlock_function("*"))) + +#else // !defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +#define ABSL_INTERNAL_IGNORE_READS_BEGIN_ATTRIBUTE // empty +#define ABSL_INTERNAL_IGNORE_READS_END_ATTRIBUTE // empty + +#endif // defined(ABSL_INTERNAL_IGNORE_READS_ATTRIBUTE_ENABLED) + +// ------------------------------------------------------------------------- +// Define IGNORE_READS_BEGIN/_END annotations. + +#if ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED == 1 + +// Request the analysis tool to ignore all reads in the current thread until +// ANNOTATE_IGNORE_READS_END is called. Useful to ignore intentional racey +// reads, while still checking other reads and all writes. +// See also ANNOTATE_UNPROTECTED_READ. +#define ANNOTATE_IGNORE_READS_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreReadsBegin)(__FILE__, __LINE__) + +// Stop ignoring reads. +#define ANNOTATE_IGNORE_READS_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreReadsEnd)(__FILE__, __LINE__) + +#elif defined(ABSL_INTERNAL_ANNOTALYSIS_ENABLED) + +// When Annotalysis is enabled without Dynamic Annotations, the use of +// static-inline functions allows the annotations to be read at compile-time, +// while still letting the compiler elide the functions from the final build. +// +// TODO(delesley) -- The exclusive lock here ignores writes as well, but +// allows IGNORE_READS_AND_WRITES to work properly. + +#define ANNOTATE_IGNORE_READS_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AbslInternalAnnotateIgnoreReadsBegin)() + +#define ANNOTATE_IGNORE_READS_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AbslInternalAnnotateIgnoreReadsEnd)() + +#else + +#define ANNOTATE_IGNORE_READS_BEGIN() // empty +#define ANNOTATE_IGNORE_READS_END() // empty + +#endif + +// ------------------------------------------------------------------------- +// Define IGNORE_WRITES_BEGIN/_END annotations. + +#if ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED == 1 + +// Similar to ANNOTATE_IGNORE_READS_BEGIN, but ignore writes instead. +#define ANNOTATE_IGNORE_WRITES_BEGIN() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreWritesBegin)(__FILE__, __LINE__) + +// Stop ignoring writes. +#define ANNOTATE_IGNORE_WRITES_END() \ + ABSL_INTERNAL_GLOBAL_SCOPED(AnnotateIgnoreWritesEnd)(__FILE__, __LINE__) + +#else + +#define ANNOTATE_IGNORE_WRITES_BEGIN() // empty +#define ANNOTATE_IGNORE_WRITES_END() // empty + +#endif + +// ------------------------------------------------------------------------- +// Define the ANNOTATE_IGNORE_READS_AND_WRITES_* annotations using the more +// primitive annotations defined above. +// +// Instead of doing +// ANNOTATE_IGNORE_READS_BEGIN(); +// ... = x; +// ANNOTATE_IGNORE_READS_END(); +// one can use +// ... = ANNOTATE_UNPROTECTED_READ(x); + +#if defined(ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED) + +// Start ignoring all memory accesses (both reads and writes). +#define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() \ + do { \ + ANNOTATE_IGNORE_READS_BEGIN(); \ + ANNOTATE_IGNORE_WRITES_BEGIN(); \ + } while (0) + +// Stop ignoring both reads and writes. +#define ANNOTATE_IGNORE_READS_AND_WRITES_END() \ + do { \ + ANNOTATE_IGNORE_WRITES_END(); \ + ANNOTATE_IGNORE_READS_END(); \ + } while (0) + +#ifdef __cplusplus +// ANNOTATE_UNPROTECTED_READ is the preferred way to annotate racey reads. +#define ANNOTATE_UNPROTECTED_READ(x) \ + absl::base_internal::AnnotateUnprotectedRead(x) + +#endif + +#else + +#define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() // empty +#define ANNOTATE_IGNORE_READS_AND_WRITES_END() // empty +#define ANNOTATE_UNPROTECTED_READ(x) (x) + +#endif + +// ------------------------------------------------------------------------- +// Address sanitizer annotations + +#ifdef ABSL_HAVE_ADDRESS_SANITIZER +// Describe the current state of a contiguous container such as e.g. +// std::vector or std::string. For more details see +// sanitizer/common_interface_defs.h, which is provided by the compiler. +#include + +#define ANNOTATE_CONTIGUOUS_CONTAINER(beg, end, old_mid, new_mid) \ + __sanitizer_annotate_contiguous_container(beg, end, old_mid, new_mid) +#define ADDRESS_SANITIZER_REDZONE(name) \ + struct { \ + char x[8] __attribute__((aligned(8))); \ + } name + +#else + +#define ANNOTATE_CONTIGUOUS_CONTAINER(beg, end, old_mid, new_mid) +#define ADDRESS_SANITIZER_REDZONE(name) static_assert(true, "") + +#endif // ABSL_HAVE_ADDRESS_SANITIZER + +// ------------------------------------------------------------------------- +// Undefine the macros intended only for this file. + +#undef ABSL_INTERNAL_RACE_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_MEMORY_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_READS_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_WRITES_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_ANNOTALYSIS_ENABLED +#undef ABSL_INTERNAL_READS_WRITES_ANNOTATIONS_ENABLED +#undef ABSL_INTERNAL_BEGIN_EXTERN_C +#undef ABSL_INTERNAL_END_EXTERN_C +#undef ABSL_INTERNAL_STATIC_INLINE + +#endif // ABSL_BASE_INTERNAL_DYNAMIC_ANNOTATIONS_H_ diff --git a/src/absl/base/internal/endian.h b/src/absl/base/internal/endian.h new file mode 100644 index 0000000..50747d7 --- /dev/null +++ b/src/absl/base/internal/endian.h @@ -0,0 +1,282 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_ENDIAN_H_ +#define ABSL_BASE_INTERNAL_ENDIAN_H_ + +#include +#include + +#include "absl/base/casts.h" +#include "absl/base/config.h" +#include "absl/base/internal/unaligned_access.h" +#include "absl/base/port.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +inline uint64_t gbswap_64(uint64_t host_int) { +#if ABSL_HAVE_BUILTIN(__builtin_bswap64) || defined(__GNUC__) + return __builtin_bswap64(host_int); +#elif defined(_MSC_VER) + return _byteswap_uint64(host_int); +#else + return (((host_int & uint64_t{0xFF}) << 56) | + ((host_int & uint64_t{0xFF00}) << 40) | + ((host_int & uint64_t{0xFF0000}) << 24) | + ((host_int & uint64_t{0xFF000000}) << 8) | + ((host_int & uint64_t{0xFF00000000}) >> 8) | + ((host_int & uint64_t{0xFF0000000000}) >> 24) | + ((host_int & uint64_t{0xFF000000000000}) >> 40) | + ((host_int & uint64_t{0xFF00000000000000}) >> 56)); +#endif +} + +inline uint32_t gbswap_32(uint32_t host_int) { +#if ABSL_HAVE_BUILTIN(__builtin_bswap32) || defined(__GNUC__) + return __builtin_bswap32(host_int); +#elif defined(_MSC_VER) + return _byteswap_ulong(host_int); +#else + return (((host_int & uint32_t{0xFF}) << 24) | + ((host_int & uint32_t{0xFF00}) << 8) | + ((host_int & uint32_t{0xFF0000}) >> 8) | + ((host_int & uint32_t{0xFF000000}) >> 24)); +#endif +} + +inline uint16_t gbswap_16(uint16_t host_int) { +#if ABSL_HAVE_BUILTIN(__builtin_bswap16) || defined(__GNUC__) + return __builtin_bswap16(host_int); +#elif defined(_MSC_VER) + return _byteswap_ushort(host_int); +#else + return (((host_int & uint16_t{0xFF}) << 8) | + ((host_int & uint16_t{0xFF00}) >> 8)); +#endif +} + +#ifdef ABSL_IS_LITTLE_ENDIAN + +// Portable definitions for htonl (host-to-network) and friends on little-endian +// architectures. +inline uint16_t ghtons(uint16_t x) { return gbswap_16(x); } +inline uint32_t ghtonl(uint32_t x) { return gbswap_32(x); } +inline uint64_t ghtonll(uint64_t x) { return gbswap_64(x); } + +#elif defined ABSL_IS_BIG_ENDIAN + +// Portable definitions for htonl (host-to-network) etc on big-endian +// architectures. These definitions are simpler since the host byte order is the +// same as network byte order. +inline uint16_t ghtons(uint16_t x) { return x; } +inline uint32_t ghtonl(uint32_t x) { return x; } +inline uint64_t ghtonll(uint64_t x) { return x; } + +#else +#error \ + "Unsupported byte order: Either ABSL_IS_BIG_ENDIAN or " \ + "ABSL_IS_LITTLE_ENDIAN must be defined" +#endif // byte order + +inline uint16_t gntohs(uint16_t x) { return ghtons(x); } +inline uint32_t gntohl(uint32_t x) { return ghtonl(x); } +inline uint64_t gntohll(uint64_t x) { return ghtonll(x); } + +// Utilities to convert numbers between the current hosts's native byte +// order and little-endian byte order +// +// Load/Store methods are alignment safe +namespace little_endian { +// Conversion functions. +#ifdef ABSL_IS_LITTLE_ENDIAN + +inline uint16_t FromHost16(uint16_t x) { return x; } +inline uint16_t ToHost16(uint16_t x) { return x; } + +inline uint32_t FromHost32(uint32_t x) { return x; } +inline uint32_t ToHost32(uint32_t x) { return x; } + +inline uint64_t FromHost64(uint64_t x) { return x; } +inline uint64_t ToHost64(uint64_t x) { return x; } + +inline constexpr bool IsLittleEndian() { return true; } + +#elif defined ABSL_IS_BIG_ENDIAN + +inline uint16_t FromHost16(uint16_t x) { return gbswap_16(x); } +inline uint16_t ToHost16(uint16_t x) { return gbswap_16(x); } + +inline uint32_t FromHost32(uint32_t x) { return gbswap_32(x); } +inline uint32_t ToHost32(uint32_t x) { return gbswap_32(x); } + +inline uint64_t FromHost64(uint64_t x) { return gbswap_64(x); } +inline uint64_t ToHost64(uint64_t x) { return gbswap_64(x); } + +inline constexpr bool IsLittleEndian() { return false; } + +#endif /* ENDIAN */ + +inline uint8_t FromHost(uint8_t x) { return x; } +inline uint16_t FromHost(uint16_t x) { return FromHost16(x); } +inline uint32_t FromHost(uint32_t x) { return FromHost32(x); } +inline uint64_t FromHost(uint64_t x) { return FromHost64(x); } +inline uint8_t ToHost(uint8_t x) { return x; } +inline uint16_t ToHost(uint16_t x) { return ToHost16(x); } +inline uint32_t ToHost(uint32_t x) { return ToHost32(x); } +inline uint64_t ToHost(uint64_t x) { return ToHost64(x); } + +inline int8_t FromHost(int8_t x) { return x; } +inline int16_t FromHost(int16_t x) { + return bit_cast(FromHost16(bit_cast(x))); +} +inline int32_t FromHost(int32_t x) { + return bit_cast(FromHost32(bit_cast(x))); +} +inline int64_t FromHost(int64_t x) { + return bit_cast(FromHost64(bit_cast(x))); +} +inline int8_t ToHost(int8_t x) { return x; } +inline int16_t ToHost(int16_t x) { + return bit_cast(ToHost16(bit_cast(x))); +} +inline int32_t ToHost(int32_t x) { + return bit_cast(ToHost32(bit_cast(x))); +} +inline int64_t ToHost(int64_t x) { + return bit_cast(ToHost64(bit_cast(x))); +} + +// Functions to do unaligned loads and stores in little-endian order. +inline uint16_t Load16(const void *p) { + return ToHost16(ABSL_INTERNAL_UNALIGNED_LOAD16(p)); +} + +inline void Store16(void *p, uint16_t v) { + ABSL_INTERNAL_UNALIGNED_STORE16(p, FromHost16(v)); +} + +inline uint32_t Load32(const void *p) { + return ToHost32(ABSL_INTERNAL_UNALIGNED_LOAD32(p)); +} + +inline void Store32(void *p, uint32_t v) { + ABSL_INTERNAL_UNALIGNED_STORE32(p, FromHost32(v)); +} + +inline uint64_t Load64(const void *p) { + return ToHost64(ABSL_INTERNAL_UNALIGNED_LOAD64(p)); +} + +inline void Store64(void *p, uint64_t v) { + ABSL_INTERNAL_UNALIGNED_STORE64(p, FromHost64(v)); +} + +} // namespace little_endian + +// Utilities to convert numbers between the current hosts's native byte +// order and big-endian byte order (same as network byte order) +// +// Load/Store methods are alignment safe +namespace big_endian { +#ifdef ABSL_IS_LITTLE_ENDIAN + +inline uint16_t FromHost16(uint16_t x) { return gbswap_16(x); } +inline uint16_t ToHost16(uint16_t x) { return gbswap_16(x); } + +inline uint32_t FromHost32(uint32_t x) { return gbswap_32(x); } +inline uint32_t ToHost32(uint32_t x) { return gbswap_32(x); } + +inline uint64_t FromHost64(uint64_t x) { return gbswap_64(x); } +inline uint64_t ToHost64(uint64_t x) { return gbswap_64(x); } + +inline constexpr bool IsLittleEndian() { return true; } + +#elif defined ABSL_IS_BIG_ENDIAN + +inline uint16_t FromHost16(uint16_t x) { return x; } +inline uint16_t ToHost16(uint16_t x) { return x; } + +inline uint32_t FromHost32(uint32_t x) { return x; } +inline uint32_t ToHost32(uint32_t x) { return x; } + +inline uint64_t FromHost64(uint64_t x) { return x; } +inline uint64_t ToHost64(uint64_t x) { return x; } + +inline constexpr bool IsLittleEndian() { return false; } + +#endif /* ENDIAN */ + +inline uint8_t FromHost(uint8_t x) { return x; } +inline uint16_t FromHost(uint16_t x) { return FromHost16(x); } +inline uint32_t FromHost(uint32_t x) { return FromHost32(x); } +inline uint64_t FromHost(uint64_t x) { return FromHost64(x); } +inline uint8_t ToHost(uint8_t x) { return x; } +inline uint16_t ToHost(uint16_t x) { return ToHost16(x); } +inline uint32_t ToHost(uint32_t x) { return ToHost32(x); } +inline uint64_t ToHost(uint64_t x) { return ToHost64(x); } + +inline int8_t FromHost(int8_t x) { return x; } +inline int16_t FromHost(int16_t x) { + return bit_cast(FromHost16(bit_cast(x))); +} +inline int32_t FromHost(int32_t x) { + return bit_cast(FromHost32(bit_cast(x))); +} +inline int64_t FromHost(int64_t x) { + return bit_cast(FromHost64(bit_cast(x))); +} +inline int8_t ToHost(int8_t x) { return x; } +inline int16_t ToHost(int16_t x) { + return bit_cast(ToHost16(bit_cast(x))); +} +inline int32_t ToHost(int32_t x) { + return bit_cast(ToHost32(bit_cast(x))); +} +inline int64_t ToHost(int64_t x) { + return bit_cast(ToHost64(bit_cast(x))); +} + +// Functions to do unaligned loads and stores in big-endian order. +inline uint16_t Load16(const void *p) { + return ToHost16(ABSL_INTERNAL_UNALIGNED_LOAD16(p)); +} + +inline void Store16(void *p, uint16_t v) { + ABSL_INTERNAL_UNALIGNED_STORE16(p, FromHost16(v)); +} + +inline uint32_t Load32(const void *p) { + return ToHost32(ABSL_INTERNAL_UNALIGNED_LOAD32(p)); +} + +inline void Store32(void *p, uint32_t v) { + ABSL_INTERNAL_UNALIGNED_STORE32(p, FromHost32(v)); +} + +inline uint64_t Load64(const void *p) { + return ToHost64(ABSL_INTERNAL_UNALIGNED_LOAD64(p)); +} + +inline void Store64(void *p, uint64_t v) { + ABSL_INTERNAL_UNALIGNED_STORE64(p, FromHost64(v)); +} + +} // namespace big_endian + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_ENDIAN_H_ diff --git a/src/absl/base/internal/errno_saver.h b/src/absl/base/internal/errno_saver.h new file mode 100644 index 0000000..251de51 --- /dev/null +++ b/src/absl/base/internal/errno_saver.h @@ -0,0 +1,43 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_ERRNO_SAVER_H_ +#define ABSL_BASE_INTERNAL_ERRNO_SAVER_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// `ErrnoSaver` captures the value of `errno` upon construction and restores it +// upon deletion. It is used in low-level code and must be super fast. Do not +// add instrumentation, even in debug modes. +class ErrnoSaver { + public: + ErrnoSaver() : saved_errno_(errno) {} + ~ErrnoSaver() { errno = saved_errno_; } + int operator()() const { return saved_errno_; } + + private: + const int saved_errno_; +}; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_ERRNO_SAVER_H_ diff --git a/src/absl/base/internal/fast_type_id.h b/src/absl/base/internal/fast_type_id.h new file mode 100644 index 0000000..a547b3a --- /dev/null +++ b/src/absl/base/internal/fast_type_id.h @@ -0,0 +1,50 @@ +// +// Copyright 2020 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_FAST_TYPE_ID_H_ +#define ABSL_BASE_INTERNAL_FAST_TYPE_ID_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +template +struct FastTypeTag { + constexpr static char dummy_var = 0; +}; + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +template +constexpr char FastTypeTag::dummy_var; +#endif + +// FastTypeId() evaluates at compile/link-time to a unique pointer for the +// passed-in type. These are meant to be good match for keys into maps or +// straight up comparisons. +using FastTypeIdType = const void*; + +template +constexpr inline FastTypeIdType FastTypeId() { + return &FastTypeTag::dummy_var; +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_FAST_TYPE_ID_H_ diff --git a/src/absl/base/internal/hide_ptr.h b/src/absl/base/internal/hide_ptr.h new file mode 100644 index 0000000..1dba809 --- /dev/null +++ b/src/absl/base/internal/hide_ptr.h @@ -0,0 +1,51 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_HIDE_PTR_H_ +#define ABSL_BASE_INTERNAL_HIDE_PTR_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Arbitrary value with high bits set. Xor'ing with it is unlikely +// to map one valid pointer to another valid pointer. +constexpr uintptr_t HideMask() { + return (uintptr_t{0xF03A5F7BU} << (sizeof(uintptr_t) - 4) * 8) | 0xF03A5F7BU; +} + +// Hide a pointer from the leak checker. For internal use only. +// Differs from absl::IgnoreLeak(ptr) in that absl::IgnoreLeak(ptr) causes ptr +// and all objects reachable from ptr to be ignored by the leak checker. +template +inline uintptr_t HidePtr(T* ptr) { + return reinterpret_cast(ptr) ^ HideMask(); +} + +// Return a pointer that has been hidden from the leak checker. +// For internal use only. +template +inline T* UnhidePtr(uintptr_t hidden) { + return reinterpret_cast(hidden ^ HideMask()); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_HIDE_PTR_H_ diff --git a/src/absl/base/internal/identity.h b/src/absl/base/internal/identity.h new file mode 100644 index 0000000..a3154ed --- /dev/null +++ b/src/absl/base/internal/identity.h @@ -0,0 +1,37 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_IDENTITY_H_ +#define ABSL_BASE_INTERNAL_IDENTITY_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace internal { + +template +struct identity { + typedef T type; +}; + +template +using identity_t = typename identity::type; + +} // namespace internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_IDENTITY_H_ diff --git a/src/absl/base/internal/inline_variable.h b/src/absl/base/internal/inline_variable.h new file mode 100644 index 0000000..130d8c2 --- /dev/null +++ b/src/absl/base/internal/inline_variable.h @@ -0,0 +1,107 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_INLINE_VARIABLE_EMULATION_H_ +#define ABSL_BASE_INTERNAL_INLINE_VARIABLE_EMULATION_H_ + +#include + +#include "absl/base/internal/identity.h" + +// File: +// This file define a macro that allows the creation of or emulation of C++17 +// inline variables based on whether or not the feature is supported. + +//////////////////////////////////////////////////////////////////////////////// +// Macro: ABSL_INTERNAL_INLINE_CONSTEXPR(type, name, init) +// +// Description: +// Expands to the equivalent of an inline constexpr instance of the specified +// `type` and `name`, initialized to the value `init`. If the compiler being +// used is detected as supporting actual inline variables as a language +// feature, then the macro expands to an actual inline variable definition. +// +// Requires: +// `type` is a type that is usable in an extern variable declaration. +// +// Requires: `name` is a valid identifier +// +// Requires: +// `init` is an expression that can be used in the following definition: +// constexpr type name = init; +// +// Usage: +// +// // Equivalent to: `inline constexpr size_t variant_npos = -1;` +// ABSL_INTERNAL_INLINE_CONSTEXPR(size_t, variant_npos, -1); +// +// Differences in implementation: +// For a direct, language-level inline variable, decltype(name) will be the +// type that was specified along with const qualification, whereas for +// emulated inline variables, decltype(name) may be different (in practice +// it will likely be a reference type). +//////////////////////////////////////////////////////////////////////////////// + +#ifdef __cpp_inline_variables + +// Clang's -Wmissing-variable-declarations option erroneously warned that +// inline constexpr objects need to be pre-declared. This has now been fixed, +// but we will need to support this workaround for people building with older +// versions of clang. +// +// Bug: https://bugs.llvm.org/show_bug.cgi?id=35862 +// +// Note: +// identity_t is used here so that the const and name are in the +// appropriate place for pointer types, reference types, function pointer +// types, etc.. +#if defined(__clang__) +#define ABSL_INTERNAL_EXTERN_DECL(type, name) \ + extern const ::absl::internal::identity_t name; +#else // Otherwise, just define the macro to do nothing. +#define ABSL_INTERNAL_EXTERN_DECL(type, name) +#endif // defined(__clang__) + +// See above comment at top of file for details. +#define ABSL_INTERNAL_INLINE_CONSTEXPR(type, name, init) \ + ABSL_INTERNAL_EXTERN_DECL(type, name) \ + inline constexpr ::absl::internal::identity_t name = init + +#else + +// See above comment at top of file for details. +// +// Note: +// identity_t is used here so that the const and name are in the +// appropriate place for pointer types, reference types, function pointer +// types, etc.. +#define ABSL_INTERNAL_INLINE_CONSTEXPR(var_type, name, init) \ + template \ + struct AbslInternalInlineVariableHolder##name { \ + static constexpr ::absl::internal::identity_t kInstance = init; \ + }; \ + \ + template \ + constexpr ::absl::internal::identity_t \ + AbslInternalInlineVariableHolder##name::kInstance; \ + \ + static constexpr const ::absl::internal::identity_t& \ + name = /* NOLINT */ \ + AbslInternalInlineVariableHolder##name<>::kInstance; \ + static_assert(sizeof(void (*)(decltype(name))) != 0, \ + "Silence unused variable warnings.") + +#endif // __cpp_inline_variables + +#endif // ABSL_BASE_INTERNAL_INLINE_VARIABLE_EMULATION_H_ diff --git a/src/absl/base/internal/invoke.h b/src/absl/base/internal/invoke.h new file mode 100644 index 0000000..80befbb --- /dev/null +++ b/src/absl/base/internal/invoke.h @@ -0,0 +1,241 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// absl::base_internal::invoke(f, args...) is an implementation of +// INVOKE(f, args...) from section [func.require] of the C++ standard. +// When compiled as C++17 and later versions, it is implemented as an alias of +// std::invoke. +// +// [func.require] +// Define INVOKE (f, t1, t2, ..., tN) as follows: +// 1. (t1.*f)(t2, ..., tN) when f is a pointer to a member function of a class T +// and t1 is an object of type T or a reference to an object of type T or a +// reference to an object of a type derived from T; +// 2. ((*t1).*f)(t2, ..., tN) when f is a pointer to a member function of a +// class T and t1 is not one of the types described in the previous item; +// 3. t1.*f when N == 1 and f is a pointer to member data of a class T and t1 is +// an object of type T or a reference to an object of type T or a reference +// to an object of a type derived from T; +// 4. (*t1).*f when N == 1 and f is a pointer to member data of a class T and t1 +// is not one of the types described in the previous item; +// 5. f(t1, t2, ..., tN) in all other cases. +// +// The implementation is SFINAE-friendly: substitution failure within invoke() +// isn't an error. + +#ifndef ABSL_BASE_INTERNAL_INVOKE_H_ +#define ABSL_BASE_INTERNAL_INVOKE_H_ + +#include "absl/base/config.h" + +#if ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L + +#include + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +using std::invoke; +using std::invoke_result_t; +using std::is_invocable_r; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#else // ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L + +#include +#include +#include + +#include "absl/meta/type_traits.h" + +// The following code is internal implementation detail. See the comment at the +// top of this file for the API documentation. + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// The five classes below each implement one of the clauses from the definition +// of INVOKE. The inner class template Accept checks whether the +// clause is applicable; static function template Invoke(f, args...) does the +// invocation. +// +// By separating the clause selection logic from invocation we make sure that +// Invoke() does exactly what the standard says. + +template +struct StrippedAccept { + template + struct Accept : Derived::template AcceptImpl::type>::type...> {}; +}; + +// (t1.*f)(t2, ..., tN) when f is a pointer to a member function of a class T +// and t1 is an object of type T or a reference to an object of type T or a +// reference to an object of a type derived from T. +struct MemFunAndRef : StrippedAccept { + template + struct AcceptImpl : std::false_type {}; + + template + struct AcceptImpl + : std::integral_constant::value && + absl::is_function::value> { + }; + + template + static decltype((std::declval().* + std::declval())(std::declval()...)) + Invoke(MemFun&& mem_fun, Obj&& obj, Args&&... args) { +// Ignore bogus GCC warnings on this line. +// See https://gcc.gnu.org/bugzilla/show_bug.cgi?id=101436 for similar example. +#if ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(11, 0) +// #pragma GCC diagnostic push +// #pragma GCC diagnostic ignored "-Warray-bounds" +// #pragma GCC diagnostic ignored "-Wmaybe-uninitialized" +#endif + return (std::forward(obj).* + std::forward(mem_fun))(std::forward(args)...); +#if ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(11, 0) +// #pragma GCC diagnostic pop +#endif + } +}; + +// ((*t1).*f)(t2, ..., tN) when f is a pointer to a member function of a +// class T and t1 is not one of the types described in the previous item. +struct MemFunAndPtr : StrippedAccept { + template + struct AcceptImpl : std::false_type {}; + + template + struct AcceptImpl + : std::integral_constant::value && + absl::is_function::value> { + }; + + template + static decltype(((*std::declval()).* + std::declval())(std::declval()...)) + Invoke(MemFun&& mem_fun, Ptr&& ptr, Args&&... args) { + return ((*std::forward(ptr)).* + std::forward(mem_fun))(std::forward(args)...); + } +}; + +// t1.*f when N == 1 and f is a pointer to member data of a class T and t1 is +// an object of type T or a reference to an object of type T or a reference +// to an object of a type derived from T. +struct DataMemAndRef : StrippedAccept { + template + struct AcceptImpl : std::false_type {}; + + template + struct AcceptImpl + : std::integral_constant::value && + !absl::is_function::value> {}; + + template + static decltype(std::declval().*std::declval()) Invoke( + DataMem&& data_mem, Ref&& ref) { + return std::forward(ref).*std::forward(data_mem); + } +}; + +// (*t1).*f when N == 1 and f is a pointer to member data of a class T and t1 +// is not one of the types described in the previous item. +struct DataMemAndPtr : StrippedAccept { + template + struct AcceptImpl : std::false_type {}; + + template + struct AcceptImpl + : std::integral_constant::value && + !absl::is_function::value> {}; + + template + static decltype((*std::declval()).*std::declval()) Invoke( + DataMem&& data_mem, Ptr&& ptr) { + return (*std::forward(ptr)).*std::forward(data_mem); + } +}; + +// f(t1, t2, ..., tN) in all other cases. +struct Callable { + // Callable doesn't have Accept because it's the last clause that gets picked + // when none of the previous clauses are applicable. + template + static decltype(std::declval()(std::declval()...)) Invoke( + F&& f, Args&&... args) { + return std::forward(f)(std::forward(args)...); + } +}; + +// Resolves to the first matching clause. +template +struct Invoker { + typedef typename std::conditional< + MemFunAndRef::Accept::value, MemFunAndRef, + typename std::conditional< + MemFunAndPtr::Accept::value, MemFunAndPtr, + typename std::conditional< + DataMemAndRef::Accept::value, DataMemAndRef, + typename std::conditional::value, + DataMemAndPtr, Callable>::type>::type>:: + type>::type type; +}; + +// The result type of Invoke. +template +using invoke_result_t = decltype(Invoker::type::Invoke( + std::declval(), std::declval()...)); + +// Invoke(f, args...) is an implementation of INVOKE(f, args...) from section +// [func.require] of the C++ standard. +template +invoke_result_t invoke(F&& f, Args&&... args) { + return Invoker::type::Invoke(std::forward(f), + std::forward(args)...); +} + +template +struct IsInvocableRImpl : std::false_type {}; + +template +struct IsInvocableRImpl< + absl::void_t >, R, F, + Args...> + : std::integral_constant< + bool, + std::is_convertible, + R>::value || + std::is_void::value> {}; + +// Type trait whose member `value` is true if invoking `F` with `Args` is valid, +// and either the return type is convertible to `R`, or `R` is void. +// C++11-compatible version of `std::is_invocable_r`. +template +using is_invocable_r = IsInvocableRImpl; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L + +#endif // ABSL_BASE_INTERNAL_INVOKE_H_ diff --git a/src/absl/base/internal/low_level_alloc.cc b/src/absl/base/internal/low_level_alloc.cc new file mode 100644 index 0000000..229ab91 --- /dev/null +++ b/src/absl/base/internal/low_level_alloc.cc @@ -0,0 +1,620 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// A low-level allocator that can be used by other low-level +// modules without introducing dependency cycles. +// This allocator is slow and wasteful of memory; +// it should not be used when performance is key. + +#include "absl/base/internal/low_level_alloc.h" + +#include + +#include "absl/base/call_once.h" +#include "absl/base/config.h" +#include "absl/base/internal/direct_mmap.h" +#include "absl/base/internal/scheduling_mode.h" +#include "absl/base/macros.h" +#include "absl/base/thread_annotations.h" + +// LowLevelAlloc requires that the platform support low-level +// allocation of virtual memory. Platforms lacking this cannot use +// LowLevelAlloc. +#ifndef ABSL_LOW_LEVEL_ALLOC_MISSING + +#ifndef _WIN32 +#include +#include +#include +#include +#else +#include +#endif + +#include +#include +#include +#include +#include +#include // for placement-new + +#include "absl/base/dynamic_annotations.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/spinlock.h" + +// MAP_ANONYMOUS +#if defined(__APPLE__) +// For mmap, Linux defines both MAP_ANONYMOUS and MAP_ANON and says MAP_ANON is +// deprecated. In Darwin, MAP_ANON is all there is. +#if !defined MAP_ANONYMOUS +#define MAP_ANONYMOUS MAP_ANON +#endif // !MAP_ANONYMOUS +#endif // __APPLE__ + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// A first-fit allocator with amortized logarithmic free() time. + +// --------------------------------------------------------------------------- +static const int kMaxLevel = 30; + +namespace { +// This struct describes one allocated block, or one free block. +struct AllocList { + struct Header { + // Size of entire region, including this field. Must be + // first. Valid in both allocated and unallocated blocks. + uintptr_t size; + + // kMagicAllocated or kMagicUnallocated xor this. + uintptr_t magic; + + // Pointer to parent arena. + LowLevelAlloc::Arena *arena; + + // Aligns regions to 0 mod 2*sizeof(void*). + void *dummy_for_alignment; + } header; + + // Next two fields: in unallocated blocks: freelist skiplist data + // in allocated blocks: overlaps with client data + + // Levels in skiplist used. + int levels; + + // Actually has levels elements. The AllocList node may not have room + // for all kMaxLevel entries. See max_fit in LLA_SkiplistLevels(). + AllocList *next[kMaxLevel]; +}; +} // namespace + +// --------------------------------------------------------------------------- +// A trivial skiplist implementation. This is used to keep the freelist +// in address order while taking only logarithmic time per insert and delete. + +// An integer approximation of log2(size/base) +// Requires size >= base. +static int IntLog2(size_t size, size_t base) { + int result = 0; + for (size_t i = size; i > base; i >>= 1) { // i == floor(size/2**result) + result++; + } + // floor(size / 2**result) <= base < floor(size / 2**(result-1)) + // => log2(size/(base+1)) <= result < 1+log2(size/base) + // => result ~= log2(size/base) + return result; +} + +// Return a random integer n: p(n)=1/(2**n) if 1 <= n; p(n)=0 if n < 1. +static int Random(uint32_t *state) { + uint32_t r = *state; + int result = 1; + while ((((r = r*1103515245 + 12345) >> 30) & 1) == 0) { + result++; + } + *state = r; + return result; +} + +// Return a number of skiplist levels for a node of size bytes, where +// base is the minimum node size. Compute level=log2(size / base)+n +// where n is 1 if random is false and otherwise a random number generated with +// the standard distribution for a skiplist: See Random() above. +// Bigger nodes tend to have more skiplist levels due to the log2(size / base) +// term, so first-fit searches touch fewer nodes. "level" is clipped so +// level(level) > max_fit) level = static_cast(max_fit); + if (level > kMaxLevel-1) level = kMaxLevel - 1; + ABSL_RAW_CHECK(level >= 1, "block not big enough for even one level"); + return level; +} + +// Return "atleast", the first element of AllocList *head s.t. *atleast >= *e. +// For 0 <= i < head->levels, set prev[i] to "no_greater", where no_greater +// points to the last element at level i in the AllocList less than *e, or is +// head if no such element exists. +static AllocList *LLA_SkiplistSearch(AllocList *head, + AllocList *e, AllocList **prev) { + AllocList *p = head; + for (int level = head->levels - 1; level >= 0; level--) { + for (AllocList *n; (n = p->next[level]) != nullptr && n < e; p = n) { + } + prev[level] = p; + } + return (head->levels == 0) ? nullptr : prev[0]->next[0]; +} + +// Insert element *e into AllocList *head. Set prev[] as LLA_SkiplistSearch. +// Requires that e->levels be previously set by the caller (using +// LLA_SkiplistLevels()) +static void LLA_SkiplistInsert(AllocList *head, AllocList *e, + AllocList **prev) { + LLA_SkiplistSearch(head, e, prev); + for (; head->levels < e->levels; head->levels++) { // extend prev pointers + prev[head->levels] = head; // to all *e's levels + } + for (int i = 0; i != e->levels; i++) { // add element to list + e->next[i] = prev[i]->next[i]; + prev[i]->next[i] = e; + } +} + +// Remove element *e from AllocList *head. Set prev[] as LLA_SkiplistSearch(). +// Requires that e->levels be previous set by the caller (using +// LLA_SkiplistLevels()) +static void LLA_SkiplistDelete(AllocList *head, AllocList *e, + AllocList **prev) { + AllocList *found = LLA_SkiplistSearch(head, e, prev); + ABSL_RAW_CHECK(e == found, "element not in freelist"); + for (int i = 0; i != e->levels && prev[i]->next[i] == e; i++) { + prev[i]->next[i] = e->next[i]; + } + while (head->levels > 0 && head->next[head->levels - 1] == nullptr) { + head->levels--; // reduce head->levels if level unused + } +} + +// --------------------------------------------------------------------------- +// Arena implementation + +// Metadata for an LowLevelAlloc arena instance. +struct LowLevelAlloc::Arena { + // Constructs an arena with the given LowLevelAlloc flags. + explicit Arena(uint32_t flags_value); + + base_internal::SpinLock mu; + // Head of free list, sorted by address + AllocList freelist ABSL_GUARDED_BY(mu); + // Count of allocated blocks + int32_t allocation_count ABSL_GUARDED_BY(mu); + // flags passed to NewArena + const uint32_t flags; + // Result of sysconf(_SC_PAGESIZE) + const size_t pagesize; + // Lowest power of two >= max(16, sizeof(AllocList)) + const size_t round_up; + // Smallest allocation block size + const size_t min_size; + // PRNG state + uint32_t random ABSL_GUARDED_BY(mu); +}; + +namespace { +// Static storage space for the lazily-constructed, default global arena +// instances. We require this space because the whole point of LowLevelAlloc +// is to avoid relying on malloc/new. +alignas(LowLevelAlloc::Arena) unsigned char default_arena_storage[sizeof( + LowLevelAlloc::Arena)]; +alignas(LowLevelAlloc::Arena) unsigned char unhooked_arena_storage[sizeof( + LowLevelAlloc::Arena)]; +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING +alignas( + LowLevelAlloc::Arena) unsigned char unhooked_async_sig_safe_arena_storage + [sizeof(LowLevelAlloc::Arena)]; +#endif + +// We must use LowLevelCallOnce here to construct the global arenas, rather than +// using function-level statics, to avoid recursively invoking the scheduler. +absl::once_flag create_globals_once; + +void CreateGlobalArenas() { + new (&default_arena_storage) + LowLevelAlloc::Arena(LowLevelAlloc::kCallMallocHook); + new (&unhooked_arena_storage) LowLevelAlloc::Arena(0); +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + new (&unhooked_async_sig_safe_arena_storage) + LowLevelAlloc::Arena(LowLevelAlloc::kAsyncSignalSafe); +#endif +} + +// Returns a global arena that does not call into hooks. Used by NewArena() +// when kCallMallocHook is not set. +LowLevelAlloc::Arena* UnhookedArena() { + base_internal::LowLevelCallOnce(&create_globals_once, CreateGlobalArenas); + return reinterpret_cast(&unhooked_arena_storage); +} + +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING +// Returns a global arena that is async-signal safe. Used by NewArena() when +// kAsyncSignalSafe is set. +LowLevelAlloc::Arena *UnhookedAsyncSigSafeArena() { + base_internal::LowLevelCallOnce(&create_globals_once, CreateGlobalArenas); + return reinterpret_cast( + &unhooked_async_sig_safe_arena_storage); +} +#endif + +} // namespace + +// Returns the default arena, as used by LowLevelAlloc::Alloc() and friends. +LowLevelAlloc::Arena *LowLevelAlloc::DefaultArena() { + base_internal::LowLevelCallOnce(&create_globals_once, CreateGlobalArenas); + return reinterpret_cast(&default_arena_storage); +} + +// magic numbers to identify allocated and unallocated blocks +static const uintptr_t kMagicAllocated = 0x4c833e95U; +static const uintptr_t kMagicUnallocated = ~kMagicAllocated; + +namespace { +class ABSL_SCOPED_LOCKABLE ArenaLock { + public: + explicit ArenaLock(LowLevelAlloc::Arena *arena) + ABSL_EXCLUSIVE_LOCK_FUNCTION(arena->mu) + : arena_(arena) { +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { + sigset_t all; + sigfillset(&all); + mask_valid_ = pthread_sigmask(SIG_BLOCK, &all, &mask_) == 0; + } +#endif + arena_->mu.Lock(); + } + ~ArenaLock() { ABSL_RAW_CHECK(left_, "haven't left Arena region"); } + void Leave() ABSL_UNLOCK_FUNCTION() { + arena_->mu.Unlock(); +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if (mask_valid_) { + const int err = pthread_sigmask(SIG_SETMASK, &mask_, nullptr); + if (err != 0) { + ABSL_RAW_LOG(FATAL, "pthread_sigmask failed: %d", err); + } + } +#endif + left_ = true; + } + + private: + bool left_ = false; // whether left region +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + bool mask_valid_ = false; + sigset_t mask_; // old mask of blocked signals +#endif + LowLevelAlloc::Arena *arena_; + ArenaLock(const ArenaLock &) = delete; + ArenaLock &operator=(const ArenaLock &) = delete; +}; +} // namespace + +// create an appropriate magic number for an object at "ptr" +// "magic" should be kMagicAllocated or kMagicUnallocated +inline static uintptr_t Magic(uintptr_t magic, AllocList::Header *ptr) { + return magic ^ reinterpret_cast(ptr); +} + +namespace { +size_t GetPageSize() { +#ifdef _WIN32 + SYSTEM_INFO system_info; + GetSystemInfo(&system_info); + return std::max(system_info.dwPageSize, system_info.dwAllocationGranularity); +#elif defined(__wasm__) || defined(__asmjs__) + return getpagesize(); +#else + return sysconf(_SC_PAGESIZE); +#endif +} + +size_t RoundedUpBlockSize() { + // Round up block sizes to a power of two close to the header size. + size_t round_up = 16; + while (round_up < sizeof(AllocList::Header)) { + round_up += round_up; + } + return round_up; +} + +} // namespace + +LowLevelAlloc::Arena::Arena(uint32_t flags_value) + : mu(base_internal::SCHEDULE_KERNEL_ONLY), + allocation_count(0), + flags(flags_value), + pagesize(GetPageSize()), + round_up(RoundedUpBlockSize()), + min_size(2 * round_up), + random(0) { + freelist.header.size = 0; + freelist.header.magic = + Magic(kMagicUnallocated, &freelist.header); + freelist.header.arena = this; + freelist.levels = 0; + memset(freelist.next, 0, sizeof(freelist.next)); +} + +// L < meta_data_arena->mu +LowLevelAlloc::Arena *LowLevelAlloc::NewArena(int32_t flags) { + Arena *meta_data_arena = DefaultArena(); +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if ((flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { + meta_data_arena = UnhookedAsyncSigSafeArena(); + } else // NOLINT(readability/braces) +#endif + if ((flags & LowLevelAlloc::kCallMallocHook) == 0) { + meta_data_arena = UnhookedArena(); + } + Arena *result = + new (AllocWithArena(sizeof (*result), meta_data_arena)) Arena(flags); + return result; +} + +// L < arena->mu, L < arena->arena->mu +bool LowLevelAlloc::DeleteArena(Arena *arena) { + ABSL_RAW_CHECK( + arena != nullptr && arena != DefaultArena() && arena != UnhookedArena(), + "may not delete default arena"); + ArenaLock section(arena); + if (arena->allocation_count != 0) { + section.Leave(); + return false; + } + while (arena->freelist.next[0] != nullptr) { + AllocList *region = arena->freelist.next[0]; + size_t size = region->header.size; + arena->freelist.next[0] = region->next[0]; + ABSL_RAW_CHECK( + region->header.magic == Magic(kMagicUnallocated, ®ion->header), + "bad magic number in DeleteArena()"); + ABSL_RAW_CHECK(region->header.arena == arena, + "bad arena pointer in DeleteArena()"); + ABSL_RAW_CHECK(size % arena->pagesize == 0, + "empty arena has non-page-aligned block size"); + ABSL_RAW_CHECK(reinterpret_cast(region) % arena->pagesize == 0, + "empty arena has non-page-aligned block"); + int munmap_result; +#ifdef _WIN32 + munmap_result = VirtualFree(region, 0, MEM_RELEASE); + ABSL_RAW_CHECK(munmap_result != 0, + "LowLevelAlloc::DeleteArena: VitualFree failed"); +#else +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) == 0) { + munmap_result = munmap(region, size); + } else { + munmap_result = base_internal::DirectMunmap(region, size); + } +#else + munmap_result = munmap(region, size); +#endif // ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if (munmap_result != 0) { + ABSL_RAW_LOG(FATAL, "LowLevelAlloc::DeleteArena: munmap failed: %d", + errno); + } +#endif // _WIN32 + } + section.Leave(); + arena->~Arena(); + Free(arena); + return true; +} + +// --------------------------------------------------------------------------- + +// Addition, checking for overflow. The intent is to die if an external client +// manages to push through a request that would cause arithmetic to fail. +static inline uintptr_t CheckedAdd(uintptr_t a, uintptr_t b) { + uintptr_t sum = a + b; + ABSL_RAW_CHECK(sum >= a, "LowLevelAlloc arithmetic overflow"); + return sum; +} + +// Return value rounded up to next multiple of align. +// align must be a power of two. +static inline uintptr_t RoundUp(uintptr_t addr, uintptr_t align) { + return CheckedAdd(addr, align - 1) & ~(align - 1); +} + +// Equivalent to "return prev->next[i]" but with sanity checking +// that the freelist is in the correct order, that it +// consists of regions marked "unallocated", and that no two regions +// are adjacent in memory (they should have been coalesced). +// L >= arena->mu +static AllocList *Next(int i, AllocList *prev, LowLevelAlloc::Arena *arena) { + ABSL_RAW_CHECK(i < prev->levels, "too few levels in Next()"); + AllocList *next = prev->next[i]; + if (next != nullptr) { + ABSL_RAW_CHECK( + next->header.magic == Magic(kMagicUnallocated, &next->header), + "bad magic number in Next()"); + ABSL_RAW_CHECK(next->header.arena == arena, "bad arena pointer in Next()"); + if (prev != &arena->freelist) { + ABSL_RAW_CHECK(prev < next, "unordered freelist"); + ABSL_RAW_CHECK(reinterpret_cast(prev) + prev->header.size < + reinterpret_cast(next), + "malformed freelist"); + } + } + return next; +} + +// Coalesce list item "a" with its successor if they are adjacent. +static void Coalesce(AllocList *a) { + AllocList *n = a->next[0]; + if (n != nullptr && reinterpret_cast(a) + a->header.size == + reinterpret_cast(n)) { + LowLevelAlloc::Arena *arena = a->header.arena; + a->header.size += n->header.size; + n->header.magic = 0; + n->header.arena = nullptr; + AllocList *prev[kMaxLevel]; + LLA_SkiplistDelete(&arena->freelist, n, prev); + LLA_SkiplistDelete(&arena->freelist, a, prev); + a->levels = LLA_SkiplistLevels(a->header.size, arena->min_size, + &arena->random); + LLA_SkiplistInsert(&arena->freelist, a, prev); + } +} + +// Adds block at location "v" to the free list +// L >= arena->mu +static void AddToFreelist(void *v, LowLevelAlloc::Arena *arena) { + AllocList *f = reinterpret_cast( + reinterpret_cast(v) - sizeof (f->header)); + ABSL_RAW_CHECK(f->header.magic == Magic(kMagicAllocated, &f->header), + "bad magic number in AddToFreelist()"); + ABSL_RAW_CHECK(f->header.arena == arena, + "bad arena pointer in AddToFreelist()"); + f->levels = LLA_SkiplistLevels(f->header.size, arena->min_size, + &arena->random); + AllocList *prev[kMaxLevel]; + LLA_SkiplistInsert(&arena->freelist, f, prev); + f->header.magic = Magic(kMagicUnallocated, &f->header); + Coalesce(f); // maybe coalesce with successor + Coalesce(prev[0]); // maybe coalesce with predecessor +} + +// Frees storage allocated by LowLevelAlloc::Alloc(). +// L < arena->mu +void LowLevelAlloc::Free(void *v) { + if (v != nullptr) { + AllocList *f = reinterpret_cast( + reinterpret_cast(v) - sizeof (f->header)); + LowLevelAlloc::Arena *arena = f->header.arena; + ArenaLock section(arena); + AddToFreelist(v, arena); + ABSL_RAW_CHECK(arena->allocation_count > 0, "nothing in arena to free"); + arena->allocation_count--; + section.Leave(); + } +} + +// allocates and returns a block of size bytes, to be freed with Free() +// L < arena->mu +static void *DoAllocWithArena(size_t request, LowLevelAlloc::Arena *arena) { + void *result = nullptr; + if (request != 0) { + AllocList *s; // will point to region that satisfies request + ArenaLock section(arena); + // round up with header + size_t req_rnd = RoundUp(CheckedAdd(request, sizeof (s->header)), + arena->round_up); + for (;;) { // loop until we find a suitable region + // find the minimum levels that a block of this size must have + int i = LLA_SkiplistLevels(req_rnd, arena->min_size, nullptr) - 1; + if (i < arena->freelist.levels) { // potential blocks exist + AllocList *before = &arena->freelist; // predecessor of s + while ((s = Next(i, before, arena)) != nullptr && + s->header.size < req_rnd) { + before = s; + } + if (s != nullptr) { // we found a region + break; + } + } + // we unlock before mmap() both because mmap() may call a callback hook, + // and because it may be slow. + arena->mu.Unlock(); + // mmap generous 64K chunks to decrease + // the chances/impact of fragmentation: + size_t new_pages_size = RoundUp(req_rnd, arena->pagesize * 16); + void *new_pages; +#ifdef _WIN32 + new_pages = VirtualAlloc(0, new_pages_size, + MEM_RESERVE | MEM_COMMIT, PAGE_READWRITE); + ABSL_RAW_CHECK(new_pages != nullptr, "VirtualAlloc failed"); +#else +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { + new_pages = base_internal::DirectMmap(nullptr, new_pages_size, + PROT_WRITE|PROT_READ, MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); + } else { + new_pages = mmap(nullptr, new_pages_size, PROT_WRITE | PROT_READ, + MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); + } +#else + new_pages = mmap(nullptr, new_pages_size, PROT_WRITE | PROT_READ, + MAP_ANONYMOUS | MAP_PRIVATE, -1, 0); +#endif // ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + if (new_pages == MAP_FAILED) { + ABSL_RAW_LOG(FATAL, "mmap error: %d", errno); + } + +#endif // _WIN32 + arena->mu.Lock(); + s = reinterpret_cast(new_pages); + s->header.size = new_pages_size; + // Pretend the block is allocated; call AddToFreelist() to free it. + s->header.magic = Magic(kMagicAllocated, &s->header); + s->header.arena = arena; + AddToFreelist(&s->levels, arena); // insert new region into free list + } + AllocList *prev[kMaxLevel]; + LLA_SkiplistDelete(&arena->freelist, s, prev); // remove from free list + // s points to the first free region that's big enough + if (CheckedAdd(req_rnd, arena->min_size) <= s->header.size) { + // big enough to split + AllocList *n = reinterpret_cast + (req_rnd + reinterpret_cast(s)); + n->header.size = s->header.size - req_rnd; + n->header.magic = Magic(kMagicAllocated, &n->header); + n->header.arena = arena; + s->header.size = req_rnd; + AddToFreelist(&n->levels, arena); + } + s->header.magic = Magic(kMagicAllocated, &s->header); + ABSL_RAW_CHECK(s->header.arena == arena, ""); + arena->allocation_count++; + section.Leave(); + result = &s->levels; + } + ABSL_ANNOTATE_MEMORY_IS_UNINITIALIZED(result, request); + return result; +} + +void *LowLevelAlloc::Alloc(size_t request) { + void *result = DoAllocWithArena(request, DefaultArena()); + return result; +} + +void *LowLevelAlloc::AllocWithArena(size_t request, Arena *arena) { + ABSL_RAW_CHECK(arena != nullptr, "must pass a valid arena"); + void *result = DoAllocWithArena(request, arena); + return result; +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_LOW_LEVEL_ALLOC_MISSING diff --git a/src/absl/base/internal/low_level_alloc.h b/src/absl/base/internal/low_level_alloc.h new file mode 100644 index 0000000..daf3ebe --- /dev/null +++ b/src/absl/base/internal/low_level_alloc.h @@ -0,0 +1,130 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_LOW_LEVEL_ALLOC_H_ +#define ABSL_BASE_INTERNAL_LOW_LEVEL_ALLOC_H_ + +#if defined(Free) +#undef Free +#endif + +// A simple thread-safe memory allocator that does not depend on +// mutexes or thread-specific data. It is intended to be used +// sparingly, and only when malloc() would introduce an unwanted +// dependency, such as inside the heap-checker, or the Mutex +// implementation. + +// IWYU pragma: private, include "base/low_level_alloc.h" + +#include + +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" + +// LowLevelAlloc requires that the platform support low-level +// allocation of virtual memory. Platforms lacking this cannot use +// LowLevelAlloc. +#ifdef ABSL_LOW_LEVEL_ALLOC_MISSING +#error ABSL_LOW_LEVEL_ALLOC_MISSING cannot be directly set +#elif !defined(ABSL_HAVE_MMAP) && !defined(_WIN32) +#define ABSL_LOW_LEVEL_ALLOC_MISSING 1 +#endif + +// Using LowLevelAlloc with kAsyncSignalSafe isn't supported on Windows or +// asm.js / WebAssembly. +// See https://kripken.github.io/emscripten-site/docs/porting/pthreads.html +// for more information. +#ifdef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING +#error ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING cannot be directly set +#elif defined(_WIN32) || defined(__asmjs__) || defined(__wasm__) +#define ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING 1 +#endif + +#include + +#include "absl/base/port.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +class LowLevelAlloc { + public: + struct Arena; // an arena from which memory may be allocated + + // Returns a pointer to a block of at least "request" bytes + // that have been newly allocated from the specific arena. + // for Alloc() call the DefaultArena() is used. + // Returns 0 if passed request==0. + // Does not return 0 under other circumstances; it crashes if memory + // is not available. + static void *Alloc(size_t request) ABSL_ATTRIBUTE_SECTION(malloc_hook); + static void *AllocWithArena(size_t request, Arena *arena) + ABSL_ATTRIBUTE_SECTION(malloc_hook); + + // Deallocates a region of memory that was previously allocated with + // Alloc(). Does nothing if passed 0. "s" must be either 0, + // or must have been returned from a call to Alloc() and not yet passed to + // Free() since that call to Alloc(). The space is returned to the arena + // from which it was allocated. + static void Free(void *s) ABSL_ATTRIBUTE_SECTION(malloc_hook); + + // ABSL_ATTRIBUTE_SECTION(malloc_hook) for Alloc* and Free + // are to put all callers of MallocHook::Invoke* in this module + // into special section, + // so that MallocHook::GetCallerStackTrace can function accurately. + + // Create a new arena. + // The root metadata for the new arena is allocated in the + // meta_data_arena; the DefaultArena() can be passed for meta_data_arena. + // These values may be ored into flags: + enum { + // Report calls to Alloc() and Free() via the MallocHook interface. + // Set in the DefaultArena. + kCallMallocHook = 0x0001, + +#ifndef ABSL_LOW_LEVEL_ALLOC_ASYNC_SIGNAL_SAFE_MISSING + // Make calls to Alloc(), Free() be async-signal-safe. Not set in + // DefaultArena(). Not supported on all platforms. + kAsyncSignalSafe = 0x0002, +#endif + }; + // Construct a new arena. The allocation of the underlying metadata honors + // the provided flags. For example, the call NewArena(kAsyncSignalSafe) + // is itself async-signal-safe, as well as generatating an arena that provides + // async-signal-safe Alloc/Free. + static Arena *NewArena(int32_t flags); + + // Destroys an arena allocated by NewArena and returns true, + // provided no allocated blocks remain in the arena. + // If allocated blocks remain in the arena, does nothing and + // returns false. + // It is illegal to attempt to destroy the DefaultArena(). + static bool DeleteArena(Arena *arena); + + // The default arena that always exists. + static Arena *DefaultArena(); + + private: + LowLevelAlloc(); // no instances +}; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_LOW_LEVEL_ALLOC_H_ diff --git a/src/absl/base/internal/low_level_scheduling.h b/src/absl/base/internal/low_level_scheduling.h new file mode 100644 index 0000000..9baccc0 --- /dev/null +++ b/src/absl/base/internal/low_level_scheduling.h @@ -0,0 +1,134 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Core interfaces and definitions used by by low-level interfaces such as +// SpinLock. + +#ifndef ABSL_BASE_INTERNAL_LOW_LEVEL_SCHEDULING_H_ +#define ABSL_BASE_INTERNAL_LOW_LEVEL_SCHEDULING_H_ + +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/scheduling_mode.h" +#include "absl/base/macros.h" + +// The following two declarations exist so SchedulingGuard may friend them with +// the appropriate language linkage. These callbacks allow libc internals, such +// as function level statics, to schedule cooperatively when locking. +extern "C" bool __google_disable_rescheduling(void); +extern "C" void __google_enable_rescheduling(bool disable_result); + +namespace absl { +ABSL_NAMESPACE_BEGIN +class CondVar; +class Mutex; + +namespace synchronization_internal { +int MutexDelay(int32_t c, int mode); +} // namespace synchronization_internal + +namespace base_internal { + +class SchedulingHelper; // To allow use of SchedulingGuard. +class SpinLock; // To allow use of SchedulingGuard. + +// SchedulingGuard +// Provides guard semantics that may be used to disable cooperative rescheduling +// of the calling thread within specific program blocks. This is used to +// protect resources (e.g. low-level SpinLocks or Domain code) that cooperative +// scheduling depends on. +// +// Domain implementations capable of rescheduling in reaction to involuntary +// kernel thread actions (e.g blocking due to a pagefault or syscall) must +// guarantee that an annotated thread is not allowed to (cooperatively) +// reschedule until the annotated region is complete. +// +// It is an error to attempt to use a cooperatively scheduled resource (e.g. +// Mutex) within a rescheduling-disabled region. +// +// All methods are async-signal safe. +class SchedulingGuard { + public: + // Returns true iff the calling thread may be cooperatively rescheduled. + static bool ReschedulingIsAllowed(); + SchedulingGuard(const SchedulingGuard&) = delete; + SchedulingGuard& operator=(const SchedulingGuard&) = delete; + + private: + // Disable cooperative rescheduling of the calling thread. It may still + // initiate scheduling operations (e.g. wake-ups), however, it may not itself + // reschedule. Nestable. The returned result is opaque, clients should not + // attempt to interpret it. + // REQUIRES: Result must be passed to a pairing EnableScheduling(). + static bool DisableRescheduling(); + + // Marks the end of a rescheduling disabled region, previously started by + // DisableRescheduling(). + // REQUIRES: Pairs with innermost call (and result) of DisableRescheduling(). + static void EnableRescheduling(bool disable_result); + + // A scoped helper for {Disable, Enable}Rescheduling(). + // REQUIRES: destructor must run in same thread as constructor. + struct ScopedDisable { + ScopedDisable() { disabled = SchedulingGuard::DisableRescheduling(); } + ~ScopedDisable() { SchedulingGuard::EnableRescheduling(disabled); } + + bool disabled; + }; + + // A scoped helper to enable rescheduling temporarily. + // REQUIRES: destructor must run in same thread as constructor. + class ScopedEnable { + public: + ScopedEnable(); + ~ScopedEnable(); + + private: + int scheduling_disabled_depth_; + }; + + // Access to SchedulingGuard is explicitly permitted. + friend class absl::CondVar; + friend class absl::Mutex; + friend class SchedulingHelper; + friend class SpinLock; + friend int absl::synchronization_internal::MutexDelay(int32_t c, int mode); +}; + +//------------------------------------------------------------------------------ +// End of public interfaces. +//------------------------------------------------------------------------------ + +inline bool SchedulingGuard::ReschedulingIsAllowed() { + return false; +} + +inline bool SchedulingGuard::DisableRescheduling() { + return false; +} + +inline void SchedulingGuard::EnableRescheduling(bool /* disable_result */) { + return; +} + +inline SchedulingGuard::ScopedEnable::ScopedEnable() + : scheduling_disabled_depth_(0) {} +inline SchedulingGuard::ScopedEnable::~ScopedEnable() { + ABSL_RAW_CHECK(scheduling_disabled_depth_ == 0, "disable unused warning"); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_LOW_LEVEL_SCHEDULING_H_ diff --git a/src/absl/base/internal/per_thread_tls.h b/src/absl/base/internal/per_thread_tls.h new file mode 100644 index 0000000..cf5e97a --- /dev/null +++ b/src/absl/base/internal/per_thread_tls.h @@ -0,0 +1,52 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_PER_THREAD_TLS_H_ +#define ABSL_BASE_INTERNAL_PER_THREAD_TLS_H_ + +// This header defines two macros: +// +// If the platform supports thread-local storage: +// +// * ABSL_PER_THREAD_TLS_KEYWORD is the C keyword needed to declare a +// thread-local variable +// * ABSL_PER_THREAD_TLS is 1 +// +// Otherwise: +// +// * ABSL_PER_THREAD_TLS_KEYWORD is empty +// * ABSL_PER_THREAD_TLS is 0 +// +// Microsoft C supports thread-local storage. +// GCC supports it if the appropriate version of glibc is available, +// which the programmer can indicate by defining ABSL_HAVE_TLS + +#include "absl/base/port.h" // For ABSL_HAVE_TLS + +#if defined(ABSL_PER_THREAD_TLS) +#error ABSL_PER_THREAD_TLS cannot be directly set +#elif defined(ABSL_PER_THREAD_TLS_KEYWORD) +#error ABSL_PER_THREAD_TLS_KEYWORD cannot be directly set +#elif defined(ABSL_HAVE_TLS) +#define ABSL_PER_THREAD_TLS_KEYWORD __thread +#define ABSL_PER_THREAD_TLS 1 +#elif defined(_MSC_VER) +#define ABSL_PER_THREAD_TLS_KEYWORD __declspec(thread) +#define ABSL_PER_THREAD_TLS 1 +#else +#define ABSL_PER_THREAD_TLS_KEYWORD +#define ABSL_PER_THREAD_TLS 0 +#endif + +#endif // ABSL_BASE_INTERNAL_PER_THREAD_TLS_H_ diff --git a/src/absl/base/internal/prefetch.h b/src/absl/base/internal/prefetch.h new file mode 100644 index 0000000..0641928 --- /dev/null +++ b/src/absl/base/internal/prefetch.h @@ -0,0 +1,138 @@ +// Copyright 2022 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_PREFETCH_H_ +#define ABSL_BASE_INTERNAL_PREFETCH_H_ + +#include "absl/base/config.h" + +#ifdef __SSE__ +#include +#endif + +#if defined(_MSC_VER) && defined(ABSL_INTERNAL_HAVE_SSE) +#include +#pragma intrinsic(_mm_prefetch) +#endif + +// Compatibility wrappers around __builtin_prefetch, to prefetch data +// for read if supported by the toolchain. + +// Move data into the cache before it is read, or "prefetch" it. +// +// The value of `addr` is the address of the memory to prefetch. If +// the target and compiler support it, data prefetch instructions are +// generated. If the prefetch is done some time before the memory is +// read, it may be in the cache by the time the read occurs. +// +// The function names specify the temporal locality heuristic applied, +// using the names of Intel prefetch instructions: +// +// T0 - high degree of temporal locality; data should be left in as +// many levels of the cache possible +// T1 - moderate degree of temporal locality +// T2 - low degree of temporal locality +// Nta - no temporal locality, data need not be left in the cache +// after the read +// +// Incorrect or gratuitous use of these functions can degrade +// performance, so use them only when representative benchmarks show +// an improvement. +// +// Example usage: +// +// absl::base_internal::PrefetchT0(addr); +// +// Currently, the different prefetch calls behave on some Intel +// architectures as follows: +// +// SNB..SKL SKX +// PrefetchT0() L1/L2/L3 L1/L2 +// PrefetchT1() L2/L3 L2 +// PrefetchT2() L2/L3 L2 +// PrefetchNta() L1/--/L3 L1* +// +// * On SKX PrefetchNta() will bring the line into L1 but will evict +// from L3 cache. This might result in surprising behavior. +// +// SNB = Sandy Bridge, SKL = Skylake, SKX = Skylake Xeon. +// +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +void PrefetchT0(const void* addr); +void PrefetchT1(const void* addr); +void PrefetchT2(const void* addr); +void PrefetchNta(const void* addr); + +// Implementation details follow. + +#if ABSL_HAVE_BUILTIN(__builtin_prefetch) || defined(__GNUC__) + +#define ABSL_INTERNAL_HAVE_PREFETCH 1 + +// See __builtin_prefetch: +// https://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html. +// +// These functions speculatively load for read only. This is +// safe for all currently supported platforms. However, prefetch for +// store may have problems depending on the target platform. +// +inline void PrefetchT0(const void* addr) { + // Note: this uses prefetcht0 on Intel. + __builtin_prefetch(addr, 0, 3); +} +inline void PrefetchT1(const void* addr) { + // Note: this uses prefetcht1 on Intel. + __builtin_prefetch(addr, 0, 2); +} +inline void PrefetchT2(const void* addr) { + // Note: this uses prefetcht2 on Intel. + __builtin_prefetch(addr, 0, 1); +} +inline void PrefetchNta(const void* addr) { + // Note: this uses prefetchtnta on Intel. + __builtin_prefetch(addr, 0, 0); +} + +#elif defined(ABSL_INTERNAL_HAVE_SSE) + +#define ABSL_INTERNAL_HAVE_PREFETCH 1 + +inline void PrefetchT0(const void* addr) { + _mm_prefetch(reinterpret_cast(addr), _MM_HINT_T0); +} +inline void PrefetchT1(const void* addr) { + _mm_prefetch(reinterpret_cast(addr), _MM_HINT_T1); +} +inline void PrefetchT2(const void* addr) { + _mm_prefetch(reinterpret_cast(addr), _MM_HINT_T2); +} +inline void PrefetchNta(const void* addr) { + _mm_prefetch(reinterpret_cast(addr), _MM_HINT_NTA); +} + +#else +inline void PrefetchT0(const void*) {} +inline void PrefetchT1(const void*) {} +inline void PrefetchT2(const void*) {} +inline void PrefetchNta(const void*) {} +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_PREFETCH_H_ diff --git a/src/absl/base/internal/pretty_function.h b/src/absl/base/internal/pretty_function.h new file mode 100644 index 0000000..35d5167 --- /dev/null +++ b/src/absl/base/internal/pretty_function.h @@ -0,0 +1,33 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_PRETTY_FUNCTION_H_ +#define ABSL_BASE_INTERNAL_PRETTY_FUNCTION_H_ + +// ABSL_PRETTY_FUNCTION +// +// In C++11, __func__ gives the undecorated name of the current function. That +// is, "main", not "int main()". Various compilers give extra macros to get the +// decorated function name, including return type and arguments, to +// differentiate between overload sets. ABSL_PRETTY_FUNCTION is a portable +// version of these macros which forwards to the correct macro on each compiler. +#if defined(_MSC_VER) +#define ABSL_PRETTY_FUNCTION __FUNCSIG__ +#elif defined(__GNUC__) +#define ABSL_PRETTY_FUNCTION __PRETTY_FUNCTION__ +#else +#error "Unsupported compiler" +#endif + +#endif // ABSL_BASE_INTERNAL_PRETTY_FUNCTION_H_ diff --git a/src/absl/base/internal/raw_logging.cc b/src/absl/base/internal/raw_logging.cc new file mode 100644 index 0000000..8092687 --- /dev/null +++ b/src/absl/base/internal/raw_logging.cc @@ -0,0 +1,249 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/raw_logging.h" + +#include +#include +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/internal/atomic_hook.h" +#include "absl/base/internal/errno_saver.h" +#include "absl/base/log_severity.h" + +// We know how to perform low-level writes to stderr in POSIX and Windows. For +// these platforms, we define the token ABSL_LOW_LEVEL_WRITE_SUPPORTED. +// Much of raw_logging.cc becomes a no-op when we can't output messages, +// although a FATAL ABSL_RAW_LOG message will still abort the process. + +// ABSL_HAVE_POSIX_WRITE is defined when the platform provides posix write() +// (as from unistd.h) +// +// This preprocessor token is also defined in raw_io.cc. If you need to copy +// this, consider moving both to config.h instead. +#if defined(__linux__) || defined(__APPLE__) || defined(__FreeBSD__) || \ + defined(__Fuchsia__) || defined(__native_client__) || \ + defined(__OpenBSD__) || defined(__EMSCRIPTEN__) || defined(__ASYLO__) + +#include + +#define ABSL_HAVE_POSIX_WRITE 1 +#define ABSL_LOW_LEVEL_WRITE_SUPPORTED 1 +#else +#undef ABSL_HAVE_POSIX_WRITE +#endif + +// ABSL_HAVE_SYSCALL_WRITE is defined when the platform provides the syscall +// syscall(SYS_write, /*int*/ fd, /*char* */ buf, /*size_t*/ len); +// for low level operations that want to avoid libc. +#if (defined(__linux__) || defined(__FreeBSD__) || defined(__OpenBSD__)) && \ + !defined(__ANDROID__) +#include +#define ABSL_HAVE_SYSCALL_WRITE 1 +#define ABSL_LOW_LEVEL_WRITE_SUPPORTED 1 +#else +#undef ABSL_HAVE_SYSCALL_WRITE +#endif + +#ifdef _WIN32 +#include + +#define ABSL_HAVE_RAW_IO 1 +#define ABSL_LOW_LEVEL_WRITE_SUPPORTED 1 +#else +#undef ABSL_HAVE_RAW_IO +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace raw_logging_internal { +namespace { + +// TODO(gfalcon): We want raw-logging to work on as many platforms as possible. +// Explicitly `#error` out when not `ABSL_LOW_LEVEL_WRITE_SUPPORTED`, except for +// a selected set of platforms for which we expect not to be able to raw log. + +#ifdef ABSL_LOW_LEVEL_WRITE_SUPPORTED +constexpr char kTruncated[] = " ... (message truncated)\n"; + +// sprintf the format to the buffer, adjusting *buf and *size to reflect the +// consumed bytes, and return whether the message fit without truncation. If +// truncation occurred, if possible leave room in the buffer for the message +// kTruncated[]. +bool VADoRawLog(char** buf, int* size, const char* format, va_list ap) + ABSL_PRINTF_ATTRIBUTE(3, 0); +bool VADoRawLog(char** buf, int* size, const char* format, va_list ap) { + int n = vsnprintf(*buf, *size, format, ap); + bool result = true; + if (n < 0 || n > *size) { + result = false; + if (static_cast(*size) > sizeof(kTruncated)) { + n = *size - sizeof(kTruncated); // room for truncation message + } else { + n = 0; // no room for truncation message + } + } + *size -= n; + *buf += n; + return result; +} +#endif // ABSL_LOW_LEVEL_WRITE_SUPPORTED + +constexpr int kLogBufSize = 3000; + +// CAVEAT: vsnprintf called from *DoRawLog below has some (exotic) code paths +// that invoke malloc() and getenv() that might acquire some locks. + +// Helper for RawLog below. +// *DoRawLog writes to *buf of *size and move them past the written portion. +// It returns true iff there was no overflow or error. +bool DoRawLog(char** buf, int* size, const char* format, ...) + ABSL_PRINTF_ATTRIBUTE(3, 4); +bool DoRawLog(char** buf, int* size, const char* format, ...) { + va_list ap; + va_start(ap, format); + int n = vsnprintf(*buf, *size, format, ap); + va_end(ap); + if (n < 0 || n > *size) return false; + *size -= n; + *buf += n; + return true; +} + +bool DefaultLogFilterAndPrefix(absl::LogSeverity, const char* file, int line, + char** buf, int* buf_size) { + DoRawLog(buf, buf_size, "[%s : %d] RAW: ", file, line); + return true; +} + +ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES +absl::base_internal::AtomicHook + log_filter_and_prefix_hook(DefaultLogFilterAndPrefix); +ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES +absl::base_internal::AtomicHook abort_hook; + +void RawLogVA(absl::LogSeverity severity, const char* file, int line, + const char* format, va_list ap) ABSL_PRINTF_ATTRIBUTE(4, 0); +void RawLogVA(absl::LogSeverity severity, const char* file, int line, + const char* format, va_list ap) { + char buffer[kLogBufSize]; + char* buf = buffer; + int size = sizeof(buffer); +#ifdef ABSL_LOW_LEVEL_WRITE_SUPPORTED + bool enabled = true; +#else + bool enabled = false; +#endif + +#ifdef ABSL_MIN_LOG_LEVEL + if (severity < static_cast(ABSL_MIN_LOG_LEVEL) && + severity < absl::LogSeverity::kFatal) { + enabled = false; + } +#endif + + enabled = log_filter_and_prefix_hook(severity, file, line, &buf, &size); + const char* const prefix_end = buf; + +#ifdef ABSL_LOW_LEVEL_WRITE_SUPPORTED + if (enabled) { + bool no_chop = VADoRawLog(&buf, &size, format, ap); + if (no_chop) { + DoRawLog(&buf, &size, "\n"); + } else { + DoRawLog(&buf, &size, "%s", kTruncated); + } + AsyncSignalSafeWriteToStderr(buffer, strlen(buffer)); + } +#else + static_cast(format); + static_cast(ap); + static_cast(enabled); +#endif + + // Abort the process after logging a FATAL message, even if the output itself + // was suppressed. + if (severity == absl::LogSeverity::kFatal) { + abort_hook(file, line, buffer, prefix_end, buffer + kLogBufSize); + throw std::runtime_error("abort()"); + } +} + +// Non-formatting version of RawLog(). +// +// TODO(gfalcon): When string_view no longer depends on base, change this +// interface to take its message as a string_view instead. +void DefaultInternalLog(absl::LogSeverity severity, const char* file, int line, + const std::string& message) { + RawLog(severity, file, line, "%.*s", static_cast(message.size()), + message.data()); +} + +} // namespace + +void AsyncSignalSafeWriteToStderr(const char* s, size_t len) { + absl::base_internal::ErrnoSaver errno_saver; +#if defined(ABSL_HAVE_SYSCALL_WRITE) + // We prefer calling write via `syscall` to minimize the risk of libc doing + // something "helpful". + syscall(SYS_write, STDERR_FILENO, s, len); +#elif defined(ABSL_HAVE_POSIX_WRITE) + write(STDERR_FILENO, s, len); +#elif defined(ABSL_HAVE_RAW_IO) + _write(/* stderr */ 2, s, len); +#else + // stderr logging unsupported on this platform + (void) s; + (void) len; +#endif +} + +void RawLog(absl::LogSeverity severity, const char* file, int line, + const char* format, ...) { + va_list ap; + va_start(ap, format); + RawLogVA(severity, file, line, format, ap); + va_end(ap); +} + +bool RawLoggingFullySupported() { +#ifdef ABSL_LOW_LEVEL_WRITE_SUPPORTED + return true; +#else // !ABSL_LOW_LEVEL_WRITE_SUPPORTED + return false; +#endif // !ABSL_LOW_LEVEL_WRITE_SUPPORTED +} + +ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES ABSL_DLL + absl::base_internal::AtomicHook + internal_log_function(DefaultInternalLog); + +void RegisterLogFilterAndPrefixHook(LogFilterAndPrefixHook func) { + log_filter_and_prefix_hook.Store(func); +} + +void RegisterAbortHook(AbortHook func) { abort_hook.Store(func); } + +void RegisterInternalLogFunction(InternalLogFunction func) { + internal_log_function.Store(func); +} + +} // namespace raw_logging_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/raw_logging.h b/src/absl/base/internal/raw_logging.h new file mode 100644 index 0000000..fa36f94 --- /dev/null +++ b/src/absl/base/internal/raw_logging.h @@ -0,0 +1,198 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Thread-safe logging routines that do not allocate any memory or +// acquire any locks, and can therefore be used by low-level memory +// allocation, synchronization, and signal-handling code. + +#ifndef ABSL_BASE_INTERNAL_RAW_LOGGING_H_ +#define ABSL_BASE_INTERNAL_RAW_LOGGING_H_ + +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/internal/atomic_hook.h" +#include "absl/base/log_severity.h" +#include "absl/base/macros.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" + +// This is similar to LOG(severity) << format..., but +// * it is to be used ONLY by low-level modules that can't use normal LOG() +// * it is designed to be a low-level logger that does not allocate any +// memory and does not need any locks, hence: +// * it logs straight and ONLY to STDERR w/o buffering +// * it uses an explicit printf-format and arguments list +// * it will silently chop off really long message strings +// Usage example: +// ABSL_RAW_LOG(ERROR, "Failed foo with %i: %s", status, error); +// This will print an almost standard log line like this to stderr only: +// E0821 211317 file.cc:123] RAW: Failed foo with 22: bad_file + +#define ABSL_RAW_LOG(severity, ...) \ + // R/Windows gives compilation warnings here that fail the CMD + // check. + // do { \ + // constexpr const char* absl_raw_logging_internal_basename = \ + // ::absl::raw_logging_internal::Basename(__FILE__, \ + // sizeof(__FILE__) - 1); \ + // ::absl::raw_logging_internal::RawLog(ABSL_RAW_LOGGING_INTERNAL_##severity, \ + // absl_raw_logging_internal_basename, \ + // __LINE__, __VA_ARGS__); \ + // } while (0) + +// Similar to CHECK(condition) << message, but for low-level modules: +// we use only ABSL_RAW_LOG that does not allocate memory. +// We do not want to provide args list here to encourage this usage: +// if (!cond) ABSL_RAW_LOG(FATAL, "foo ...", hard_to_compute_args); +// so that the args are not computed when not needed. +#define ABSL_RAW_CHECK(condition, message) \ + do { \ + if (ABSL_PREDICT_FALSE(!(condition))) { \ + ABSL_RAW_LOG(FATAL, "Check %s failed: %s", #condition, message); \ + } \ + } while (0) + +// ABSL_INTERNAL_LOG and ABSL_INTERNAL_CHECK work like the RAW variants above, +// except that if the richer log library is linked into the binary, we dispatch +// to that instead. This is potentially useful for internal logging and +// assertions, where we are using RAW_LOG neither for its async-signal-safety +// nor for its non-allocating nature, but rather because raw logging has very +// few other dependencies. +// +// The API is a subset of the above: each macro only takes two arguments. Use +// StrCat if you need to build a richer message. +#define ABSL_INTERNAL_LOG(severity, message) \ + do { \ + constexpr const char* absl_raw_logging_internal_filename = __FILE__; \ + ::absl::raw_logging_internal::internal_log_function( \ + ABSL_RAW_LOGGING_INTERNAL_##severity, \ + absl_raw_logging_internal_filename, __LINE__, message); \ + if (ABSL_RAW_LOGGING_INTERNAL_##severity == ::absl::LogSeverity::kFatal) \ + ABSL_INTERNAL_UNREACHABLE; \ + } while (0) + +#define ABSL_INTERNAL_CHECK(condition, message) \ + do { \ + if (ABSL_PREDICT_FALSE(!(condition))) { \ + std::string death_message = "Check " #condition " failed: "; \ + death_message += std::string(message); \ + ABSL_INTERNAL_LOG(FATAL, death_message); \ + } \ + } while (0) + +#define ABSL_RAW_LOGGING_INTERNAL_INFO ::absl::LogSeverity::kInfo +#define ABSL_RAW_LOGGING_INTERNAL_WARNING ::absl::LogSeverity::kWarning +#define ABSL_RAW_LOGGING_INTERNAL_ERROR ::absl::LogSeverity::kError +#define ABSL_RAW_LOGGING_INTERNAL_FATAL ::absl::LogSeverity::kFatal +#define ABSL_RAW_LOGGING_INTERNAL_LEVEL(severity) \ + ::absl::NormalizeLogSeverity(severity) + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace raw_logging_internal { + +// Helper function to implement ABSL_RAW_LOG +// Logs format... at "severity" level, reporting it +// as called from file:line. +// This does not allocate memory or acquire locks. +void RawLog(absl::LogSeverity severity, const char* file, int line, + const char* format, ...) ABSL_PRINTF_ATTRIBUTE(4, 5); + +// Writes the provided buffer directly to stderr, in a signal-safe, low-level +// manner. +void AsyncSignalSafeWriteToStderr(const char* s, size_t len); + +// compile-time function to get the "base" filename, that is, the part of +// a filename after the last "/" or "\" path separator. The search starts at +// the end of the string; the second parameter is the length of the string. +constexpr const char* Basename(const char* fname, int offset) { + return offset == 0 || fname[offset - 1] == '/' || fname[offset - 1] == '\\' + ? fname + offset + : Basename(fname, offset - 1); +} + +// For testing only. +// Returns true if raw logging is fully supported. When it is not +// fully supported, no messages will be emitted, but a log at FATAL +// severity will cause an abort. +// +// TODO(gfalcon): Come up with a better name for this method. +bool RawLoggingFullySupported(); + +// Function type for a raw_logging customization hook for suppressing messages +// by severity, and for writing custom prefixes on non-suppressed messages. +// +// The installed hook is called for every raw log invocation. The message will +// be logged to stderr only if the hook returns true. FATAL errors will cause +// the process to abort, even if writing to stderr is suppressed. The hook is +// also provided with an output buffer, where it can write a custom log message +// prefix. +// +// The raw_logging system does not allocate memory or grab locks. User-provided +// hooks must avoid these operations, and must not throw exceptions. +// +// 'severity' is the severity level of the message being written. +// 'file' and 'line' are the file and line number where the ABSL_RAW_LOG macro +// was located. +// 'buf' and 'buf_size' are pointers to the buffer and buffer size. If the +// hook writes a prefix, it must increment *buf and decrement *buf_size +// accordingly. +using LogFilterAndPrefixHook = bool (*)(absl::LogSeverity severity, + const char* file, int line, char** buf, + int* buf_size); + +// Function type for a raw_logging customization hook called to abort a process +// when a FATAL message is logged. If the provided AbortHook() returns, the +// logging system will call abort(). +// +// 'file' and 'line' are the file and line number where the ABSL_RAW_LOG macro +// was located. +// The NUL-terminated logged message lives in the buffer between 'buf_start' +// and 'buf_end'. 'prefix_end' points to the first non-prefix character of the +// buffer (as written by the LogFilterAndPrefixHook.) +// +// The lifetime of the filename and message buffers will not end while the +// process remains alive. +using AbortHook = void (*)(const char* file, int line, const char* buf_start, + const char* prefix_end, const char* buf_end); + +// Internal logging function for ABSL_INTERNAL_LOG to dispatch to. +// +// TODO(gfalcon): When string_view no longer depends on base, change this +// interface to take its message as a string_view instead. +using InternalLogFunction = void (*)(absl::LogSeverity severity, + const char* file, int line, + const std::string& message); + +ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES ABSL_DLL extern base_internal::AtomicHook< + InternalLogFunction> + internal_log_function; + +// Registers hooks of the above types. Only a single hook of each type may be +// registered. It is an error to call these functions multiple times with +// different input arguments. +// +// These functions are safe to call at any point during initialization; they do +// not block or malloc, and are async-signal safe. +void RegisterLogFilterAndPrefixHook(LogFilterAndPrefixHook func); +void RegisterAbortHook(AbortHook func); +void RegisterInternalLogFunction(InternalLogFunction func); + +} // namespace raw_logging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_RAW_LOGGING_H_ diff --git a/src/absl/base/internal/scheduling_mode.h b/src/absl/base/internal/scheduling_mode.h new file mode 100644 index 0000000..8be5ab6 --- /dev/null +++ b/src/absl/base/internal/scheduling_mode.h @@ -0,0 +1,58 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Core interfaces and definitions used by by low-level interfaces such as +// SpinLock. + +#ifndef ABSL_BASE_INTERNAL_SCHEDULING_MODE_H_ +#define ABSL_BASE_INTERNAL_SCHEDULING_MODE_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Used to describe how a thread may be scheduled. Typically associated with +// the declaration of a resource supporting synchronized access. +// +// SCHEDULE_COOPERATIVE_AND_KERNEL: +// Specifies that when waiting, a cooperative thread (e.g. a Fiber) may +// reschedule (using base::scheduling semantics); allowing other cooperative +// threads to proceed. +// +// SCHEDULE_KERNEL_ONLY: (Also described as "non-cooperative") +// Specifies that no cooperative scheduling semantics may be used, even if the +// current thread is itself cooperatively scheduled. This means that +// cooperative threads will NOT allow other cooperative threads to execute in +// their place while waiting for a resource of this type. Host operating system +// semantics (e.g. a futex) may still be used. +// +// When optional, clients should strongly prefer SCHEDULE_COOPERATIVE_AND_KERNEL +// by default. SCHEDULE_KERNEL_ONLY should only be used for resources on which +// base::scheduling (e.g. the implementation of a Scheduler) may depend. +// +// NOTE: Cooperative resources may not be nested below non-cooperative ones. +// This means that it is invalid to to acquire a SCHEDULE_COOPERATIVE_AND_KERNEL +// resource if a SCHEDULE_KERNEL_ONLY resource is already held. +enum SchedulingMode { + SCHEDULE_KERNEL_ONLY = 0, // Allow scheduling only the host OS. + SCHEDULE_COOPERATIVE_AND_KERNEL, // Also allow cooperative scheduling. +}; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_SCHEDULING_MODE_H_ diff --git a/src/absl/base/internal/scoped_set_env.cc b/src/absl/base/internal/scoped_set_env.cc new file mode 100644 index 0000000..8a934cb --- /dev/null +++ b/src/absl/base/internal/scoped_set_env.cc @@ -0,0 +1,81 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/scoped_set_env.h" + +#ifdef _WIN32 +#include +#endif + +#include + +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +namespace { + +#ifdef _WIN32 +const int kMaxEnvVarValueSize = 1024; +#endif + +void SetEnvVar(const char* name, const char* value) { +#ifdef _WIN32 + SetEnvironmentVariableA(name, value); +#else + if (value == nullptr) { + ::unsetenv(name); + } else { + ::setenv(name, value, 1); + } +#endif +} + +} // namespace + +ScopedSetEnv::ScopedSetEnv(const char* var_name, const char* new_value) + : var_name_(var_name), was_unset_(false) { +#ifdef _WIN32 + char buf[kMaxEnvVarValueSize]; + auto get_res = GetEnvironmentVariableA(var_name_.c_str(), buf, sizeof(buf)); + ABSL_INTERNAL_CHECK(get_res < sizeof(buf), "value exceeds buffer size"); + + if (get_res == 0) { + was_unset_ = (GetLastError() == ERROR_ENVVAR_NOT_FOUND); + } else { + old_value_.assign(buf, get_res); + } + + SetEnvironmentVariableA(var_name_.c_str(), new_value); +#else + const char* val = ::getenv(var_name_.c_str()); + if (val == nullptr) { + was_unset_ = true; + } else { + old_value_ = val; + } +#endif + + SetEnvVar(var_name_.c_str(), new_value); +} + +ScopedSetEnv::~ScopedSetEnv() { + SetEnvVar(var_name_.c_str(), was_unset_ ? nullptr : old_value_.c_str()); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/scoped_set_env.h b/src/absl/base/internal/scoped_set_env.h new file mode 100644 index 0000000..19ec7b5 --- /dev/null +++ b/src/absl/base/internal/scoped_set_env.h @@ -0,0 +1,45 @@ +// +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_SCOPED_SET_ENV_H_ +#define ABSL_BASE_INTERNAL_SCOPED_SET_ENV_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +class ScopedSetEnv { + public: + ScopedSetEnv(const char* var_name, const char* new_value); + ~ScopedSetEnv(); + + private: + std::string var_name_; + std::string old_value_; + + // True if the environment variable was initially not set. + bool was_unset_; +}; + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_SCOPED_SET_ENV_H_ diff --git a/src/absl/base/internal/spinlock.cc b/src/absl/base/internal/spinlock.cc new file mode 100644 index 0000000..9b5ed6e --- /dev/null +++ b/src/absl/base/internal/spinlock.cc @@ -0,0 +1,232 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/spinlock.h" + +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/internal/atomic_hook.h" +#include "absl/base/internal/cycleclock.h" +#include "absl/base/internal/spinlock_wait.h" +#include "absl/base/internal/sysinfo.h" /* For NumCPUs() */ +#include "absl/base/call_once.h" + +// Description of lock-word: +// 31..00: [............................3][2][1][0] +// +// [0]: kSpinLockHeld +// [1]: kSpinLockCooperative +// [2]: kSpinLockDisabledScheduling +// [31..3]: ONLY kSpinLockSleeper OR +// Wait time in cycles >> PROFILE_TIMESTAMP_SHIFT +// +// Detailed descriptions: +// +// Bit [0]: The lock is considered held iff kSpinLockHeld is set. +// +// Bit [1]: Eligible waiters (e.g. Fibers) may co-operatively reschedule when +// contended iff kSpinLockCooperative is set. +// +// Bit [2]: This bit is exclusive from bit [1]. It is used only by a +// non-cooperative lock. When set, indicates that scheduling was +// successfully disabled when the lock was acquired. May be unset, +// even if non-cooperative, if a ThreadIdentity did not yet exist at +// time of acquisition. +// +// Bit [3]: If this is the only upper bit ([31..3]) set then this lock was +// acquired without contention, however, at least one waiter exists. +// +// Otherwise, bits [31..3] represent the time spent by the current lock +// holder to acquire the lock. There may be outstanding waiter(s). + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +ABSL_INTERNAL_ATOMIC_HOOK_ATTRIBUTES static base_internal::AtomicHook + submit_profile_data; + +void RegisterSpinLockProfiler(void (*fn)(const void *contendedlock, + int64_t wait_cycles)) { + submit_profile_data.Store(fn); +} + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +// Static member variable definitions. +constexpr uint32_t SpinLock::kSpinLockHeld; +constexpr uint32_t SpinLock::kSpinLockCooperative; +constexpr uint32_t SpinLock::kSpinLockDisabledScheduling; +constexpr uint32_t SpinLock::kSpinLockSleeper; +constexpr uint32_t SpinLock::kWaitTimeMask; +#endif + +// Uncommon constructors. +SpinLock::SpinLock(base_internal::SchedulingMode mode) + : lockword_(IsCooperative(mode) ? kSpinLockCooperative : 0) { + ABSL_TSAN_MUTEX_CREATE(this, __tsan_mutex_not_static); +} + +// Monitor the lock to see if its value changes within some time period +// (adaptive_spin_count loop iterations). The last value read from the lock +// is returned from the method. +uint32_t SpinLock::SpinLoop() { + // We are already in the slow path of SpinLock, initialize the + // adaptive_spin_count here. + ABSL_CONST_INIT static absl::once_flag init_adaptive_spin_count; + ABSL_CONST_INIT static int adaptive_spin_count = 0; + base_internal::LowLevelCallOnce(&init_adaptive_spin_count, []() { + adaptive_spin_count = base_internal::NumCPUs() > 1 ? 1000 : 1; + }); + + int c = adaptive_spin_count; + uint32_t lock_value; + do { + lock_value = lockword_.load(std::memory_order_relaxed); + } while ((lock_value & kSpinLockHeld) != 0 && --c > 0); + return lock_value; +} + +void SpinLock::SlowLock() { + uint32_t lock_value = SpinLoop(); + lock_value = TryLockInternal(lock_value, 0); + if ((lock_value & kSpinLockHeld) == 0) { + return; + } + + base_internal::SchedulingMode scheduling_mode; + if ((lock_value & kSpinLockCooperative) != 0) { + scheduling_mode = base_internal::SCHEDULE_COOPERATIVE_AND_KERNEL; + } else { + scheduling_mode = base_internal::SCHEDULE_KERNEL_ONLY; + } + + // The lock was not obtained initially, so this thread needs to wait for + // it. Record the current timestamp in the local variable wait_start_time + // so the total wait time can be stored in the lockword once this thread + // obtains the lock. + int64_t wait_start_time = CycleClock::Now(); + uint32_t wait_cycles = 0; + int lock_wait_call_count = 0; + while ((lock_value & kSpinLockHeld) != 0) { + // If the lock is currently held, but not marked as having a sleeper, mark + // it as having a sleeper. + if ((lock_value & kWaitTimeMask) == 0) { + // Here, just "mark" that the thread is going to sleep. Don't store the + // lock wait time in the lock -- the lock word stores the amount of time + // that the current holder waited before acquiring the lock, not the wait + // time of any thread currently waiting to acquire it. + if (lockword_.compare_exchange_strong( + lock_value, lock_value | kSpinLockSleeper, + std::memory_order_relaxed, std::memory_order_relaxed)) { + // Successfully transitioned to kSpinLockSleeper. Pass + // kSpinLockSleeper to the SpinLockWait routine to properly indicate + // the last lock_value observed. + lock_value |= kSpinLockSleeper; + } else if ((lock_value & kSpinLockHeld) == 0) { + // Lock is free again, so try and acquire it before sleeping. The + // new lock state will be the number of cycles this thread waited if + // this thread obtains the lock. + lock_value = TryLockInternal(lock_value, wait_cycles); + continue; // Skip the delay at the end of the loop. + } else if ((lock_value & kWaitTimeMask) == 0) { + // The lock is still held, without a waiter being marked, but something + // else about the lock word changed, causing our CAS to fail. For + // example, a new lock holder may have acquired the lock with + // kSpinLockDisabledScheduling set, whereas the previous holder had not + // set that flag. In this case, attempt again to mark ourselves as a + // waiter. + continue; + } + } + + // SpinLockDelay() calls into fiber scheduler, we need to see + // synchronization there to avoid false positives. + ABSL_TSAN_MUTEX_PRE_DIVERT(this, 0); + // Wait for an OS specific delay. + base_internal::SpinLockDelay(&lockword_, lock_value, ++lock_wait_call_count, + scheduling_mode); + ABSL_TSAN_MUTEX_POST_DIVERT(this, 0); + // Spin again after returning from the wait routine to give this thread + // some chance of obtaining the lock. + lock_value = SpinLoop(); + wait_cycles = EncodeWaitCycles(wait_start_time, CycleClock::Now()); + lock_value = TryLockInternal(lock_value, wait_cycles); + } +} + +void SpinLock::SlowUnlock(uint32_t lock_value) { + base_internal::SpinLockWake(&lockword_, + false); // wake waiter if necessary + + // If our acquisition was contended, collect contentionz profile info. We + // reserve a unitary wait time to represent that a waiter exists without our + // own acquisition having been contended. + if ((lock_value & kWaitTimeMask) != kSpinLockSleeper) { + const uint64_t wait_cycles = DecodeWaitCycles(lock_value); + ABSL_TSAN_MUTEX_PRE_DIVERT(this, 0); + submit_profile_data(this, wait_cycles); + ABSL_TSAN_MUTEX_POST_DIVERT(this, 0); + } +} + +// We use the upper 29 bits of the lock word to store the time spent waiting to +// acquire this lock. This is reported by contentionz profiling. Since the +// lower bits of the cycle counter wrap very quickly on high-frequency +// processors we divide to reduce the granularity to 2^kProfileTimestampShift +// sized units. On a 4Ghz machine this will lose track of wait times greater +// than (2^29/4 Ghz)*128 =~ 17.2 seconds. Such waits should be extremely rare. +static constexpr int kProfileTimestampShift = 7; + +// We currently reserve the lower 3 bits. +static constexpr int kLockwordReservedShift = 3; + +uint32_t SpinLock::EncodeWaitCycles(int64_t wait_start_time, + int64_t wait_end_time) { + static const int64_t kMaxWaitTime = + std::numeric_limits::max() >> kLockwordReservedShift; + int64_t scaled_wait_time = + (wait_end_time - wait_start_time) >> kProfileTimestampShift; + + // Return a representation of the time spent waiting that can be stored in + // the lock word's upper bits. + uint32_t clamped = static_cast( + std::min(scaled_wait_time, kMaxWaitTime) << kLockwordReservedShift); + + if (clamped == 0) { + return kSpinLockSleeper; // Just wake waiters, but don't record contention. + } + // Bump up value if necessary to avoid returning kSpinLockSleeper. + const uint32_t kMinWaitTime = + kSpinLockSleeper + (1 << kLockwordReservedShift); + if (clamped == kSpinLockSleeper) { + return kMinWaitTime; + } + return clamped; +} + +uint64_t SpinLock::DecodeWaitCycles(uint32_t lock_value) { + // Cast to uint32_t first to ensure bits [63:32] are cleared. + const uint64_t scaled_wait_time = + static_cast(lock_value & kWaitTimeMask); + return scaled_wait_time << (kProfileTimestampShift - kLockwordReservedShift); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/spinlock.h b/src/absl/base/internal/spinlock.h new file mode 100644 index 0000000..6d8d8dd --- /dev/null +++ b/src/absl/base/internal/spinlock.h @@ -0,0 +1,256 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +// Most users requiring mutual exclusion should use Mutex. +// SpinLock is provided for use in two situations: +// - for use by Abseil internal code that Mutex itself depends on +// - for async signal safety (see below) + +// SpinLock is async signal safe. If a spinlock is used within a signal +// handler, all code that acquires the lock must ensure that the signal cannot +// arrive while they are holding the lock. Typically, this is done by blocking +// the signal. +// +// Threads waiting on a SpinLock may be woken in an arbitrary order. + +#ifndef ABSL_BASE_INTERNAL_SPINLOCK_H_ +#define ABSL_BASE_INTERNAL_SPINLOCK_H_ + +#include +#include + +#include + +#include "absl/base/attributes.h" +#include "absl/base/const_init.h" +#include "absl/base/dynamic_annotations.h" +#include "absl/base/internal/low_level_scheduling.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/scheduling_mode.h" +#include "absl/base/internal/tsan_mutex_interface.h" +#include "absl/base/macros.h" +#include "absl/base/port.h" +#include "absl/base/thread_annotations.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +class ABSL_LOCKABLE SpinLock { + public: + SpinLock() : lockword_(kSpinLockCooperative) { + ABSL_TSAN_MUTEX_CREATE(this, __tsan_mutex_not_static); + } + + // Constructors that allow non-cooperative spinlocks to be created for use + // inside thread schedulers. Normal clients should not use these. + explicit SpinLock(base_internal::SchedulingMode mode); + + // Constructor for global SpinLock instances. See absl/base/const_init.h. + constexpr SpinLock(absl::ConstInitType, base_internal::SchedulingMode mode) + : lockword_(IsCooperative(mode) ? kSpinLockCooperative : 0) {} + + // For global SpinLock instances prefer trivial destructor when possible. + // Default but non-trivial destructor in some build configurations causes an + // extra static initializer. +#ifdef ABSL_INTERNAL_HAVE_TSAN_INTERFACE + ~SpinLock() { ABSL_TSAN_MUTEX_DESTROY(this, __tsan_mutex_not_static); } +#else + ~SpinLock() = default; +#endif + + // Acquire this SpinLock. + inline void Lock() ABSL_EXCLUSIVE_LOCK_FUNCTION() { + ABSL_TSAN_MUTEX_PRE_LOCK(this, 0); + if (!TryLockImpl()) { + SlowLock(); + } + ABSL_TSAN_MUTEX_POST_LOCK(this, 0, 0); + } + + // Try to acquire this SpinLock without blocking and return true if the + // acquisition was successful. If the lock was not acquired, false is + // returned. If this SpinLock is free at the time of the call, TryLock + // will return true with high probability. + inline bool TryLock() ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(true) { + ABSL_TSAN_MUTEX_PRE_LOCK(this, __tsan_mutex_try_lock); + bool res = TryLockImpl(); + ABSL_TSAN_MUTEX_POST_LOCK( + this, __tsan_mutex_try_lock | (res ? 0 : __tsan_mutex_try_lock_failed), + 0); + return res; + } + + // Release this SpinLock, which must be held by the calling thread. + inline void Unlock() ABSL_UNLOCK_FUNCTION() { + ABSL_TSAN_MUTEX_PRE_UNLOCK(this, 0); + uint32_t lock_value = lockword_.load(std::memory_order_relaxed); + lock_value = lockword_.exchange(lock_value & kSpinLockCooperative, + std::memory_order_release); + + if ((lock_value & kSpinLockDisabledScheduling) != 0) { + base_internal::SchedulingGuard::EnableRescheduling(true); + } + if ((lock_value & kWaitTimeMask) != 0) { + // Collect contentionz profile info, and speed the wakeup of any waiter. + // The wait_cycles value indicates how long this thread spent waiting + // for the lock. + SlowUnlock(lock_value); + } + ABSL_TSAN_MUTEX_POST_UNLOCK(this, 0); + } + + // Determine if the lock is held. When the lock is held by the invoking + // thread, true will always be returned. Intended to be used as + // CHECK(lock.IsHeld()). + inline bool IsHeld() const { + return (lockword_.load(std::memory_order_relaxed) & kSpinLockHeld) != 0; + } + + // Return immediately if this thread holds the SpinLock exclusively. + // Otherwise, report an error by crashing with a diagnostic. + inline void AssertHeld() const ABSL_ASSERT_EXCLUSIVE_LOCK() { + if (!IsHeld()) { + ABSL_RAW_LOG(FATAL, "thread should hold the lock on SpinLock"); + } + } + + protected: + // These should not be exported except for testing. + + // Store number of cycles between wait_start_time and wait_end_time in a + // lock value. + static uint32_t EncodeWaitCycles(int64_t wait_start_time, + int64_t wait_end_time); + + // Extract number of wait cycles in a lock value. + static uint64_t DecodeWaitCycles(uint32_t lock_value); + + // Provide access to protected method above. Use for testing only. + friend struct SpinLockTest; + + private: + // lockword_ is used to store the following: + // + // bit[0] encodes whether a lock is being held. + // bit[1] encodes whether a lock uses cooperative scheduling. + // bit[2] encodes whether the current lock holder disabled scheduling when + // acquiring the lock. Only set when kSpinLockHeld is also set. + // bit[3:31] encodes time a lock spent on waiting as a 29-bit unsigned int. + // This is set by the lock holder to indicate how long it waited on + // the lock before eventually acquiring it. The number of cycles is + // encoded as a 29-bit unsigned int, or in the case that the current + // holder did not wait but another waiter is queued, the LSB + // (kSpinLockSleeper) is set. The implementation does not explicitly + // track the number of queued waiters beyond this. It must always be + // assumed that waiters may exist if the current holder was required to + // queue. + // + // Invariant: if the lock is not held, the value is either 0 or + // kSpinLockCooperative. + static constexpr uint32_t kSpinLockHeld = 1; + static constexpr uint32_t kSpinLockCooperative = 2; + static constexpr uint32_t kSpinLockDisabledScheduling = 4; + static constexpr uint32_t kSpinLockSleeper = 8; + // Includes kSpinLockSleeper. + static constexpr uint32_t kWaitTimeMask = + ~(kSpinLockHeld | kSpinLockCooperative | kSpinLockDisabledScheduling); + + // Returns true if the provided scheduling mode is cooperative. + static constexpr bool IsCooperative( + base_internal::SchedulingMode scheduling_mode) { + return scheduling_mode == base_internal::SCHEDULE_COOPERATIVE_AND_KERNEL; + } + + uint32_t TryLockInternal(uint32_t lock_value, uint32_t wait_cycles); + void SlowLock() ABSL_ATTRIBUTE_COLD; + void SlowUnlock(uint32_t lock_value) ABSL_ATTRIBUTE_COLD; + uint32_t SpinLoop(); + + inline bool TryLockImpl() { + uint32_t lock_value = lockword_.load(std::memory_order_relaxed); + return (TryLockInternal(lock_value, 0) & kSpinLockHeld) == 0; + } + + std::atomic lockword_; + + SpinLock(const SpinLock&) = delete; + SpinLock& operator=(const SpinLock&) = delete; +}; + +// Corresponding locker object that arranges to acquire a spinlock for +// the duration of a C++ scope. +class ABSL_SCOPED_LOCKABLE SpinLockHolder { + public: + inline explicit SpinLockHolder(SpinLock* l) ABSL_EXCLUSIVE_LOCK_FUNCTION(l) + : lock_(l) { + l->Lock(); + } + inline ~SpinLockHolder() ABSL_UNLOCK_FUNCTION() { lock_->Unlock(); } + + SpinLockHolder(const SpinLockHolder&) = delete; + SpinLockHolder& operator=(const SpinLockHolder&) = delete; + + private: + SpinLock* lock_; +}; + +// Register a hook for profiling support. +// +// The function pointer registered here will be called whenever a spinlock is +// contended. The callback is given an opaque handle to the contended spinlock +// and the number of wait cycles. This is thread-safe, but only a single +// profiler can be registered. It is an error to call this function multiple +// times with different arguments. +void RegisterSpinLockProfiler(void (*fn)(const void* lock, + int64_t wait_cycles)); + +//------------------------------------------------------------------------------ +// Public interface ends here. +//------------------------------------------------------------------------------ + +// If (result & kSpinLockHeld) == 0, then *this was successfully locked. +// Otherwise, returns last observed value for lockword_. +inline uint32_t SpinLock::TryLockInternal(uint32_t lock_value, + uint32_t wait_cycles) { + if ((lock_value & kSpinLockHeld) != 0) { + return lock_value; + } + + uint32_t sched_disabled_bit = 0; + if ((lock_value & kSpinLockCooperative) == 0) { + // For non-cooperative locks we must make sure we mark ourselves as + // non-reschedulable before we attempt to CompareAndSwap. + if (base_internal::SchedulingGuard::DisableRescheduling()) { + sched_disabled_bit = kSpinLockDisabledScheduling; + } + } + + if (!lockword_.compare_exchange_strong( + lock_value, + kSpinLockHeld | lock_value | wait_cycles | sched_disabled_bit, + std::memory_order_acquire, std::memory_order_relaxed)) { + base_internal::SchedulingGuard::EnableRescheduling(sched_disabled_bit != 0); + } + + return lock_value; +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_SPINLOCK_H_ diff --git a/src/absl/base/internal/spinlock_akaros.inc b/src/absl/base/internal/spinlock_akaros.inc new file mode 100644 index 0000000..7b0cada --- /dev/null +++ b/src/absl/base/internal/spinlock_akaros.inc @@ -0,0 +1,35 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file is an Akaros-specific part of spinlock_wait.cc + +#include + +#include "absl/base/internal/scheduling_mode.h" + +extern "C" { + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay)( + std::atomic* /* lock_word */, uint32_t /* value */, + int /* loop */, absl::base_internal::SchedulingMode /* mode */) { + // In Akaros, one must take care not to call anything that could cause a + // malloc(), a blocking system call, or a uthread_yield() while holding a + // spinlock. Our callers assume will not call into libraries or other + // arbitrary code. +} + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)( + std::atomic* /* lock_word */, bool /* all */) {} + +} // extern "C" diff --git a/src/absl/base/internal/spinlock_linux.inc b/src/absl/base/internal/spinlock_linux.inc new file mode 100644 index 0000000..fe8ba67 --- /dev/null +++ b/src/absl/base/internal/spinlock_linux.inc @@ -0,0 +1,71 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file is a Linux-specific part of spinlock_wait.cc + +#include +#include +#include + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/internal/errno_saver.h" + +// The SpinLock lockword is `std::atomic`. Here we assert that +// `std::atomic` is bitwise equivalent of the `int` expected +// by SYS_futex. We also assume that reads/writes done to the lockword +// by SYS_futex have rational semantics with regard to the +// std::atomic<> API. C++ provides no guarantees of these assumptions, +// but they are believed to hold in practice. +static_assert(sizeof(std::atomic) == sizeof(int), + "SpinLock lockword has the wrong size for a futex"); + +// Some Android headers are missing these definitions even though they +// support these futex operations. +#ifdef __BIONIC__ +#ifndef SYS_futex +#define SYS_futex __NR_futex +#endif +#ifndef FUTEX_PRIVATE_FLAG +#define FUTEX_PRIVATE_FLAG 128 +#endif +#endif + +#if defined(__NR_futex_time64) && !defined(SYS_futex_time64) +#define SYS_futex_time64 __NR_futex_time64 +#endif + +#if defined(SYS_futex_time64) && !defined(SYS_futex) +#define SYS_futex SYS_futex_time64 +#endif + +extern "C" { + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay)( + std::atomic *w, uint32_t value, int, + absl::base_internal::SchedulingMode) { + absl::base_internal::ErrnoSaver errno_saver; + syscall(SYS_futex, w, FUTEX_WAIT | FUTEX_PRIVATE_FLAG, value, nullptr); +} + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)( + std::atomic *w, bool all) { + syscall(SYS_futex, w, FUTEX_WAKE | FUTEX_PRIVATE_FLAG, all ? INT_MAX : 1, 0); +} + +} // extern "C" diff --git a/src/absl/base/internal/spinlock_posix.inc b/src/absl/base/internal/spinlock_posix.inc new file mode 100644 index 0000000..4f6f887 --- /dev/null +++ b/src/absl/base/internal/spinlock_posix.inc @@ -0,0 +1,46 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file is a Posix-specific part of spinlock_wait.cc + +#include + +#include +#include + +#include "absl/base/internal/errno_saver.h" +#include "absl/base/internal/scheduling_mode.h" +#include "absl/base/port.h" + +extern "C" { + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay)( + std::atomic* /* lock_word */, uint32_t /* value */, int loop, + absl::base_internal::SchedulingMode /* mode */) { + absl::base_internal::ErrnoSaver errno_saver; + if (loop == 0) { + } else if (loop == 1) { + sched_yield(); + } else { + struct timespec tm; + tm.tv_sec = 0; + tm.tv_nsec = absl::base_internal::SpinLockSuggestedDelayNS(loop); + nanosleep(&tm, nullptr); + } +} + +ABSL_ATTRIBUTE_WEAK void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)( + std::atomic* /* lock_word */, bool /* all */) {} + +} // extern "C" diff --git a/src/absl/base/internal/spinlock_wait.cc b/src/absl/base/internal/spinlock_wait.cc new file mode 100644 index 0000000..fa824be --- /dev/null +++ b/src/absl/base/internal/spinlock_wait.cc @@ -0,0 +1,81 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// The OS-specific header included below must provide two calls: +// AbslInternalSpinLockDelay() and AbslInternalSpinLockWake(). +// See spinlock_wait.h for the specs. + +#include +#include + +#include "absl/base/internal/spinlock_wait.h" + +#if defined(_WIN32) +#include "absl/base/internal/spinlock_win32.inc" +#elif defined(__linux__) +#include "absl/base/internal/spinlock_linux.inc" +#elif defined(__akaros__) +#include "absl/base/internal/spinlock_akaros.inc" +#else +#include "absl/base/internal/spinlock_posix.inc" +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// See spinlock_wait.h for spec. +uint32_t SpinLockWait(std::atomic *w, int n, + const SpinLockWaitTransition trans[], + base_internal::SchedulingMode scheduling_mode) { + int loop = 0; + for (;;) { + uint32_t v = w->load(std::memory_order_acquire); + int i; + for (i = 0; i != n && v != trans[i].from; i++) { + } + if (i == n) { + SpinLockDelay(w, v, ++loop, scheduling_mode); // no matching transition + } else if (trans[i].to == v || // null transition + w->compare_exchange_strong(v, trans[i].to, + std::memory_order_acquire, + std::memory_order_relaxed)) { + if (trans[i].done) return v; + } + } +} + +static std::atomic delay_rand; + +// Return a suggested delay in nanoseconds for iteration number "loop" +int SpinLockSuggestedDelayNS(int loop) { + // Weak pseudo-random number generator to get some spread between threads + // when many are spinning. + uint64_t r = delay_rand.load(std::memory_order_relaxed); + r = 0x5deece66dLL * r + 0xb; // numbers from nrand48() + delay_rand.store(r, std::memory_order_relaxed); + + if (loop < 0 || loop > 32) { // limit loop to 0..32 + loop = 32; + } + const int kMinDelay = 128 << 10; // 128us + // Double delay every 8 iterations, up to 16x (2ms). + int delay = kMinDelay << (loop / 8); + // Randomize in delay..2*delay range, for resulting 128us..4ms range. + return delay | ((delay - 1) & static_cast(r)); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/spinlock_wait.h b/src/absl/base/internal/spinlock_wait.h new file mode 100644 index 0000000..9a1adcd --- /dev/null +++ b/src/absl/base/internal/spinlock_wait.h @@ -0,0 +1,95 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_SPINLOCK_WAIT_H_ +#define ABSL_BASE_INTERNAL_SPINLOCK_WAIT_H_ + +// Operations to make atomic transitions on a word, and to allow +// waiting for those transitions to become possible. + +#include +#include + +#include "absl/base/internal/scheduling_mode.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// SpinLockWait() waits until it can perform one of several transitions from +// "from" to "to". It returns when it performs a transition where done==true. +struct SpinLockWaitTransition { + uint32_t from; + uint32_t to; + bool done; +}; + +// Wait until *w can transition from trans[i].from to trans[i].to for some i +// satisfying 0<=i *w, int n, + const SpinLockWaitTransition trans[], + SchedulingMode scheduling_mode); + +// If possible, wake some thread that has called SpinLockDelay(w, ...). If `all` +// is true, wake all such threads. On some systems, this may be a no-op; on +// those systems, threads calling SpinLockDelay() will always wake eventually +// even if SpinLockWake() is never called. +void SpinLockWake(std::atomic *w, bool all); + +// Wait for an appropriate spin delay on iteration "loop" of a +// spin loop on location *w, whose previously observed value was "value". +// SpinLockDelay() may do nothing, may yield the CPU, may sleep a clock tick, +// or may wait for a call to SpinLockWake(w). +void SpinLockDelay(std::atomic *w, uint32_t value, int loop, + base_internal::SchedulingMode scheduling_mode); + +// Helper used by AbslInternalSpinLockDelay. +// Returns a suggested delay in nanoseconds for iteration number "loop". +int SpinLockSuggestedDelayNS(int loop); + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +// In some build configurations we pass --detect-odr-violations to the +// gold linker. This causes it to flag weak symbol overrides as ODR +// violations. Because ODR only applies to C++ and not C, +// --detect-odr-violations ignores symbols not mangled with C++ names. +// By changing our extension points to be extern "C", we dodge this +// check. +extern "C" { +void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)(std::atomic *w, + bool all); +void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay)( + std::atomic *w, uint32_t value, int loop, + absl::base_internal::SchedulingMode scheduling_mode); +} + +inline void absl::base_internal::SpinLockWake(std::atomic *w, + bool all) { + ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)(w, all); +} + +inline void absl::base_internal::SpinLockDelay( + std::atomic *w, uint32_t value, int loop, + absl::base_internal::SchedulingMode scheduling_mode) { + ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay) + (w, value, loop, scheduling_mode); +} + +#endif // ABSL_BASE_INTERNAL_SPINLOCK_WAIT_H_ diff --git a/src/absl/base/internal/spinlock_win32.inc b/src/absl/base/internal/spinlock_win32.inc new file mode 100644 index 0000000..9d22481 --- /dev/null +++ b/src/absl/base/internal/spinlock_win32.inc @@ -0,0 +1,37 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file is a Win32-specific part of spinlock_wait.cc + +#include +#include +#include "absl/base/internal/scheduling_mode.h" + +extern "C" { + +void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockDelay)( + std::atomic* /* lock_word */, uint32_t /* value */, int loop, + absl::base_internal::SchedulingMode /* mode */) { + if (loop == 0) { + } else if (loop == 1) { + Sleep(0); + } else { + Sleep(absl::base_internal::SpinLockSuggestedDelayNS(loop) / 1000000); + } +} + +void ABSL_INTERNAL_C_SYMBOL(AbslInternalSpinLockWake)( + std::atomic* /* lock_word */, bool /* all */) {} + +} // extern "C" diff --git a/src/absl/base/internal/strerror.cc b/src/absl/base/internal/strerror.cc new file mode 100644 index 0000000..0d6226f --- /dev/null +++ b/src/absl/base/internal/strerror.cc @@ -0,0 +1,88 @@ +// Copyright 2020 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/strerror.h" + +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/internal/errno_saver.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { +namespace { + +const char* StrErrorAdaptor(int errnum, char* buf, size_t buflen) { +#if defined(_WIN32) + int rc = strerror_s(buf, buflen, errnum); + buf[buflen - 1] = '\0'; // guarantee NUL termination + if (rc == 0 && strncmp(buf, "Unknown error", buflen) == 0) *buf = '\0'; + return buf; +#else + // The type of `ret` is platform-specific; both of these branches must compile + // either way but only one will execute on any given platform: + auto ret = strerror_r(errnum, buf, buflen); + if (std::is_same::value) { + // XSI `strerror_r`; `ret` is `int`: + if (ret) *buf = '\0'; + return buf; + } else { + // GNU `strerror_r`; `ret` is `char *`: + return reinterpret_cast(ret); + } +#endif +} + +std::string StrErrorInternal(int errnum) { + char buf[100]; + const char* str = StrErrorAdaptor(errnum, buf, sizeof buf); + if (*str == '\0') { + snprintf(buf, sizeof buf, "Unknown error %d", errnum); + str = buf; + } + return str; +} + +// kSysNerr is the number of errors from a recent glibc. `StrError()` falls back +// to `StrErrorAdaptor()` if the value is larger than this. +constexpr int kSysNerr = 135; + +std::array* NewStrErrorTable() { + auto* table = new std::array; + for (int i = 0; i < static_cast(table->size()); ++i) { + (*table)[i] = StrErrorInternal(i); + } + return table; +} + +} // namespace + +std::string StrError(int errnum) { + absl::base_internal::ErrnoSaver errno_saver; + static const auto* table = NewStrErrorTable(); + if (errnum >= 0 && errnum < static_cast(table->size())) { + return (*table)[errnum]; + } + return StrErrorInternal(errnum); +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/strerror.h b/src/absl/base/internal/strerror.h new file mode 100644 index 0000000..3500973 --- /dev/null +++ b/src/absl/base/internal/strerror.h @@ -0,0 +1,39 @@ +// Copyright 2020 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_INTERNAL_STRERROR_H_ +#define ABSL_BASE_INTERNAL_STRERROR_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// A portable and thread-safe alternative to C89's `strerror`. +// +// The C89 specification of `strerror` is not suitable for use in a +// multi-threaded application as the returned string may be changed by calls to +// `strerror` from another thread. The many non-stdlib alternatives differ +// enough in their names, availability, and semantics to justify this wrapper +// around them. `errno` will not be modified by a call to `absl::StrError`. +std::string StrError(int errnum); + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_STRERROR_H_ diff --git a/src/absl/base/internal/sysinfo.cc b/src/absl/base/internal/sysinfo.cc new file mode 100644 index 0000000..3d78f8a --- /dev/null +++ b/src/absl/base/internal/sysinfo.cc @@ -0,0 +1,507 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/sysinfo.h" + +#include "absl/base/attributes.h" + +#ifdef _WIN32 +#include +#else +#include +#include +#include +#include +#include +#endif + +#ifdef __linux__ +#include +#endif + +#if defined(__APPLE__) || defined(__FreeBSD__) +#include +#endif + +#if defined(__myriad2__) +#include +#endif + +#include + +#include +#include +#include +#include +#include +#include +#include // NOLINT(build/c++11) +#include +#include + +#include "absl/base/call_once.h" +#include "absl/base/config.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/spinlock.h" +#include "absl/base/internal/unscaledcycleclock.h" +#include "absl/base/thread_annotations.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +namespace { + +#if defined(_WIN32) + +// Returns number of bits set in `bitMask` +DWORD Win32CountSetBits(ULONG_PTR bitMask) { + for (DWORD bitSetCount = 0; ; ++bitSetCount) { + if (bitMask == 0) return bitSetCount; + bitMask &= bitMask - 1; + } +} + +// Returns the number of logical CPUs using GetLogicalProcessorInformation(), or +// 0 if the number of processors is not available or can not be computed. +// https://docs.microsoft.com/en-us/windows/win32/api/sysinfoapi/nf-sysinfoapi-getlogicalprocessorinformation +int Win32NumCPUs() { +#pragma comment(lib, "kernel32.lib") + using Info = SYSTEM_LOGICAL_PROCESSOR_INFORMATION; + + DWORD info_size = sizeof(Info); + Info* info(static_cast(malloc(info_size))); + if (info == nullptr) return 0; + + bool success = GetLogicalProcessorInformation(info, &info_size); + if (!success && GetLastError() == ERROR_INSUFFICIENT_BUFFER) { + free(info); + info = static_cast(malloc(info_size)); + if (info == nullptr) return 0; + success = GetLogicalProcessorInformation(info, &info_size); + } + + DWORD logicalProcessorCount = 0; + if (success) { + Info* ptr = info; + DWORD byteOffset = 0; + while (byteOffset + sizeof(Info) <= info_size) { + switch (ptr->Relationship) { + case RelationProcessorCore: + logicalProcessorCount += Win32CountSetBits(ptr->ProcessorMask); + break; + + case RelationNumaNode: + case RelationCache: + case RelationProcessorPackage: + // Ignore other entries + break; + + default: + // Ignore unknown entries + break; + } + byteOffset += sizeof(Info); + ptr++; + } + } + free(info); + return logicalProcessorCount; +} + +#endif + +} // namespace + +static int GetNumCPUs() { +#if defined(__myriad2__) + return 1; +#elif defined(_WIN32) + const unsigned hardware_concurrency = Win32NumCPUs(); + return hardware_concurrency ? hardware_concurrency : 1; +#elif defined(_AIX) + return sysconf(_SC_NPROCESSORS_ONLN); +#else + // Other possibilities: + // - Read /sys/devices/system/cpu/online and use cpumask_parse() + // - sysconf(_SC_NPROCESSORS_ONLN) + return std::thread::hardware_concurrency(); +#endif +} + +#if defined(_WIN32) + +static double GetNominalCPUFrequency() { +#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP) && \ + !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP) + // UWP apps don't have access to the registry and currently don't provide an + // API informing about CPU nominal frequency. + return 1.0; +#else +#pragma comment(lib, "advapi32.lib") // For Reg* functions. + HKEY key; + // Use the Reg* functions rather than the SH functions because shlwapi.dll + // pulls in gdi32.dll which makes process destruction much more costly. + if (RegOpenKeyExA(HKEY_LOCAL_MACHINE, + "HARDWARE\\DESCRIPTION\\System\\CentralProcessor\\0", 0, + KEY_READ, &key) == ERROR_SUCCESS) { + DWORD type = 0; + DWORD data = 0; + DWORD data_size = sizeof(data); + auto result = RegQueryValueExA(key, "~MHz", 0, &type, + reinterpret_cast(&data), &data_size); + RegCloseKey(key); + if (result == ERROR_SUCCESS && type == REG_DWORD && + data_size == sizeof(data)) { + return data * 1e6; // Value is MHz. + } + } + return 1.0; +#endif // WINAPI_PARTITION_APP && !WINAPI_PARTITION_DESKTOP +} + +#elif defined(CTL_HW) && defined(HW_CPU_FREQ) + +static double GetNominalCPUFrequency() { + unsigned freq; + size_t size = sizeof(freq); + int mib[2] = {CTL_HW, HW_CPU_FREQ}; + if (sysctl(mib, 2, &freq, &size, nullptr, 0) == 0) { + return static_cast(freq); + } + return 1.0; +} + +#else + +// Helper function for reading a long from a file. Returns true if successful +// and the memory location pointed to by value is set to the value read. +static bool ReadLongFromFile(const char *file, long *value) { + bool ret = false; + int fd = open(file, O_RDONLY); + if (fd != -1) { + char line[1024]; + char *err; + memset(line, '\0', sizeof(line)); + int len = read(fd, line, sizeof(line) - 1); + if (len <= 0) { + ret = false; + } else { + const long temp_value = strtol(line, &err, 10); + if (line[0] != '\0' && (*err == '\n' || *err == '\0')) { + *value = temp_value; + ret = true; + } + } + close(fd); + } + return ret; +} + +#if defined(ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY) + +// Reads a monotonic time source and returns a value in +// nanoseconds. The returned value uses an arbitrary epoch, not the +// Unix epoch. +static int64_t ReadMonotonicClockNanos() { + struct timespec t; +#ifdef CLOCK_MONOTONIC_RAW + int rc = clock_gettime(CLOCK_MONOTONIC_RAW, &t); +#else + int rc = clock_gettime(CLOCK_MONOTONIC, &t); +#endif + if (rc != 0) { + perror("clock_gettime() failed"); + throw std::runtime_error("abort()"); + } + return int64_t{t.tv_sec} * 1000000000 + t.tv_nsec; +} + +class UnscaledCycleClockWrapperForInitializeFrequency { + public: + static int64_t Now() { return base_internal::UnscaledCycleClock::Now(); } +}; + +struct TimeTscPair { + int64_t time; // From ReadMonotonicClockNanos(). + int64_t tsc; // From UnscaledCycleClock::Now(). +}; + +// Returns a pair of values (monotonic kernel time, TSC ticks) that +// approximately correspond to each other. This is accomplished by +// doing several reads and picking the reading with the lowest +// latency. This approach is used to minimize the probability that +// our thread was preempted between clock reads. +static TimeTscPair GetTimeTscPair() { + int64_t best_latency = std::numeric_limits::max(); + TimeTscPair best; + for (int i = 0; i < 10; ++i) { + int64_t t0 = ReadMonotonicClockNanos(); + int64_t tsc = UnscaledCycleClockWrapperForInitializeFrequency::Now(); + int64_t t1 = ReadMonotonicClockNanos(); + int64_t latency = t1 - t0; + if (latency < best_latency) { + best_latency = latency; + best.time = t0; + best.tsc = tsc; + } + } + return best; +} + +// Measures and returns the TSC frequency by taking a pair of +// measurements approximately `sleep_nanoseconds` apart. +static double MeasureTscFrequencyWithSleep(int sleep_nanoseconds) { + auto t0 = GetTimeTscPair(); + struct timespec ts; + ts.tv_sec = 0; + ts.tv_nsec = sleep_nanoseconds; + while (nanosleep(&ts, &ts) != 0 && errno == EINTR) {} + auto t1 = GetTimeTscPair(); + double elapsed_ticks = t1.tsc - t0.tsc; + double elapsed_time = (t1.time - t0.time) * 1e-9; + return elapsed_ticks / elapsed_time; +} + +// Measures and returns the TSC frequency by calling +// MeasureTscFrequencyWithSleep(), doubling the sleep interval until the +// frequency measurement stabilizes. +static double MeasureTscFrequency() { + double last_measurement = -1.0; + int sleep_nanoseconds = 1000000; // 1 millisecond. + for (int i = 0; i < 8; ++i) { + double measurement = MeasureTscFrequencyWithSleep(sleep_nanoseconds); + if (measurement * 0.99 < last_measurement && + last_measurement < measurement * 1.01) { + // Use the current measurement if it is within 1% of the + // previous measurement. + return measurement; + } + last_measurement = measurement; + sleep_nanoseconds *= 2; + } + return last_measurement; +} + +#endif // ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY + +static double GetNominalCPUFrequency() { + long freq = 0; + + // Google's production kernel has a patch to export the TSC + // frequency through sysfs. If the kernel is exporting the TSC + // frequency use that. There are issues where cpuinfo_max_freq + // cannot be relied on because the BIOS may be exporting an invalid + // p-state (on x86) or p-states may be used to put the processor in + // a new mode (turbo mode). Essentially, those frequencies cannot + // always be relied upon. The same reasons apply to /proc/cpuinfo as + // well. + if (ReadLongFromFile("/sys/devices/system/cpu/cpu0/tsc_freq_khz", &freq)) { + return freq * 1e3; // Value is kHz. + } + +#if defined(ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY) + // On these platforms, the TSC frequency is the nominal CPU + // frequency. But without having the kernel export it directly + // though /sys/devices/system/cpu/cpu0/tsc_freq_khz, there is no + // other way to reliably get the TSC frequency, so we have to + // measure it ourselves. Some CPUs abuse cpuinfo_max_freq by + // exporting "fake" frequencies for implementing new features. For + // example, Intel's turbo mode is enabled by exposing a p-state + // value with a higher frequency than that of the real TSC + // rate. Because of this, we prefer to measure the TSC rate + // ourselves on i386 and x86-64. + return MeasureTscFrequency(); +#else + + // If CPU scaling is in effect, we want to use the *maximum* + // frequency, not whatever CPU speed some random processor happens + // to be using now. + if (ReadLongFromFile("/sys/devices/system/cpu/cpu0/cpufreq/cpuinfo_max_freq", + &freq)) { + return freq * 1e3; // Value is kHz. + } + + return 1.0; +#endif // !ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY +} + +#endif + +ABSL_CONST_INIT static once_flag init_num_cpus_once; +ABSL_CONST_INIT static int num_cpus = 0; + +// NumCPUs() may be called before main() and before malloc is properly +// initialized, therefore this must not allocate memory. +int NumCPUs() { + base_internal::LowLevelCallOnce( + &init_num_cpus_once, []() { num_cpus = GetNumCPUs(); }); + return num_cpus; +} + +// A default frequency of 0.0 might be dangerous if it is used in division. +ABSL_CONST_INIT static once_flag init_nominal_cpu_frequency_once; +ABSL_CONST_INIT static double nominal_cpu_frequency = 1.0; + +// NominalCPUFrequency() may be called before main() and before malloc is +// properly initialized, therefore this must not allocate memory. +double NominalCPUFrequency() { + base_internal::LowLevelCallOnce( + &init_nominal_cpu_frequency_once, + []() { nominal_cpu_frequency = GetNominalCPUFrequency(); }); + return nominal_cpu_frequency; +} + +#if defined(_WIN32) + +pid_t GetTID() { + return pid_t{GetCurrentThreadId()}; +} + +#elif defined(__linux__) + +#ifndef SYS_gettid +#define SYS_gettid __NR_gettid +#endif + +pid_t GetTID() { + return syscall(SYS_gettid); +} + +#elif defined(__akaros__) + +pid_t GetTID() { + // Akaros has a concept of "vcore context", which is the state the program + // is forced into when we need to make a user-level scheduling decision, or + // run a signal handler. This is analogous to the interrupt context that a + // CPU might enter if it encounters some kind of exception. + // + // There is no current thread context in vcore context, but we need to give + // a reasonable answer if asked for a thread ID (e.g., in a signal handler). + // Thread 0 always exists, so if we are in vcore context, we return that. + // + // Otherwise, we know (since we are using pthreads) that the uthread struct + // current_uthread is pointing to is the first element of a + // struct pthread_tcb, so we extract and return the thread ID from that. + // + // TODO(dcross): Akaros anticipates moving the thread ID to the uthread + // structure at some point. We should modify this code to remove the cast + // when that happens. + if (in_vcore_context()) + return 0; + return reinterpret_cast(current_uthread)->id; +} + +#elif defined(__myriad2__) + +pid_t GetTID() { + uint32_t tid; + rtems_task_ident(RTEMS_SELF, 0, &tid); + return tid; +} + +#else + +// Fallback implementation of GetTID using pthread_getspecific. +ABSL_CONST_INIT static once_flag tid_once; +ABSL_CONST_INIT static pthread_key_t tid_key; +ABSL_CONST_INIT static absl::base_internal::SpinLock tid_lock( + absl::kConstInit, base_internal::SCHEDULE_KERNEL_ONLY); + +// We set a bit per thread in this array to indicate that an ID is in +// use. ID 0 is unused because it is the default value returned by +// pthread_getspecific(). +ABSL_CONST_INIT static std::vector *tid_array + ABSL_GUARDED_BY(tid_lock) = nullptr; +static constexpr int kBitsPerWord = 32; // tid_array is uint32_t. + +// Returns the TID to tid_array. +static void FreeTID(void *v) { + intptr_t tid = reinterpret_cast(v); + int word = tid / kBitsPerWord; + uint32_t mask = ~(1u << (tid % kBitsPerWord)); + absl::base_internal::SpinLockHolder lock(&tid_lock); + assert(0 <= word && static_cast(word) < tid_array->size()); + (*tid_array)[word] &= mask; +} + +static void InitGetTID() { + if (pthread_key_create(&tid_key, FreeTID) != 0) { + // The logging system calls GetTID() so it can't be used here. + perror("pthread_key_create failed"); + throw std::runtime_error("abort()"); + } + + // Initialize tid_array. + absl::base_internal::SpinLockHolder lock(&tid_lock); + tid_array = new std::vector(1); + (*tid_array)[0] = 1; // ID 0 is never-allocated. +} + +// Return a per-thread small integer ID from pthread's thread-specific data. +pid_t GetTID() { + absl::call_once(tid_once, InitGetTID); + + intptr_t tid = reinterpret_cast(pthread_getspecific(tid_key)); + if (tid != 0) { + return tid; + } + + int bit; // tid_array[word] = 1u << bit; + size_t word; + { + // Search for the first unused ID. + absl::base_internal::SpinLockHolder lock(&tid_lock); + // First search for a word in the array that is not all ones. + word = 0; + while (word < tid_array->size() && ~(*tid_array)[word] == 0) { + ++word; + } + if (word == tid_array->size()) { + tid_array->push_back(0); // No space left, add kBitsPerWord more IDs. + } + // Search for a zero bit in the word. + bit = 0; + while (bit < kBitsPerWord && (((*tid_array)[word] >> bit) & 1) != 0) { + ++bit; + } + tid = (word * kBitsPerWord) + bit; + (*tid_array)[word] |= 1u << bit; // Mark the TID as allocated. + } + + if (pthread_setspecific(tid_key, reinterpret_cast(tid)) != 0) { + perror("pthread_setspecific failed"); + throw std::runtime_error("abort()"); + } + + return static_cast(tid); +} + +#endif + +// GetCachedTID() caches the thread ID in thread-local storage (which is a +// userspace construct) to avoid unnecessary system calls. Without this caching, +// it can take roughly 98ns, while it takes roughly 1ns with this caching. +pid_t GetCachedTID() { +#ifdef ABSL_HAVE_THREAD_LOCAL + static thread_local pid_t thread_id = GetTID(); + return thread_id; +#else + return GetTID(); +#endif // ABSL_HAVE_THREAD_LOCAL +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/sysinfo.h b/src/absl/base/internal/sysinfo.h new file mode 100644 index 0000000..119cf1f --- /dev/null +++ b/src/absl/base/internal/sysinfo.h @@ -0,0 +1,74 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file includes routines to find out characteristics +// of the machine a program is running on. It is undoubtedly +// system-dependent. + +// Functions listed here that accept a pid_t as an argument act on the +// current process if the pid_t argument is 0 +// All functions here are thread-hostile due to file caching unless +// commented otherwise. + +#ifndef ABSL_BASE_INTERNAL_SYSINFO_H_ +#define ABSL_BASE_INTERNAL_SYSINFO_H_ + +#ifndef _WIN32 +#include +#endif + +#include + +#include "absl/base/config.h" +#include "absl/base/port.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Nominal core processor cycles per second of each processor. This is _not_ +// necessarily the frequency of the CycleClock counter (see cycleclock.h) +// Thread-safe. +double NominalCPUFrequency(); + +// Number of logical processors (hyperthreads) in system. Thread-safe. +int NumCPUs(); + +// Return the thread id of the current thread, as told by the system. +// No two currently-live threads implemented by the OS shall have the same ID. +// Thread ids of exited threads may be reused. Multiple user-level threads +// may have the same thread ID if multiplexed on the same OS thread. +// +// On Linux, you may send a signal to the resulting ID with kill(). However, +// it is recommended for portability that you use pthread_kill() instead. +#ifdef _WIN32 +// On Windows, process id and thread id are of the same type according to the +// return types of GetProcessId() and GetThreadId() are both DWORD, an unsigned +// 32-bit type. +using pid_t = uint32_t; +#endif +pid_t GetTID(); + +// Like GetTID(), but caches the result in thread-local storage in order +// to avoid unnecessary system calls. Note that there are some cases where +// one must call through to GetTID directly, which is why this exists as a +// separate function. For example, GetCachedTID() is not safe to call in +// an asynchronous signal-handling context nor right after a call to fork(). +pid_t GetCachedTID(); + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_SYSINFO_H_ diff --git a/src/absl/base/internal/thread_annotations.h b/src/absl/base/internal/thread_annotations.h new file mode 100644 index 0000000..4dab6a9 --- /dev/null +++ b/src/absl/base/internal/thread_annotations.h @@ -0,0 +1,271 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: thread_annotations.h +// ----------------------------------------------------------------------------- +// +// WARNING: This is a backwards compatible header and it will be removed after +// the migration to prefixed thread annotations is finished; please include +// "absl/base/thread_annotations.h". +// +// This header file contains macro definitions for thread safety annotations +// that allow developers to document the locking policies of multi-threaded +// code. The annotations can also help program analysis tools to identify +// potential thread safety issues. +// +// These annotations are implemented using compiler attributes. Using the macros +// defined here instead of raw attributes allow for portability and future +// compatibility. +// +// When referring to mutexes in the arguments of the attributes, you should +// use variable names or more complex expressions (e.g. my_object->mutex_) +// that evaluate to a concrete mutex object whenever possible. If the mutex +// you want to refer to is not in scope, you may use a member pointer +// (e.g. &MyClass::mutex_) to refer to a mutex in some (unknown) object. + +#ifndef ABSL_BASE_INTERNAL_THREAD_ANNOTATIONS_H_ +#define ABSL_BASE_INTERNAL_THREAD_ANNOTATIONS_H_ + +#if defined(__clang__) +#define THREAD_ANNOTATION_ATTRIBUTE__(x) __attribute__((x)) +#else +#define THREAD_ANNOTATION_ATTRIBUTE__(x) // no-op +#endif + +// GUARDED_BY() +// +// Documents if a shared field or global variable needs to be protected by a +// mutex. GUARDED_BY() allows the user to specify a particular mutex that +// should be held when accessing the annotated variable. +// +// Although this annotation (and PT_GUARDED_BY, below) cannot be applied to +// local variables, a local variable and its associated mutex can often be +// combined into a small class or struct, thereby allowing the annotation. +// +// Example: +// +// class Foo { +// Mutex mu_; +// int p1_ GUARDED_BY(mu_); +// ... +// }; +#define GUARDED_BY(x) THREAD_ANNOTATION_ATTRIBUTE__(guarded_by(x)) + +// PT_GUARDED_BY() +// +// Documents if the memory location pointed to by a pointer should be guarded +// by a mutex when dereferencing the pointer. +// +// Example: +// class Foo { +// Mutex mu_; +// int *p1_ PT_GUARDED_BY(mu_); +// ... +// }; +// +// Note that a pointer variable to a shared memory location could itself be a +// shared variable. +// +// Example: +// +// // `q_`, guarded by `mu1_`, points to a shared memory location that is +// // guarded by `mu2_`: +// int *q_ GUARDED_BY(mu1_) PT_GUARDED_BY(mu2_); +#define PT_GUARDED_BY(x) THREAD_ANNOTATION_ATTRIBUTE__(pt_guarded_by(x)) + +// ACQUIRED_AFTER() / ACQUIRED_BEFORE() +// +// Documents the acquisition order between locks that can be held +// simultaneously by a thread. For any two locks that need to be annotated +// to establish an acquisition order, only one of them needs the annotation. +// (i.e. You don't have to annotate both locks with both ACQUIRED_AFTER +// and ACQUIRED_BEFORE.) +// +// As with GUARDED_BY, this is only applicable to mutexes that are shared +// fields or global variables. +// +// Example: +// +// Mutex m1_; +// Mutex m2_ ACQUIRED_AFTER(m1_); +#define ACQUIRED_AFTER(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(acquired_after(__VA_ARGS__)) + +#define ACQUIRED_BEFORE(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(acquired_before(__VA_ARGS__)) + +// EXCLUSIVE_LOCKS_REQUIRED() / SHARED_LOCKS_REQUIRED() +// +// Documents a function that expects a mutex to be held prior to entry. +// The mutex is expected to be held both on entry to, and exit from, the +// function. +// +// An exclusive lock allows read-write access to the guarded data member(s), and +// only one thread can acquire a lock exclusively at any one time. A shared lock +// allows read-only access, and any number of threads can acquire a shared lock +// concurrently. +// +// Generally, non-const methods should be annotated with +// EXCLUSIVE_LOCKS_REQUIRED, while const methods should be annotated with +// SHARED_LOCKS_REQUIRED. +// +// Example: +// +// Mutex mu1, mu2; +// int a GUARDED_BY(mu1); +// int b GUARDED_BY(mu2); +// +// void foo() EXCLUSIVE_LOCKS_REQUIRED(mu1, mu2) { ... } +// void bar() const SHARED_LOCKS_REQUIRED(mu1, mu2) { ... } +#define EXCLUSIVE_LOCKS_REQUIRED(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(exclusive_locks_required(__VA_ARGS__)) + +#define SHARED_LOCKS_REQUIRED(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(shared_locks_required(__VA_ARGS__)) + +// LOCKS_EXCLUDED() +// +// Documents the locks acquired in the body of the function. These locks +// cannot be held when calling this function (as Abseil's `Mutex` locks are +// non-reentrant). +#define LOCKS_EXCLUDED(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(locks_excluded(__VA_ARGS__)) + +// LOCK_RETURNED() +// +// Documents a function that returns a mutex without acquiring it. For example, +// a public getter method that returns a pointer to a private mutex should +// be annotated with LOCK_RETURNED. +#define LOCK_RETURNED(x) \ + THREAD_ANNOTATION_ATTRIBUTE__(lock_returned(x)) + +// LOCKABLE +// +// Documents if a class/type is a lockable type (such as the `Mutex` class). +#define LOCKABLE \ + THREAD_ANNOTATION_ATTRIBUTE__(lockable) + +// SCOPED_LOCKABLE +// +// Documents if a class does RAII locking (such as the `MutexLock` class). +// The constructor should use `LOCK_FUNCTION()` to specify the mutex that is +// acquired, and the destructor should use `UNLOCK_FUNCTION()` with no +// arguments; the analysis will assume that the destructor unlocks whatever the +// constructor locked. +#define SCOPED_LOCKABLE \ + THREAD_ANNOTATION_ATTRIBUTE__(scoped_lockable) + +// EXCLUSIVE_LOCK_FUNCTION() +// +// Documents functions that acquire a lock in the body of a function, and do +// not release it. +#define EXCLUSIVE_LOCK_FUNCTION(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(exclusive_lock_function(__VA_ARGS__)) + +// SHARED_LOCK_FUNCTION() +// +// Documents functions that acquire a shared (reader) lock in the body of a +// function, and do not release it. +#define SHARED_LOCK_FUNCTION(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(shared_lock_function(__VA_ARGS__)) + +// UNLOCK_FUNCTION() +// +// Documents functions that expect a lock to be held on entry to the function, +// and release it in the body of the function. +#define UNLOCK_FUNCTION(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(unlock_function(__VA_ARGS__)) + +// EXCLUSIVE_TRYLOCK_FUNCTION() / SHARED_TRYLOCK_FUNCTION() +// +// Documents functions that try to acquire a lock, and return success or failure +// (or a non-boolean value that can be interpreted as a boolean). +// The first argument should be `true` for functions that return `true` on +// success, or `false` for functions that return `false` on success. The second +// argument specifies the mutex that is locked on success. If unspecified, this +// mutex is assumed to be `this`. +#define EXCLUSIVE_TRYLOCK_FUNCTION(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(exclusive_trylock_function(__VA_ARGS__)) + +#define SHARED_TRYLOCK_FUNCTION(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(shared_trylock_function(__VA_ARGS__)) + +// ASSERT_EXCLUSIVE_LOCK() / ASSERT_SHARED_LOCK() +// +// Documents functions that dynamically check to see if a lock is held, and fail +// if it is not held. +#define ASSERT_EXCLUSIVE_LOCK(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(assert_exclusive_lock(__VA_ARGS__)) + +#define ASSERT_SHARED_LOCK(...) \ + THREAD_ANNOTATION_ATTRIBUTE__(assert_shared_lock(__VA_ARGS__)) + +// NO_THREAD_SAFETY_ANALYSIS +// +// Turns off thread safety checking within the body of a particular function. +// This annotation is used to mark functions that are known to be correct, but +// the locking behavior is more complicated than the analyzer can handle. +#define NO_THREAD_SAFETY_ANALYSIS \ + THREAD_ANNOTATION_ATTRIBUTE__(no_thread_safety_analysis) + +//------------------------------------------------------------------------------ +// Tool-Supplied Annotations +//------------------------------------------------------------------------------ + +// TS_UNCHECKED should be placed around lock expressions that are not valid +// C++ syntax, but which are present for documentation purposes. These +// annotations will be ignored by the analysis. +#define TS_UNCHECKED(x) "" + +// TS_FIXME is used to mark lock expressions that are not valid C++ syntax. +// It is used by automated tools to mark and disable invalid expressions. +// The annotation should either be fixed, or changed to TS_UNCHECKED. +#define TS_FIXME(x) "" + +// Like NO_THREAD_SAFETY_ANALYSIS, this turns off checking within the body of +// a particular function. However, this attribute is used to mark functions +// that are incorrect and need to be fixed. It is used by automated tools to +// avoid breaking the build when the analysis is updated. +// Code owners are expected to eventually fix the routine. +#define NO_THREAD_SAFETY_ANALYSIS_FIXME NO_THREAD_SAFETY_ANALYSIS + +// Similar to NO_THREAD_SAFETY_ANALYSIS_FIXME, this macro marks a GUARDED_BY +// annotation that needs to be fixed, because it is producing thread safety +// warning. It disables the GUARDED_BY. +#define GUARDED_BY_FIXME(x) + +// Disables warnings for a single read operation. This can be used to avoid +// warnings when it is known that the read is not actually involved in a race, +// but the compiler cannot confirm that. +#define TS_UNCHECKED_READ(x) thread_safety_analysis::ts_unchecked_read(x) + + +namespace thread_safety_analysis { + +// Takes a reference to a guarded data member, and returns an unguarded +// reference. +template +inline const T& ts_unchecked_read(const T& v) NO_THREAD_SAFETY_ANALYSIS { + return v; +} + +template +inline T& ts_unchecked_read(T& v) NO_THREAD_SAFETY_ANALYSIS { + return v; +} + +} // namespace thread_safety_analysis + +#endif // ABSL_BASE_INTERNAL_THREAD_ANNOTATIONS_H_ diff --git a/src/absl/base/internal/thread_identity.cc b/src/absl/base/internal/thread_identity.cc new file mode 100644 index 0000000..79853f0 --- /dev/null +++ b/src/absl/base/internal/thread_identity.cc @@ -0,0 +1,156 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/thread_identity.h" + +#if !defined(_WIN32) || defined(__MINGW32__) +#include +#include +#endif + +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/call_once.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/spinlock.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +#if ABSL_THREAD_IDENTITY_MODE != ABSL_THREAD_IDENTITY_MODE_USE_CPP11 +namespace { +// Used to co-ordinate one-time creation of our pthread_key +absl::once_flag init_thread_identity_key_once; +pthread_key_t thread_identity_pthread_key; +std::atomic pthread_key_initialized(false); + +void AllocateThreadIdentityKey(ThreadIdentityReclaimerFunction reclaimer) { + pthread_key_create(&thread_identity_pthread_key, reclaimer); + pthread_key_initialized.store(true, std::memory_order_release); +} +} // namespace +#endif + +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_TLS || \ + ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_CPP11 +// The actual TLS storage for a thread's currently associated ThreadIdentity. +// This is referenced by inline accessors in the header. +// "protected" visibility ensures that if multiple instances of Abseil code +// exist within a process (via dlopen() or similar), references to +// thread_identity_ptr from each instance of the code will refer to +// *different* instances of this ptr. +// Apple platforms have the visibility attribute, but issue a compile warning +// that protected visibility is unsupported. +ABSL_CONST_INIT // Must come before __attribute__((visibility("protected"))) +#if ABSL_HAVE_ATTRIBUTE(visibility) && !defined(__APPLE__) +__attribute__((visibility("protected"))) +#endif // ABSL_HAVE_ATTRIBUTE(visibility) && !defined(__APPLE__) +#if ABSL_PER_THREAD_TLS +// Prefer __thread to thread_local as benchmarks indicate it is a bit faster. +ABSL_PER_THREAD_TLS_KEYWORD ThreadIdentity* thread_identity_ptr = nullptr; +#elif defined(ABSL_HAVE_THREAD_LOCAL) +thread_local ThreadIdentity* thread_identity_ptr = nullptr; +#endif // ABSL_PER_THREAD_TLS +#endif // TLS or CPP11 + +void SetCurrentThreadIdentity( + ThreadIdentity* identity, ThreadIdentityReclaimerFunction reclaimer) { + assert(CurrentThreadIdentityIfPresent() == nullptr); + // Associate our destructor. + // NOTE: This call to pthread_setspecific is currently the only immovable + // barrier to CurrentThreadIdentity() always being async signal safe. +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC + // NOTE: Not async-safe. But can be open-coded. + absl::call_once(init_thread_identity_key_once, AllocateThreadIdentityKey, + reclaimer); + +#if defined(__EMSCRIPTEN__) || defined(__MINGW32__) + // Emscripten and MinGW pthread implementations does not support signals. + // See https://kripken.github.io/emscripten-site/docs/porting/pthreads.html + // for more information. + pthread_setspecific(thread_identity_pthread_key, + reinterpret_cast(identity)); +#else + // We must mask signals around the call to setspecific as with current glibc, + // a concurrent getspecific (needed for GetCurrentThreadIdentityIfPresent()) + // may zero our value. + // + // While not officially async-signal safe, getspecific within a signal handler + // is otherwise OK. + sigset_t all_signals; + sigset_t curr_signals; + sigfillset(&all_signals); + pthread_sigmask(SIG_SETMASK, &all_signals, &curr_signals); + pthread_setspecific(thread_identity_pthread_key, + reinterpret_cast(identity)); + pthread_sigmask(SIG_SETMASK, &curr_signals, nullptr); +#endif // !__EMSCRIPTEN__ && !__MINGW32__ + +#elif ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_TLS + // NOTE: Not async-safe. But can be open-coded. + absl::call_once(init_thread_identity_key_once, AllocateThreadIdentityKey, + reclaimer); + pthread_setspecific(thread_identity_pthread_key, + reinterpret_cast(identity)); + thread_identity_ptr = identity; +#elif ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_CPP11 + thread_local std::unique_ptr + holder(identity, reclaimer); + thread_identity_ptr = identity; +#else +#error Unimplemented ABSL_THREAD_IDENTITY_MODE +#endif +} + +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_TLS || \ + ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_CPP11 + +// Please see the comment on `CurrentThreadIdentityIfPresent` in +// thread_identity.h. When we cannot expose thread_local variables in +// headers, we opt for the correct-but-slower option of not inlining this +// function. +#ifndef ABSL_INTERNAL_INLINE_CURRENT_THREAD_IDENTITY_IF_PRESENT +ThreadIdentity* CurrentThreadIdentityIfPresent() { return thread_identity_ptr; } +#endif +#endif + +void ClearCurrentThreadIdentity() { +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_TLS || \ + ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_CPP11 + thread_identity_ptr = nullptr; +#elif ABSL_THREAD_IDENTITY_MODE == \ + ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC + // pthread_setspecific expected to clear value on destruction + assert(CurrentThreadIdentityIfPresent() == nullptr); +#endif +} + +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC +ThreadIdentity* CurrentThreadIdentityIfPresent() { + bool initialized = pthread_key_initialized.load(std::memory_order_acquire); + if (!initialized) { + return nullptr; + } + return reinterpret_cast( + pthread_getspecific(thread_identity_pthread_key)); +} +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/thread_identity.h b/src/absl/base/internal/thread_identity.h new file mode 100644 index 0000000..659694b --- /dev/null +++ b/src/absl/base/internal/thread_identity.h @@ -0,0 +1,265 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Each active thread has an ThreadIdentity that may represent the thread in +// various level interfaces. ThreadIdentity objects are never deallocated. +// When a thread terminates, its ThreadIdentity object may be reused for a +// thread created later. + +#ifndef ABSL_BASE_INTERNAL_THREAD_IDENTITY_H_ +#define ABSL_BASE_INTERNAL_THREAD_IDENTITY_H_ + +#ifndef _WIN32 +#include +// Defines __GOOGLE_GRTE_VERSION__ (via glibc-specific features.h) when +// supported. +#include +#endif + +#include +#include + +#include "absl/base/config.h" +#include "absl/base/internal/per_thread_tls.h" +#include "absl/base/optimization.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +struct SynchLocksHeld; +struct SynchWaitParams; + +namespace base_internal { + +class SpinLock; +struct ThreadIdentity; + +// Used by the implementation of absl::Mutex and absl::CondVar. +struct PerThreadSynch { + // The internal representation of absl::Mutex and absl::CondVar rely + // on the alignment of PerThreadSynch. Both store the address of the + // PerThreadSynch in the high-order bits of their internal state, + // which means the low kLowZeroBits of the address of PerThreadSynch + // must be zero. + static constexpr int kLowZeroBits = 8; + static constexpr int kAlignment = 1 << kLowZeroBits; + + // Returns the associated ThreadIdentity. + // This can be implemented as a cast because we guarantee + // PerThreadSynch is the first element of ThreadIdentity. + ThreadIdentity* thread_identity() { + return reinterpret_cast(this); + } + + PerThreadSynch *next; // Circular waiter queue; initialized to 0. + PerThreadSynch *skip; // If non-zero, all entries in Mutex queue + // up to and including "skip" have same + // condition as this, and will be woken later + bool may_skip; // if false while on mutex queue, a mutex unlocker + // is using this PerThreadSynch as a terminator. Its + // skip field must not be filled in because the loop + // might then skip over the terminator. + bool wake; // This thread is to be woken from a Mutex. + // If "x" is on a waiter list for a mutex, "x->cond_waiter" is true iff the + // waiter is waiting on the mutex as part of a CV Wait or Mutex Await. + // + // The value of "x->cond_waiter" is meaningless if "x" is not on a + // Mutex waiter list. + bool cond_waiter; + bool maybe_unlocking; // Valid at head of Mutex waiter queue; + // true if UnlockSlow could be searching + // for a waiter to wake. Used for an optimization + // in Enqueue(). true is always a valid value. + // Can be reset to false when the unlocker or any + // writer releases the lock, or a reader fully + // releases the lock. It may not be set to false + // by a reader that decrements the count to + // non-zero. protected by mutex spinlock + bool suppress_fatal_errors; // If true, try to proceed even in the face + // of broken invariants. This is used within + // fatal signal handlers to improve the + // chances of debug logging information being + // output successfully. + int priority; // Priority of thread (updated every so often). + + // State values: + // kAvailable: This PerThreadSynch is available. + // kQueued: This PerThreadSynch is unavailable, it's currently queued on a + // Mutex or CondVar waistlist. + // + // Transitions from kQueued to kAvailable require a release + // barrier. This is needed as a waiter may use "state" to + // independently observe that it's no longer queued. + // + // Transitions from kAvailable to kQueued require no barrier, they + // are externally ordered by the Mutex. + enum State { + kAvailable, + kQueued + }; + std::atomic state; + + // The wait parameters of the current wait. waitp is null if the + // thread is not waiting. Transitions from null to non-null must + // occur before the enqueue commit point (state = kQueued in + // Enqueue() and CondVarEnqueue()). Transitions from non-null to + // null must occur after the wait is finished (state = kAvailable in + // Mutex::Block() and CondVar::WaitCommon()). This field may be + // changed only by the thread that describes this PerThreadSynch. A + // special case is Fer(), which calls Enqueue() on another thread, + // but with an identical SynchWaitParams pointer, thus leaving the + // pointer unchanged. + SynchWaitParams* waitp; + + intptr_t readers; // Number of readers in mutex. + + // When priority will next be read (cycles). + int64_t next_priority_read_cycles; + + // Locks held; used during deadlock detection. + // Allocated in Synch_GetAllLocks() and freed in ReclaimThreadIdentity(). + SynchLocksHeld *all_locks; +}; + +// The instances of this class are allocated in NewThreadIdentity() with an +// alignment of PerThreadSynch::kAlignment. +struct ThreadIdentity { + // Must be the first member. The Mutex implementation requires that + // the PerThreadSynch object associated with each thread is + // PerThreadSynch::kAlignment aligned. We provide this alignment on + // ThreadIdentity itself. + PerThreadSynch per_thread_synch; + + // Private: Reserved for absl::synchronization_internal::Waiter. + struct WaiterState { + alignas(void*) char data[128]; + } waiter_state; + + // Used by PerThreadSem::{Get,Set}ThreadBlockedCounter(). + std::atomic* blocked_count_ptr; + + // The following variables are mostly read/written just by the + // thread itself. The only exception is that these are read by + // a ticker thread as a hint. + std::atomic ticker; // Tick counter, incremented once per second. + std::atomic wait_start; // Ticker value when thread started waiting. + std::atomic is_idle; // Has thread become idle yet? + + ThreadIdentity* next; +}; + +// Returns the ThreadIdentity object representing the calling thread; guaranteed +// to be unique for its lifetime. The returned object will remain valid for the +// program's lifetime; although it may be re-assigned to a subsequent thread. +// If one does not exist, return nullptr instead. +// +// Does not malloc(*), and is async-signal safe. +// [*] Technically pthread_setspecific() does malloc on first use; however this +// is handled internally within tcmalloc's initialization already. +// +// New ThreadIdentity objects can be constructed and associated with a thread +// by calling GetOrCreateCurrentThreadIdentity() in per-thread-sem.h. +ThreadIdentity* CurrentThreadIdentityIfPresent(); + +using ThreadIdentityReclaimerFunction = void (*)(void*); + +// Sets the current thread identity to the given value. 'reclaimer' is a +// pointer to the global function for cleaning up instances on thread +// destruction. +void SetCurrentThreadIdentity(ThreadIdentity* identity, + ThreadIdentityReclaimerFunction reclaimer); + +// Removes the currently associated ThreadIdentity from the running thread. +// This must be called from inside the ThreadIdentityReclaimerFunction, and only +// from that function. +void ClearCurrentThreadIdentity(); + +// May be chosen at compile time via: -DABSL_FORCE_THREAD_IDENTITY_MODE= +#ifdef ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC +#error ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC cannot be directly set +#else +#define ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC 0 +#endif + +#ifdef ABSL_THREAD_IDENTITY_MODE_USE_TLS +#error ABSL_THREAD_IDENTITY_MODE_USE_TLS cannot be directly set +#else +#define ABSL_THREAD_IDENTITY_MODE_USE_TLS 1 +#endif + +#ifdef ABSL_THREAD_IDENTITY_MODE_USE_CPP11 +#error ABSL_THREAD_IDENTITY_MODE_USE_CPP11 cannot be directly set +#else +#define ABSL_THREAD_IDENTITY_MODE_USE_CPP11 2 +#endif + +#ifdef ABSL_THREAD_IDENTITY_MODE +#error ABSL_THREAD_IDENTITY_MODE cannot be directly set +#elif defined(ABSL_FORCE_THREAD_IDENTITY_MODE) +#define ABSL_THREAD_IDENTITY_MODE ABSL_FORCE_THREAD_IDENTITY_MODE +#elif defined(_WIN32) && !defined(__MINGW32__) +#define ABSL_THREAD_IDENTITY_MODE ABSL_THREAD_IDENTITY_MODE_USE_CPP11 +#elif defined(__APPLE__) && defined(ABSL_HAVE_THREAD_LOCAL) +#define ABSL_THREAD_IDENTITY_MODE ABSL_THREAD_IDENTITY_MODE_USE_CPP11 +#elif ABSL_PER_THREAD_TLS && defined(__GOOGLE_GRTE_VERSION__) && \ + (__GOOGLE_GRTE_VERSION__ >= 20140228L) +// Support for async-safe TLS was specifically added in GRTEv4. It's not +// present in the upstream eglibc. +// Note: Current default for production systems. +#define ABSL_THREAD_IDENTITY_MODE ABSL_THREAD_IDENTITY_MODE_USE_TLS +#else +#define ABSL_THREAD_IDENTITY_MODE \ + ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC +#endif + +#if ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_TLS || \ + ABSL_THREAD_IDENTITY_MODE == ABSL_THREAD_IDENTITY_MODE_USE_CPP11 + +#if ABSL_PER_THREAD_TLS +ABSL_CONST_INIT extern ABSL_PER_THREAD_TLS_KEYWORD ThreadIdentity* + thread_identity_ptr; +#elif defined(ABSL_HAVE_THREAD_LOCAL) +ABSL_CONST_INIT extern thread_local ThreadIdentity* thread_identity_ptr; +#else +#error Thread-local storage not detected on this platform +#endif + +// thread_local variables cannot be in headers exposed by DLLs or in certain +// build configurations on Apple platforms. However, it is important for +// performance reasons in general that `CurrentThreadIdentityIfPresent` be +// inlined. In the other cases we opt to have the function not be inlined. Note +// that `CurrentThreadIdentityIfPresent` is declared above so we can exclude +// this entire inline definition. +#if !defined(__APPLE__) && !defined(ABSL_BUILD_DLL) && \ + !defined(ABSL_CONSUME_DLL) +#define ABSL_INTERNAL_INLINE_CURRENT_THREAD_IDENTITY_IF_PRESENT 1 +#endif + +#ifdef ABSL_INTERNAL_INLINE_CURRENT_THREAD_IDENTITY_IF_PRESENT +inline ThreadIdentity* CurrentThreadIdentityIfPresent() { + return thread_identity_ptr; +} +#endif + +#elif ABSL_THREAD_IDENTITY_MODE != \ + ABSL_THREAD_IDENTITY_MODE_USE_POSIX_SETSPECIFIC +#error Unknown ABSL_THREAD_IDENTITY_MODE +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_THREAD_IDENTITY_H_ diff --git a/src/absl/base/internal/throw_delegate.cc b/src/absl/base/internal/throw_delegate.cc new file mode 100644 index 0000000..c260ff1 --- /dev/null +++ b/src/absl/base/internal/throw_delegate.cc @@ -0,0 +1,212 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/throw_delegate.h" + +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// NOTE: The various STL exception throwing functions are placed within the +// #ifdef blocks so the symbols aren't exposed on platforms that don't support +// them, such as the Android NDK. For example, ANGLE fails to link when building +// within AOSP without them, since the STL functions don't exist. +namespace { +#ifdef ABSL_HAVE_EXCEPTIONS +template +[[noreturn]] void Throw(const T& error) { + throw error; +} +#endif +} // namespace + +void ThrowStdLogicError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::logic_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdLogicError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::logic_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} +void ThrowStdInvalidArgument(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::invalid_argument(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdInvalidArgument(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::invalid_argument(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdDomainError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::domain_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdDomainError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::domain_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdLengthError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::length_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdLengthError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::length_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdOutOfRange(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::out_of_range(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdOutOfRange(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::out_of_range(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdRuntimeError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::runtime_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdRuntimeError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::runtime_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdRangeError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::range_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdRangeError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::range_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdOverflowError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::overflow_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdOverflowError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::overflow_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdUnderflowError(const std::string& what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::underflow_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg.c_str()); + std::abort(); +#endif +} +void ThrowStdUnderflowError(const char* what_arg) { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::underflow_error(what_arg)); +#else + ABSL_RAW_LOG(FATAL, "%s", what_arg); + std::abort(); +#endif +} + +void ThrowStdBadFunctionCall() { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::bad_function_call()); +#else + std::abort(); +#endif +} + +void ThrowStdBadAlloc() { +#ifdef ABSL_HAVE_EXCEPTIONS + Throw(std::bad_alloc()); +#else + std::abort(); +#endif +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/internal/throw_delegate.h b/src/absl/base/internal/throw_delegate.h new file mode 100644 index 0000000..075f527 --- /dev/null +++ b/src/absl/base/internal/throw_delegate.h @@ -0,0 +1,75 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_THROW_DELEGATE_H_ +#define ABSL_BASE_INTERNAL_THROW_DELEGATE_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Helper functions that allow throwing exceptions consistently from anywhere. +// The main use case is for header-based libraries (eg templates), as they will +// be built by many different targets with their own compiler options. +// In particular, this will allow a safe way to throw exceptions even if the +// caller is compiled with -fno-exceptions. This is intended for implementing +// things like map<>::at(), which the standard documents as throwing an +// exception on error. +// +// Using other techniques like #if tricks could lead to ODR violations. +// +// You shouldn't use it unless you're writing code that you know will be built +// both with and without exceptions and you need to conform to an interface +// that uses exceptions. + +[[noreturn]] void ThrowStdLogicError(const std::string& what_arg); +[[noreturn]] void ThrowStdLogicError(const char* what_arg); +[[noreturn]] void ThrowStdInvalidArgument(const std::string& what_arg); +[[noreturn]] void ThrowStdInvalidArgument(const char* what_arg); +[[noreturn]] void ThrowStdDomainError(const std::string& what_arg); +[[noreturn]] void ThrowStdDomainError(const char* what_arg); +[[noreturn]] void ThrowStdLengthError(const std::string& what_arg); +[[noreturn]] void ThrowStdLengthError(const char* what_arg); +[[noreturn]] void ThrowStdOutOfRange(const std::string& what_arg); +[[noreturn]] void ThrowStdOutOfRange(const char* what_arg); +[[noreturn]] void ThrowStdRuntimeError(const std::string& what_arg); +[[noreturn]] void ThrowStdRuntimeError(const char* what_arg); +[[noreturn]] void ThrowStdRangeError(const std::string& what_arg); +[[noreturn]] void ThrowStdRangeError(const char* what_arg); +[[noreturn]] void ThrowStdOverflowError(const std::string& what_arg); +[[noreturn]] void ThrowStdOverflowError(const char* what_arg); +[[noreturn]] void ThrowStdUnderflowError(const std::string& what_arg); +[[noreturn]] void ThrowStdUnderflowError(const char* what_arg); + +[[noreturn]] void ThrowStdBadFunctionCall(); +[[noreturn]] void ThrowStdBadAlloc(); + +// ThrowStdBadArrayNewLength() cannot be consistently supported because +// std::bad_array_new_length is missing in libstdc++ until 4.9.0. +// https://gcc.gnu.org/onlinedocs/gcc-4.8.3/libstdc++/api/a01379_source.html +// https://gcc.gnu.org/onlinedocs/gcc-4.9.0/libstdc++/api/a01327_source.html +// libcxx (as of 3.2) and msvc (as of 2015) both have it. +// [[noreturn]] void ThrowStdBadArrayNewLength(); + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_INTERNAL_THROW_DELEGATE_H_ diff --git a/src/absl/base/internal/tsan_mutex_interface.h b/src/absl/base/internal/tsan_mutex_interface.h new file mode 100644 index 0000000..39207d8 --- /dev/null +++ b/src/absl/base/internal/tsan_mutex_interface.h @@ -0,0 +1,68 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This file is intended solely for spinlock.h. +// It provides ThreadSanitizer annotations for custom mutexes. +// See for meaning of these annotations. + +#ifndef ABSL_BASE_INTERNAL_TSAN_MUTEX_INTERFACE_H_ +#define ABSL_BASE_INTERNAL_TSAN_MUTEX_INTERFACE_H_ + +#include "absl/base/config.h" + +// ABSL_INTERNAL_HAVE_TSAN_INTERFACE +// Macro intended only for internal use. +// +// Checks whether LLVM Thread Sanitizer interfaces are available. +// First made available in LLVM 5.0 (Sep 2017). +#ifdef ABSL_INTERNAL_HAVE_TSAN_INTERFACE +#error "ABSL_INTERNAL_HAVE_TSAN_INTERFACE cannot be directly set." +#endif + +#if defined(ABSL_HAVE_THREAD_SANITIZER) && defined(__has_include) +#if __has_include() +#define ABSL_INTERNAL_HAVE_TSAN_INTERFACE 1 +#endif +#endif + +#ifdef ABSL_INTERNAL_HAVE_TSAN_INTERFACE +#include + +#define ABSL_TSAN_MUTEX_CREATE __tsan_mutex_create +#define ABSL_TSAN_MUTEX_DESTROY __tsan_mutex_destroy +#define ABSL_TSAN_MUTEX_PRE_LOCK __tsan_mutex_pre_lock +#define ABSL_TSAN_MUTEX_POST_LOCK __tsan_mutex_post_lock +#define ABSL_TSAN_MUTEX_PRE_UNLOCK __tsan_mutex_pre_unlock +#define ABSL_TSAN_MUTEX_POST_UNLOCK __tsan_mutex_post_unlock +#define ABSL_TSAN_MUTEX_PRE_SIGNAL __tsan_mutex_pre_signal +#define ABSL_TSAN_MUTEX_POST_SIGNAL __tsan_mutex_post_signal +#define ABSL_TSAN_MUTEX_PRE_DIVERT __tsan_mutex_pre_divert +#define ABSL_TSAN_MUTEX_POST_DIVERT __tsan_mutex_post_divert + +#else + +#define ABSL_TSAN_MUTEX_CREATE(...) +#define ABSL_TSAN_MUTEX_DESTROY(...) +#define ABSL_TSAN_MUTEX_PRE_LOCK(...) +#define ABSL_TSAN_MUTEX_POST_LOCK(...) +#define ABSL_TSAN_MUTEX_PRE_UNLOCK(...) +#define ABSL_TSAN_MUTEX_POST_UNLOCK(...) +#define ABSL_TSAN_MUTEX_PRE_SIGNAL(...) +#define ABSL_TSAN_MUTEX_POST_SIGNAL(...) +#define ABSL_TSAN_MUTEX_PRE_DIVERT(...) +#define ABSL_TSAN_MUTEX_POST_DIVERT(...) + +#endif + +#endif // ABSL_BASE_INTERNAL_TSAN_MUTEX_INTERFACE_H_ diff --git a/src/absl/base/internal/unaligned_access.h b/src/absl/base/internal/unaligned_access.h new file mode 100644 index 0000000..093dd9b --- /dev/null +++ b/src/absl/base/internal/unaligned_access.h @@ -0,0 +1,82 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_BASE_INTERNAL_UNALIGNED_ACCESS_H_ +#define ABSL_BASE_INTERNAL_UNALIGNED_ACCESS_H_ + +#include + +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" + +// unaligned APIs + +// Portable handling of unaligned loads, stores, and copies. + +// The unaligned API is C++ only. The declarations use C++ features +// (namespaces, inline) which are absent or incompatible in C. +#if defined(__cplusplus) +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +inline uint16_t UnalignedLoad16(const void *p) { + uint16_t t; + memcpy(&t, p, sizeof t); + return t; +} + +inline uint32_t UnalignedLoad32(const void *p) { + uint32_t t; + memcpy(&t, p, sizeof t); + return t; +} + +inline uint64_t UnalignedLoad64(const void *p) { + uint64_t t; + memcpy(&t, p, sizeof t); + return t; +} + +inline void UnalignedStore16(void *p, uint16_t v) { memcpy(p, &v, sizeof v); } + +inline void UnalignedStore32(void *p, uint32_t v) { memcpy(p, &v, sizeof v); } + +inline void UnalignedStore64(void *p, uint64_t v) { memcpy(p, &v, sizeof v); } + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#define ABSL_INTERNAL_UNALIGNED_LOAD16(_p) \ + (absl::base_internal::UnalignedLoad16(_p)) +#define ABSL_INTERNAL_UNALIGNED_LOAD32(_p) \ + (absl::base_internal::UnalignedLoad32(_p)) +#define ABSL_INTERNAL_UNALIGNED_LOAD64(_p) \ + (absl::base_internal::UnalignedLoad64(_p)) + +#define ABSL_INTERNAL_UNALIGNED_STORE16(_p, _val) \ + (absl::base_internal::UnalignedStore16(_p, _val)) +#define ABSL_INTERNAL_UNALIGNED_STORE32(_p, _val) \ + (absl::base_internal::UnalignedStore32(_p, _val)) +#define ABSL_INTERNAL_UNALIGNED_STORE64(_p, _val) \ + (absl::base_internal::UnalignedStore64(_p, _val)) + +#endif // defined(__cplusplus), end of unaligned API + +#endif // ABSL_BASE_INTERNAL_UNALIGNED_ACCESS_H_ diff --git a/src/absl/base/internal/unscaledcycleclock.cc b/src/absl/base/internal/unscaledcycleclock.cc new file mode 100644 index 0000000..b1c396c --- /dev/null +++ b/src/absl/base/internal/unscaledcycleclock.cc @@ -0,0 +1,153 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/internal/unscaledcycleclock.h" + +#if ABSL_USE_UNSCALED_CYCLECLOCK + +#if defined(_WIN32) +#include +#endif + +#if defined(__powerpc__) || defined(__ppc__) +#ifdef __GLIBC__ +#include +#elif defined(__FreeBSD__) +// clang-format off +// This order does actually matter =(. +#include +#include +// clang-format on + +#include "absl/base/call_once.h" +#endif +#endif + +#include "absl/base/internal/sysinfo.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +#if defined(__i386__) + +int64_t UnscaledCycleClock::Now() { + int64_t ret; + __asm__ volatile("rdtsc" : "=A"(ret)); + return ret; +} + +double UnscaledCycleClock::Frequency() { + return base_internal::NominalCPUFrequency(); +} + +#elif defined(__x86_64__) + +double UnscaledCycleClock::Frequency() { + return base_internal::NominalCPUFrequency(); +} + +#elif defined(__powerpc__) || defined(__ppc__) + +int64_t UnscaledCycleClock::Now() { +#ifdef __GLIBC__ + return __ppc_get_timebase(); +#else +#ifdef __powerpc64__ + int64_t tbr; + asm volatile("mfspr %0, 268" : "=r"(tbr)); + return tbr; +#else + int32_t tbu, tbl, tmp; + asm volatile( + "0:\n" + "mftbu %[hi32]\n" + "mftb %[lo32]\n" + "mftbu %[tmp]\n" + "cmpw %[tmp],%[hi32]\n" + "bne 0b\n" + : [ hi32 ] "=r"(tbu), [ lo32 ] "=r"(tbl), [ tmp ] "=r"(tmp)); + return (static_cast(tbu) << 32) | tbl; +#endif +#endif +} + +double UnscaledCycleClock::Frequency() { +#ifdef __GLIBC__ + return __ppc_get_timebase_freq(); +#elif defined(_AIX) + // This is the same constant value as returned by + // __ppc_get_timebase_freq(). + return static_cast(512000000); +#elif defined(__FreeBSD__) + static once_flag init_timebase_frequency_once; + static double timebase_frequency = 0.0; + base_internal::LowLevelCallOnce(&init_timebase_frequency_once, [&]() { + size_t length = sizeof(timebase_frequency); + sysctlbyname("kern.timecounter.tc.timebase.frequency", &timebase_frequency, + &length, nullptr, 0); + }); + return timebase_frequency; +#else +#error Must implement UnscaledCycleClock::Frequency() +#endif +} + +#elif defined(__aarch64__) + +// System timer of ARMv8 runs at a different frequency than the CPU's. +// The frequency is fixed, typically in the range 1-50MHz. It can be +// read at CNTFRQ special register. We assume the OS has set up +// the virtual timer properly. +int64_t UnscaledCycleClock::Now() { + int64_t virtual_timer_value; + asm volatile("mrs %0, cntvct_el0" : "=r"(virtual_timer_value)); + return virtual_timer_value; +} + +double UnscaledCycleClock::Frequency() { + uint64_t aarch64_timer_frequency; + asm volatile("mrs %0, cntfrq_el0" : "=r"(aarch64_timer_frequency)); + return aarch64_timer_frequency; +} + +#elif defined(__riscv) + +int64_t UnscaledCycleClock::Now() { + int64_t virtual_timer_value; + asm volatile("rdcycle %0" : "=r"(virtual_timer_value)); + return virtual_timer_value; +} + +double UnscaledCycleClock::Frequency() { + return base_internal::NominalCPUFrequency(); +} + +#elif defined(_M_IX86) || defined(_M_X64) + +#pragma intrinsic(__rdtsc) + +int64_t UnscaledCycleClock::Now() { return __rdtsc(); } + +double UnscaledCycleClock::Frequency() { + return base_internal::NominalCPUFrequency(); +} + +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_USE_UNSCALED_CYCLECLOCK diff --git a/src/absl/base/internal/unscaledcycleclock.h b/src/absl/base/internal/unscaledcycleclock.h new file mode 100644 index 0000000..2cbeae3 --- /dev/null +++ b/src/absl/base/internal/unscaledcycleclock.h @@ -0,0 +1,133 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// UnscaledCycleClock +// An UnscaledCycleClock yields the value and frequency of a cycle counter +// that increments at a rate that is approximately constant. +// This class is for internal use only, you should consider using CycleClock +// instead. +// +// Notes: +// The cycle counter frequency is not necessarily the core clock frequency. +// That is, CycleCounter cycles are not necessarily "CPU cycles". +// +// An arbitrary offset may have been added to the counter at power on. +// +// On some platforms, the rate and offset of the counter may differ +// slightly when read from different CPUs of a multiprocessor. Usually, +// we try to ensure that the operating system adjusts values periodically +// so that values agree approximately. If you need stronger guarantees, +// consider using alternate interfaces. +// +// The CPU is not required to maintain the ordering of a cycle counter read +// with respect to surrounding instructions. + +#ifndef ABSL_BASE_INTERNAL_UNSCALEDCYCLECLOCK_H_ +#define ABSL_BASE_INTERNAL_UNSCALEDCYCLECLOCK_H_ + +#include + +#if defined(__APPLE__) +#include +#endif + +#include "absl/base/port.h" + +// The following platforms have an implementation of a hardware counter. +#if defined(__i386__) || defined(__x86_64__) || defined(__aarch64__) || \ + defined(__powerpc__) || defined(__ppc__) || defined(__riscv) || \ + defined(_M_IX86) || (defined(_M_X64) && !defined(_M_ARM64EC)) +#define ABSL_HAVE_UNSCALED_CYCLECLOCK_IMPLEMENTATION 1 +#else +#define ABSL_HAVE_UNSCALED_CYCLECLOCK_IMPLEMENTATION 0 +#endif + +// The following platforms often disable access to the hardware +// counter (through a sandbox) even if the underlying hardware has a +// usable counter. The CycleTimer interface also requires a *scaled* +// CycleClock that runs at atleast 1 MHz. We've found some Android +// ARM64 devices where this is not the case, so we disable it by +// default on Android ARM64. +#if defined(__native_client__) || (defined(__APPLE__)) || \ + (defined(__ANDROID__) && defined(__aarch64__)) +#define ABSL_USE_UNSCALED_CYCLECLOCK_DEFAULT 0 +#else +#define ABSL_USE_UNSCALED_CYCLECLOCK_DEFAULT 1 +#endif + +// UnscaledCycleClock is an optional internal feature. +// Use "#if ABSL_USE_UNSCALED_CYCLECLOCK" to test for its presence. +// Can be overridden at compile-time via -DABSL_USE_UNSCALED_CYCLECLOCK=0|1 +#if !defined(ABSL_USE_UNSCALED_CYCLECLOCK) +#define ABSL_USE_UNSCALED_CYCLECLOCK \ + (ABSL_HAVE_UNSCALED_CYCLECLOCK_IMPLEMENTATION && \ + ABSL_USE_UNSCALED_CYCLECLOCK_DEFAULT) +#endif + +#if ABSL_USE_UNSCALED_CYCLECLOCK + +// This macro can be used to test if UnscaledCycleClock::Frequency() +// is NominalCPUFrequency() on a particular platform. +#if (defined(__i386__) || defined(__x86_64__) || defined(__riscv) || \ + defined(_M_IX86) || defined(_M_X64)) +#define ABSL_INTERNAL_UNSCALED_CYCLECLOCK_FREQUENCY_IS_CPU_FREQUENCY +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace time_internal { +class UnscaledCycleClockWrapperForGetCurrentTime; +} // namespace time_internal + +namespace base_internal { +class CycleClock; +class UnscaledCycleClockWrapperForInitializeFrequency; + +class UnscaledCycleClock { + private: + UnscaledCycleClock() = delete; + + // Return the value of a cycle counter that counts at a rate that is + // approximately constant. + static int64_t Now(); + + // Return the how much UnscaledCycleClock::Now() increases per second. + // This is not necessarily the core CPU clock frequency. + // It may be the nominal value report by the kernel, rather than a measured + // value. + static double Frequency(); + + // Allowed users + friend class base_internal::CycleClock; + friend class time_internal::UnscaledCycleClockWrapperForGetCurrentTime; + friend class base_internal::UnscaledCycleClockWrapperForInitializeFrequency; +}; + +#if defined(__x86_64__) + +inline int64_t UnscaledCycleClock::Now() { + uint64_t low, high; + __asm__ volatile("rdtsc" : "=a"(low), "=d"(high)); + return (high << 32) | low; +} + +#endif + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_USE_UNSCALED_CYCLECLOCK + +#endif // ABSL_BASE_INTERNAL_UNSCALEDCYCLECLOCK_H_ diff --git a/src/absl/base/log_severity.cc b/src/absl/base/log_severity.cc new file mode 100644 index 0000000..60a8fc1 --- /dev/null +++ b/src/absl/base/log_severity.cc @@ -0,0 +1,55 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/base/log_severity.h" + +#include + +#include "absl/base/attributes.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +std::ostream& operator<<(std::ostream& os, absl::LogSeverity s) { + if (s == absl::NormalizeLogSeverity(s)) return os << absl::LogSeverityName(s); + return os << "absl::LogSeverity(" << static_cast(s) << ")"; +} + +std::ostream& operator<<(std::ostream& os, absl::LogSeverityAtLeast s) { + switch (s) { + case absl::LogSeverityAtLeast::kInfo: + case absl::LogSeverityAtLeast::kWarning: + case absl::LogSeverityAtLeast::kError: + case absl::LogSeverityAtLeast::kFatal: + return os << ">=" << static_cast(s); + case absl::LogSeverityAtLeast::kInfinity: + return os << "INFINITY"; + } + return os; +} + +std::ostream& operator<<(std::ostream& os, absl::LogSeverityAtMost s) { + switch (s) { + case absl::LogSeverityAtMost::kInfo: + case absl::LogSeverityAtMost::kWarning: + case absl::LogSeverityAtMost::kError: + case absl::LogSeverityAtMost::kFatal: + return os << "<=" << static_cast(s); + case absl::LogSeverityAtMost::kNegativeInfinity: + return os << "NEGATIVE_INFINITY"; + } + return os; +} +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/base/log_severity.h b/src/absl/base/log_severity.h new file mode 100644 index 0000000..8bdca38 --- /dev/null +++ b/src/absl/base/log_severity.h @@ -0,0 +1,172 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_BASE_LOG_SEVERITY_H_ +#define ABSL_BASE_LOG_SEVERITY_H_ + +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// absl::LogSeverity +// +// Four severity levels are defined. Logging APIs should terminate the program +// when a message is logged at severity `kFatal`; the other levels have no +// special semantics. +// +// Values other than the four defined levels (e.g. produced by `static_cast`) +// are valid, but their semantics when passed to a function, macro, or flag +// depend on the function, macro, or flag. The usual behavior is to normalize +// such values to a defined severity level, however in some cases values other +// than the defined levels are useful for comparison. +// +// Example: +// +// // Effectively disables all logging: +// SetMinLogLevel(static_cast(100)); +// +// Abseil flags may be defined with type `LogSeverity`. Dependency layering +// constraints require that the `AbslParseFlag()` overload be declared and +// defined in the flags library itself rather than here. The `AbslUnparseFlag()` +// overload is defined there as well for consistency. +// +// absl::LogSeverity Flag String Representation +// +// An `absl::LogSeverity` has a string representation used for parsing +// command-line flags based on the enumerator name (e.g. `kFatal`) or +// its unprefixed name (without the `k`) in any case-insensitive form. (E.g. +// "FATAL", "fatal" or "Fatal" are all valid.) Unparsing such flags produces an +// unprefixed string representation in all caps (e.g. "FATAL") or an integer. +// +// Additionally, the parser accepts arbitrary integers (as if the type were +// `int`). +// +// Examples: +// +// --my_log_level=kInfo +// --my_log_level=INFO +// --my_log_level=info +// --my_log_level=0 +// +// Unparsing a flag produces the same result as `absl::LogSeverityName()` for +// the standard levels and a base-ten integer otherwise. +enum class LogSeverity : int { + kInfo = 0, + kWarning = 1, + kError = 2, + kFatal = 3, +}; + +// LogSeverities() +// +// Returns an iterable of all standard `absl::LogSeverity` values, ordered from +// least to most severe. +constexpr std::array LogSeverities() { + return {{absl::LogSeverity::kInfo, absl::LogSeverity::kWarning, + absl::LogSeverity::kError, absl::LogSeverity::kFatal}}; +} + +// LogSeverityName() +// +// Returns the all-caps string representation (e.g. "INFO") of the specified +// severity level if it is one of the standard levels and "UNKNOWN" otherwise. +constexpr const char* LogSeverityName(absl::LogSeverity s) { + return s == absl::LogSeverity::kInfo + ? "INFO" + : s == absl::LogSeverity::kWarning + ? "WARNING" + : s == absl::LogSeverity::kError + ? "ERROR" + : s == absl::LogSeverity::kFatal ? "FATAL" : "UNKNOWN"; +} + +// NormalizeLogSeverity() +// +// Values less than `kInfo` normalize to `kInfo`; values greater than `kFatal` +// normalize to `kError` (**NOT** `kFatal`). +constexpr absl::LogSeverity NormalizeLogSeverity(absl::LogSeverity s) { + return s < absl::LogSeverity::kInfo + ? absl::LogSeverity::kInfo + : s > absl::LogSeverity::kFatal ? absl::LogSeverity::kError : s; +} +constexpr absl::LogSeverity NormalizeLogSeverity(int s) { + return absl::NormalizeLogSeverity(static_cast(s)); +} + +// operator<< +// +// The exact representation of a streamed `absl::LogSeverity` is deliberately +// unspecified; do not rely on it. +std::ostream& operator<<(std::ostream& os, absl::LogSeverity s); + +// Enums representing a lower bound for LogSeverity. APIs that only operate on +// messages of at least a certain level (for example, `SetMinLogLevel()`) use +// this type to specify that level. absl::LogSeverityAtLeast::kInfinity is +// a level above all threshold levels and therefore no log message will +// ever meet this threshold. +enum class LogSeverityAtLeast : int { + kInfo = static_cast(absl::LogSeverity::kInfo), + kWarning = static_cast(absl::LogSeverity::kWarning), + kError = static_cast(absl::LogSeverity::kError), + kFatal = static_cast(absl::LogSeverity::kFatal), + kInfinity = 1000, +}; + +std::ostream& operator<<(std::ostream& os, absl::LogSeverityAtLeast s); + +// Enums representing an upper bound for LogSeverity. APIs that only operate on +// messages of at most a certain level (for example, buffer all messages at or +// below a certain level) use this type to specify that level. +// absl::LogSeverityAtMost::kNegativeInfinity is a level below all threshold +// levels and therefore will exclude all log messages. +enum class LogSeverityAtMost : int { + kNegativeInfinity = -1000, + kInfo = static_cast(absl::LogSeverity::kInfo), + kWarning = static_cast(absl::LogSeverity::kWarning), + kError = static_cast(absl::LogSeverity::kError), + kFatal = static_cast(absl::LogSeverity::kFatal), +}; + +std::ostream& operator<<(std::ostream& os, absl::LogSeverityAtMost s); + +#define COMPOP(op1, op2, T) \ + constexpr bool operator op1(absl::T lhs, absl::LogSeverity rhs) { \ + return static_cast(lhs) op1 rhs; \ + } \ + constexpr bool operator op2(absl::LogSeverity lhs, absl::T rhs) { \ + return lhs op2 static_cast(rhs); \ + } + +// Comparisons between `LogSeverity` and `LogSeverityAtLeast`/ +// `LogSeverityAtMost` are only supported in one direction. +// Valid checks are: +// LogSeverity >= LogSeverityAtLeast +// LogSeverity < LogSeverityAtLeast +// LogSeverity <= LogSeverityAtMost +// LogSeverity > LogSeverityAtMost +COMPOP(>, <, LogSeverityAtLeast) +COMPOP(<=, >=, LogSeverityAtLeast) +COMPOP(<, >, LogSeverityAtMost) +COMPOP(>=, <=, LogSeverityAtMost) +#undef COMPOP + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_LOG_SEVERITY_H_ diff --git a/src/absl/base/macros.h b/src/absl/base/macros.h new file mode 100644 index 0000000..3e085a9 --- /dev/null +++ b/src/absl/base/macros.h @@ -0,0 +1,158 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: macros.h +// ----------------------------------------------------------------------------- +// +// This header file defines the set of language macros used within Abseil code. +// For the set of macros used to determine supported compilers and platforms, +// see absl/base/config.h instead. +// +// This code is compiled directly on many platforms, including client +// platforms like Windows, Mac, and embedded systems. Before making +// any changes here, make sure that you're not breaking any platforms. + +#ifndef ABSL_BASE_MACROS_H_ +#define ABSL_BASE_MACROS_H_ + +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" + +// ABSL_ARRAYSIZE() +// +// Returns the number of elements in an array as a compile-time constant, which +// can be used in defining new arrays. If you use this macro on a pointer by +// mistake, you will get a compile-time error. +#define ABSL_ARRAYSIZE(array) \ + (sizeof(::absl::macros_internal::ArraySizeHelper(array))) + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace macros_internal { +// Note: this internal template function declaration is used by ABSL_ARRAYSIZE. +// The function doesn't need a definition, as we only use its type. +template +auto ArraySizeHelper(const T (&array)[N]) -> char (&)[N]; +} // namespace macros_internal +ABSL_NAMESPACE_END +} // namespace absl + +// ABSL_BAD_CALL_IF() +// +// Used on a function overload to trap bad calls: any call that matches the +// overload will cause a compile-time error. This macro uses a clang-specific +// "enable_if" attribute, as described at +// https://clang.llvm.org/docs/AttributeReference.html#enable-if +// +// Overloads which use this macro should be bracketed by +// `#ifdef ABSL_BAD_CALL_IF`. +// +// Example: +// +// int isdigit(int c); +// #ifdef ABSL_BAD_CALL_IF +// int isdigit(int c) +// ABSL_BAD_CALL_IF(c <= -1 || c > 255, +// "'c' must have the value of an unsigned char or EOF"); +// #endif // ABSL_BAD_CALL_IF +#if ABSL_HAVE_ATTRIBUTE(enable_if) +#define ABSL_BAD_CALL_IF(expr, msg) \ + __attribute__((enable_if(expr, "Bad call trap"), unavailable(msg))) +#endif + +// ABSL_ASSERT() +// +// In C++11, `assert` can't be used portably within constexpr functions. +// ABSL_ASSERT functions as a runtime assert but works in C++11 constexpr +// functions. Example: +// +// constexpr double Divide(double a, double b) { +// return ABSL_ASSERT(b != 0), a / b; +// } +// +// This macro is inspired by +// https://akrzemi1.wordpress.com/2017/05/18/asserts-in-constexpr-functions/ +#if defined(NDEBUG) +#define ABSL_ASSERT(expr) \ + (false ? static_cast(expr) : static_cast(0)) +#else +#define ABSL_ASSERT(expr) \ + (ABSL_PREDICT_TRUE((expr)) ? static_cast(0) \ + : [] { assert(false && #expr); }()) // NOLINT +#endif + +// `ABSL_INTERNAL_HARDENING_ABORT()` controls how `ABSL_HARDENING_ASSERT()` +// aborts the program in release mode (when NDEBUG is defined). The +// implementation should abort the program as quickly as possible and ideally it +// should not be possible to ignore the abort request. +#if (ABSL_HAVE_BUILTIN(__builtin_trap) && \ + ABSL_HAVE_BUILTIN(__builtin_unreachable)) || \ + (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_INTERNAL_HARDENING_ABORT() \ + do { \ + __builtin_trap(); \ + __builtin_unreachable(); \ + } while (false) +#else +#define ABSL_INTERNAL_HARDENING_ABORT() abort() +#endif + +// ABSL_HARDENING_ASSERT() +// +// `ABSL_HARDENING_ASSERT()` is like `ABSL_ASSERT()`, but used to implement +// runtime assertions that should be enabled in hardened builds even when +// `NDEBUG` is defined. +// +// When `NDEBUG` is not defined, `ABSL_HARDENING_ASSERT()` is identical to +// `ABSL_ASSERT()`. +// +// See `ABSL_OPTION_HARDENED` in `absl/base/options.h` for more information on +// hardened mode. +#if ABSL_OPTION_HARDENED == 1 && defined(NDEBUG) +#define ABSL_HARDENING_ASSERT(expr) \ + (ABSL_PREDICT_TRUE((expr)) ? static_cast(0) \ + : [] { ABSL_INTERNAL_HARDENING_ABORT(); }()) +#else +#define ABSL_HARDENING_ASSERT(expr) ABSL_ASSERT(expr) +#endif + +#ifdef ABSL_HAVE_EXCEPTIONS +#define ABSL_INTERNAL_TRY try +#define ABSL_INTERNAL_CATCH_ANY catch (...) +#define ABSL_INTERNAL_RETHROW do { throw; } while (false) +#else // ABSL_HAVE_EXCEPTIONS +#define ABSL_INTERNAL_TRY if (true) +#define ABSL_INTERNAL_CATCH_ANY else if (false) +#define ABSL_INTERNAL_RETHROW do {} while (false) +#endif // ABSL_HAVE_EXCEPTIONS + +// `ABSL_INTERNAL_UNREACHABLE` is an unreachable statement. A program which +// reaches one has undefined behavior, and the compiler may optimize +// accordingly. +#if defined(__GNUC__) || ABSL_HAVE_BUILTIN(__builtin_unreachable) +#define ABSL_INTERNAL_UNREACHABLE __builtin_unreachable() +#elif defined(_MSC_VER) +#define ABSL_INTERNAL_UNREACHABLE __assume(0) +#else +#define ABSL_INTERNAL_UNREACHABLE +#endif + +#endif // ABSL_BASE_MACROS_H_ diff --git a/src/absl/base/optimization.h b/src/absl/base/optimization.h new file mode 100644 index 0000000..db5cc09 --- /dev/null +++ b/src/absl/base/optimization.h @@ -0,0 +1,252 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: optimization.h +// ----------------------------------------------------------------------------- +// +// This header file defines portable macros for performance optimization. + +#ifndef ABSL_BASE_OPTIMIZATION_H_ +#define ABSL_BASE_OPTIMIZATION_H_ + +#include + +#include "absl/base/config.h" + +// ABSL_BLOCK_TAIL_CALL_OPTIMIZATION +// +// Instructs the compiler to avoid optimizing tail-call recursion. This macro is +// useful when you wish to preserve the existing function order within a stack +// trace for logging, debugging, or profiling purposes. +// +// Example: +// +// int f() { +// int result = g(); +// ABSL_BLOCK_TAIL_CALL_OPTIMIZATION(); +// return result; +// } +#if defined(__pnacl__) +#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() if (volatile int x = 0) { (void)x; } +#elif defined(__clang__) +// Clang will not tail call given inline volatile assembly. +#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __asm__ __volatile__("") +#elif defined(__GNUC__) +// GCC will not tail call given inline volatile assembly. +#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __asm__ __volatile__("") +#elif defined(_MSC_VER) +#include +// The __nop() intrinsic blocks the optimisation. +#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() __nop() +#else +#define ABSL_BLOCK_TAIL_CALL_OPTIMIZATION() if (volatile int x = 0) { (void)x; } +#endif + +// ABSL_CACHELINE_SIZE +// +// Explicitly defines the size of the L1 cache for purposes of alignment. +// Setting the cacheline size allows you to specify that certain objects be +// aligned on a cacheline boundary with `ABSL_CACHELINE_ALIGNED` declarations. +// (See below.) +// +// NOTE: this macro should be replaced with the following C++17 features, when +// those are generally available: +// +// * `std::hardware_constructive_interference_size` +// * `std::hardware_destructive_interference_size` +// +// See http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0154r1.html +// for more information. +#if defined(__GNUC__) +// Cache line alignment +#if defined(__i386__) || defined(__x86_64__) +#define ABSL_CACHELINE_SIZE 64 +#elif defined(__powerpc64__) +#define ABSL_CACHELINE_SIZE 128 +#elif defined(__aarch64__) +// We would need to read special register ctr_el0 to find out L1 dcache size. +// This value is a good estimate based on a real aarch64 machine. +#define ABSL_CACHELINE_SIZE 64 +#elif defined(__arm__) +// Cache line sizes for ARM: These values are not strictly correct since +// cache line sizes depend on implementations, not architectures. There +// are even implementations with cache line sizes configurable at boot +// time. +#if defined(__ARM_ARCH_5T__) +#define ABSL_CACHELINE_SIZE 32 +#elif defined(__ARM_ARCH_7A__) +#define ABSL_CACHELINE_SIZE 64 +#endif +#endif + +#ifndef ABSL_CACHELINE_SIZE +// A reasonable default guess. Note that overestimates tend to waste more +// space, while underestimates tend to waste more time. +#define ABSL_CACHELINE_SIZE 64 +#endif + +// ABSL_CACHELINE_ALIGNED +// +// Indicates that the declared object be cache aligned using +// `ABSL_CACHELINE_SIZE` (see above). Cacheline aligning objects allows you to +// load a set of related objects in the L1 cache for performance improvements. +// Cacheline aligning objects properly allows constructive memory sharing and +// prevents destructive (or "false") memory sharing. +// +// NOTE: callers should replace uses of this macro with `alignas()` using +// `std::hardware_constructive_interference_size` and/or +// `std::hardware_destructive_interference_size` when C++17 becomes available to +// them. +// +// See http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2016/p0154r1.html +// for more information. +// +// On some compilers, `ABSL_CACHELINE_ALIGNED` expands to an `__attribute__` +// or `__declspec` attribute. For compilers where this is not known to work, +// the macro expands to nothing. +// +// No further guarantees are made here. The result of applying the macro +// to variables and types is always implementation-defined. +// +// WARNING: It is easy to use this attribute incorrectly, even to the point +// of causing bugs that are difficult to diagnose, crash, etc. It does not +// of itself guarantee that objects are aligned to a cache line. +// +// NOTE: Some compilers are picky about the locations of annotations such as +// this attribute, so prefer to put it at the beginning of your declaration. +// For example, +// +// ABSL_CACHELINE_ALIGNED static Foo* foo = ... +// +// class ABSL_CACHELINE_ALIGNED Bar { ... +// +// Recommendations: +// +// 1) Consult compiler documentation; this comment is not kept in sync as +// toolchains evolve. +// 2) Verify your use has the intended effect. This often requires inspecting +// the generated machine code. +// 3) Prefer applying this attribute to individual variables. Avoid +// applying it to types. This tends to localize the effect. +#define ABSL_CACHELINE_ALIGNED __attribute__((aligned(ABSL_CACHELINE_SIZE))) +#elif defined(_MSC_VER) +#define ABSL_CACHELINE_SIZE 64 +#define ABSL_CACHELINE_ALIGNED __declspec(align(ABSL_CACHELINE_SIZE)) +#else +#define ABSL_CACHELINE_SIZE 64 +#define ABSL_CACHELINE_ALIGNED +#endif + +// ABSL_PREDICT_TRUE, ABSL_PREDICT_FALSE +// +// Enables the compiler to prioritize compilation using static analysis for +// likely paths within a boolean branch. +// +// Example: +// +// if (ABSL_PREDICT_TRUE(expression)) { +// return result; // Faster if more likely +// } else { +// return 0; +// } +// +// Compilers can use the information that a certain branch is not likely to be +// taken (for instance, a CHECK failure) to optimize for the common case in +// the absence of better information (ie. compiling gcc with `-fprofile-arcs`). +// +// Recommendation: Modern CPUs dynamically predict branch execution paths, +// typically with accuracy greater than 97%. As a result, annotating every +// branch in a codebase is likely counterproductive; however, annotating +// specific branches that are both hot and consistently mispredicted is likely +// to yield performance improvements. +#if ABSL_HAVE_BUILTIN(__builtin_expect) || \ + (defined(__GNUC__) && !defined(__clang__)) +#define ABSL_PREDICT_FALSE(x) (__builtin_expect(false || (x), false)) +#define ABSL_PREDICT_TRUE(x) (__builtin_expect(false || (x), true)) +#else +#define ABSL_PREDICT_FALSE(x) (x) +#define ABSL_PREDICT_TRUE(x) (x) +#endif + +// ABSL_ASSUME(cond) +// +// Informs the compiler that a condition is always true and that it can assume +// it to be true for optimization purposes. +// +// WARNING: If the condition is false, the program can produce undefined and +// potentially dangerous behavior. +// +// In !NDEBUG mode, the condition is checked with an assert(). +// +// NOTE: The expression must not have side effects, as it may only be evaluated +// in some compilation modes and not others. Some compilers may issue a warning +// if the compiler cannot prove the expression has no side effects. For example, +// the expression should not use a function call since the compiler cannot prove +// that a function call does not have side effects. +// +// Example: +// +// int x = ...; +// ABSL_ASSUME(x >= 0); +// // The compiler can optimize the division to a simple right shift using the +// // assumption specified above. +// int y = x / 16; +// +#if !defined(NDEBUG) +#define ABSL_ASSUME(cond) assert(cond) +#elif ABSL_HAVE_BUILTIN(__builtin_assume) +#define ABSL_ASSUME(cond) __builtin_assume(cond) +#elif defined(__GNUC__) || ABSL_HAVE_BUILTIN(__builtin_unreachable) +#define ABSL_ASSUME(cond) \ + do { \ + if (!(cond)) __builtin_unreachable(); \ + } while (0) +#elif defined(_MSC_VER) +#define ABSL_ASSUME(cond) __assume(cond) +#else +#define ABSL_ASSUME(cond) \ + do { \ + static_cast(false && (cond)); \ + } while (0) +#endif + +// ABSL_INTERNAL_UNIQUE_SMALL_NAME(cond) +// This macro forces small unique name on a static file level symbols like +// static local variables or static functions. This is intended to be used in +// macro definitions to optimize the cost of generated code. Do NOT use it on +// symbols exported from translation unit since it may cause a link time +// conflict. +// +// Example: +// +// #define MY_MACRO(txt) +// namespace { +// char VeryVeryLongVarName[] ABSL_INTERNAL_UNIQUE_SMALL_NAME() = txt; +// const char* VeryVeryLongFuncName() ABSL_INTERNAL_UNIQUE_SMALL_NAME(); +// const char* VeryVeryLongFuncName() { return txt; } +// } +// + +#if defined(__GNUC__) +#define ABSL_INTERNAL_UNIQUE_SMALL_NAME2(x) #x +#define ABSL_INTERNAL_UNIQUE_SMALL_NAME1(x) ABSL_INTERNAL_UNIQUE_SMALL_NAME2(x) +#define ABSL_INTERNAL_UNIQUE_SMALL_NAME() \ + asm(ABSL_INTERNAL_UNIQUE_SMALL_NAME1(.absl.__COUNTER__)) +#else +#define ABSL_INTERNAL_UNIQUE_SMALL_NAME() +#endif + +#endif // ABSL_BASE_OPTIMIZATION_H_ diff --git a/src/absl/base/options.h b/src/absl/base/options.h new file mode 100644 index 0000000..bc59847 --- /dev/null +++ b/src/absl/base/options.h @@ -0,0 +1,238 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: options.h +// ----------------------------------------------------------------------------- +// +// This file contains Abseil configuration options for setting specific +// implementations instead of letting Abseil determine which implementation to +// use at compile-time. Setting these options may be useful for package or build +// managers who wish to guarantee ABI stability within binary builds (which are +// otherwise difficult to enforce). +// +// *** IMPORTANT NOTICE FOR PACKAGE MANAGERS: It is important that +// maintainers of package managers who wish to package Abseil read and +// understand this file! *** +// +// Abseil contains a number of possible configuration endpoints, based on +// parameters such as the detected platform, language version, or command-line +// flags used to invoke the underlying binary. As is the case with all +// libraries, binaries which contain Abseil code must ensure that separate +// packages use the same compiled copy of Abseil to avoid a diamond dependency +// problem, which can occur if two packages built with different Abseil +// configuration settings are linked together. Diamond dependency problems in +// C++ may manifest as violations to the One Definition Rule (ODR) (resulting in +// linker errors), or undefined behavior (resulting in crashes). +// +// Diamond dependency problems can be avoided if all packages utilize the same +// exact version of Abseil. Building from source code with the same compilation +// parameters is the easiest way to avoid such dependency problems. However, for +// package managers who cannot control such compilation parameters, we are +// providing the file to allow you to inject ABI (Application Binary Interface) +// stability across builds. Settings options in this file will neither change +// API nor ABI, providing a stable copy of Abseil between packages. +// +// Care must be taken to keep options within these configurations isolated +// from any other dynamic settings, such as command-line flags which could alter +// these options. This file is provided specifically to help build and package +// managers provide a stable copy of Abseil within their libraries and binaries; +// other developers should not have need to alter the contents of this file. +// +// ----------------------------------------------------------------------------- +// Usage +// ----------------------------------------------------------------------------- +// +// For any particular package release, set the appropriate definitions within +// this file to whatever value makes the most sense for your package(s). Note +// that, by default, most of these options, at the moment, affect the +// implementation of types; future options may affect other implementation +// details. +// +// NOTE: the defaults within this file all assume that Abseil can select the +// proper Abseil implementation at compile-time, which will not be sufficient +// to guarantee ABI stability to package managers. + +#ifndef ABSL_BASE_OPTIONS_H_ +#define ABSL_BASE_OPTIONS_H_ + +// Include a standard library header to allow configuration based on the +// standard library in use. +#ifdef __cplusplus +#include +#endif + +// ----------------------------------------------------------------------------- +// Type Compatibility Options +// ----------------------------------------------------------------------------- +// +// ABSL_OPTION_USE_STD_ANY +// +// This option controls whether absl::any is implemented as an alias to +// std::any, or as an independent implementation. +// +// A value of 0 means to use Abseil's implementation. This requires only C++11 +// support, and is expected to work on every toolchain we support. +// +// A value of 1 means to use an alias to std::any. This requires that all code +// using Abseil is built in C++17 mode or later. +// +// A value of 2 means to detect the C++ version being used to compile Abseil, +// and use an alias only if a working std::any is available. This option is +// useful when you are building your entire program, including all of its +// dependencies, from source. It should not be used otherwise -- for example, +// if you are distributing Abseil in a binary package manager -- since in +// mode 2, absl::any will name a different type, with a different mangled name +// and binary layout, depending on the compiler flags passed by the end user. +// For more info, see https://abseil.io/about/design/dropin-types. +// +// User code should not inspect this macro. To check in the preprocessor if +// absl::any is a typedef of std::any, use the feature macro ABSL_USES_STD_ANY. + +#define ABSL_OPTION_USE_STD_ANY 2 + + +// ABSL_OPTION_USE_STD_OPTIONAL +// +// This option controls whether absl::optional is implemented as an alias to +// std::optional, or as an independent implementation. +// +// A value of 0 means to use Abseil's implementation. This requires only C++11 +// support, and is expected to work on every toolchain we support. +// +// A value of 1 means to use an alias to std::optional. This requires that all +// code using Abseil is built in C++17 mode or later. +// +// A value of 2 means to detect the C++ version being used to compile Abseil, +// and use an alias only if a working std::optional is available. This option +// is useful when you are building your program from source. It should not be +// used otherwise -- for example, if you are distributing Abseil in a binary +// package manager -- since in mode 2, absl::optional will name a different +// type, with a different mangled name and binary layout, depending on the +// compiler flags passed by the end user. For more info, see +// https://abseil.io/about/design/dropin-types. + +// User code should not inspect this macro. To check in the preprocessor if +// absl::optional is a typedef of std::optional, use the feature macro +// ABSL_USES_STD_OPTIONAL. + +#define ABSL_OPTION_USE_STD_OPTIONAL 2 + + +// ABSL_OPTION_USE_STD_STRING_VIEW +// +// This option controls whether absl::string_view is implemented as an alias to +// std::string_view, or as an independent implementation. +// +// A value of 0 means to use Abseil's implementation. This requires only C++11 +// support, and is expected to work on every toolchain we support. +// +// A value of 1 means to use an alias to std::string_view. This requires that +// all code using Abseil is built in C++17 mode or later. +// +// A value of 2 means to detect the C++ version being used to compile Abseil, +// and use an alias only if a working std::string_view is available. This +// option is useful when you are building your program from source. It should +// not be used otherwise -- for example, if you are distributing Abseil in a +// binary package manager -- since in mode 2, absl::string_view will name a +// different type, with a different mangled name and binary layout, depending on +// the compiler flags passed by the end user. For more info, see +// https://abseil.io/about/design/dropin-types. +// +// User code should not inspect this macro. To check in the preprocessor if +// absl::string_view is a typedef of std::string_view, use the feature macro +// ABSL_USES_STD_STRING_VIEW. + +#define ABSL_OPTION_USE_STD_STRING_VIEW 2 + +// ABSL_OPTION_USE_STD_VARIANT +// +// This option controls whether absl::variant is implemented as an alias to +// std::variant, or as an independent implementation. +// +// A value of 0 means to use Abseil's implementation. This requires only C++11 +// support, and is expected to work on every toolchain we support. +// +// A value of 1 means to use an alias to std::variant. This requires that all +// code using Abseil is built in C++17 mode or later. +// +// A value of 2 means to detect the C++ version being used to compile Abseil, +// and use an alias only if a working std::variant is available. This option +// is useful when you are building your program from source. It should not be +// used otherwise -- for example, if you are distributing Abseil in a binary +// package manager -- since in mode 2, absl::variant will name a different +// type, with a different mangled name and binary layout, depending on the +// compiler flags passed by the end user. For more info, see +// https://abseil.io/about/design/dropin-types. +// +// User code should not inspect this macro. To check in the preprocessor if +// absl::variant is a typedef of std::variant, use the feature macro +// ABSL_USES_STD_VARIANT. + +#define ABSL_OPTION_USE_STD_VARIANT 2 + + +// ABSL_OPTION_USE_INLINE_NAMESPACE +// ABSL_OPTION_INLINE_NAMESPACE_NAME +// +// These options controls whether all entities in the absl namespace are +// contained within an inner inline namespace. This does not affect the +// user-visible API of Abseil, but it changes the mangled names of all symbols. +// +// This can be useful as a version tag if you are distributing Abseil in +// precompiled form. This will prevent a binary library build of Abseil with +// one inline namespace being used with headers configured with a different +// inline namespace name. Binary packagers are reminded that Abseil does not +// guarantee any ABI stability in Abseil, so any update of Abseil or +// configuration change in such a binary package should be combined with a +// new, unique value for the inline namespace name. +// +// A value of 0 means not to use inline namespaces. +// +// A value of 1 means to use an inline namespace with the given name inside +// namespace absl. If this is set, ABSL_OPTION_INLINE_NAMESPACE_NAME must also +// be changed to a new, unique identifier name. In particular "head" is not +// allowed. + +#define ABSL_OPTION_USE_INLINE_NAMESPACE 1 +#define ABSL_OPTION_INLINE_NAMESPACE_NAME lts_20220623 + +// ABSL_OPTION_HARDENED +// +// This option enables a "hardened" build in release mode (in this context, +// release mode is defined as a build where the `NDEBUG` macro is defined). +// +// A value of 0 means that "hardened" mode is not enabled. +// +// A value of 1 means that "hardened" mode is enabled. +// +// Hardened builds have additional security checks enabled when `NDEBUG` is +// defined. Defining `NDEBUG` is normally used to turn `assert()` macro into a +// no-op, as well as disabling other bespoke program consistency checks. By +// defining ABSL_OPTION_HARDENED to 1, a select set of checks remain enabled in +// release mode. These checks guard against programming errors that may lead to +// security vulnerabilities. In release mode, when one of these programming +// errors is encountered, the program will immediately abort, possibly without +// any attempt at logging. +// +// The checks enabled by this option are not free; they do incur runtime cost. +// +// The checks enabled by this option are always active when `NDEBUG` is not +// defined, even in the case when ABSL_OPTION_HARDENED is defined to 0. The +// checks enabled by this option may abort the program in a different way and +// log additional information when `NDEBUG` is not defined. + +#define ABSL_OPTION_HARDENED 0 + +#endif // ABSL_BASE_OPTIONS_H_ diff --git a/src/absl/base/policy_checks.h b/src/absl/base/policy_checks.h new file mode 100644 index 0000000..06b3243 --- /dev/null +++ b/src/absl/base/policy_checks.h @@ -0,0 +1,111 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: policy_checks.h +// ----------------------------------------------------------------------------- +// +// This header enforces a minimum set of policies at build time, such as the +// supported compiler and library versions. Unsupported configurations are +// reported with `#error`. This enforcement is best effort, so successfully +// compiling this header does not guarantee a supported configuration. + +#ifndef ABSL_BASE_POLICY_CHECKS_H_ +#define ABSL_BASE_POLICY_CHECKS_H_ + +// Included for the __GLIBC_PREREQ macro used below. +#include + +// Included for the _STLPORT_VERSION macro used below. +#if defined(__cplusplus) +#include +#endif + +// ----------------------------------------------------------------------------- +// Operating System Check +// ----------------------------------------------------------------------------- + +#if defined(__CYGWIN__) +#error "Cygwin is not supported." +#endif + +// ----------------------------------------------------------------------------- +// Toolchain Check +// ----------------------------------------------------------------------------- + +// We support MSVC++ 14.0 update 2 and later. +// This minimum will go up. +#if defined(_MSC_FULL_VER) && _MSC_FULL_VER < 190023918 && !defined(__clang__) +#error "This package requires Visual Studio 2015 Update 2 or higher." +#endif + +// We support gcc 4.7 and later. +// This minimum will go up. +#if defined(__GNUC__) && !defined(__clang__) +#if __GNUC__ < 4 || (__GNUC__ == 4 && __GNUC_MINOR__ < 7) +#error "This package requires gcc 4.7 or higher." +#endif +#endif + +// We support Apple Xcode clang 4.2.1 (version 421.11.65) and later. +// This corresponds to Apple Xcode version 4.5. +// This minimum will go up. +#if defined(__apple_build_version__) && __apple_build_version__ < 4211165 +#error "This package requires __apple_build_version__ of 4211165 or higher." +#endif + +// ----------------------------------------------------------------------------- +// C++ Version Check +// ----------------------------------------------------------------------------- + +// Enforce C++11 as the minimum. Note that Visual Studio has not +// advanced __cplusplus despite being good enough for our purposes, so +// so we exempt it from the check. +#if defined(__cplusplus) && !defined(_MSC_VER) +#if __cplusplus < 201103L +#error "C++ versions less than C++11 are not supported." +#endif +#endif + +// ----------------------------------------------------------------------------- +// Standard Library Check +// ----------------------------------------------------------------------------- + +#if defined(_STLPORT_VERSION) +#error "STLPort is not supported." +#endif + +// ----------------------------------------------------------------------------- +// `char` Size Check +// ----------------------------------------------------------------------------- + +// Abseil currently assumes CHAR_BIT == 8. If you would like to use Abseil on a +// platform where this is not the case, please provide us with the details about +// your platform so we can consider relaxing this requirement. +#if CHAR_BIT != 8 +#error "Abseil assumes CHAR_BIT == 8." +#endif + +// ----------------------------------------------------------------------------- +// `int` Size Check +// ----------------------------------------------------------------------------- + +// Abseil currently assumes that an int is 4 bytes. If you would like to use +// Abseil on a platform where this is not the case, please provide us with the +// details about your platform so we can consider relaxing this requirement. +#if INT_MAX < 2147483647 +#error "Abseil assumes that int is at least 4 bytes. " +#endif + +#endif // ABSL_BASE_POLICY_CHECKS_H_ diff --git a/src/absl/base/port.h b/src/absl/base/port.h new file mode 100644 index 0000000..5bc4d6c --- /dev/null +++ b/src/absl/base/port.h @@ -0,0 +1,25 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This files is a forwarding header for other headers containing various +// portability macros and functions. + +#ifndef ABSL_BASE_PORT_H_ +#define ABSL_BASE_PORT_H_ + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/base/optimization.h" + +#endif // ABSL_BASE_PORT_H_ diff --git a/src/absl/base/thread_annotations.h b/src/absl/base/thread_annotations.h new file mode 100644 index 0000000..bc8a620 --- /dev/null +++ b/src/absl/base/thread_annotations.h @@ -0,0 +1,335 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: thread_annotations.h +// ----------------------------------------------------------------------------- +// +// This header file contains macro definitions for thread safety annotations +// that allow developers to document the locking policies of multi-threaded +// code. The annotations can also help program analysis tools to identify +// potential thread safety issues. +// +// These annotations are implemented using compiler attributes. Using the macros +// defined here instead of raw attributes allow for portability and future +// compatibility. +// +// When referring to mutexes in the arguments of the attributes, you should +// use variable names or more complex expressions (e.g. my_object->mutex_) +// that evaluate to a concrete mutex object whenever possible. If the mutex +// you want to refer to is not in scope, you may use a member pointer +// (e.g. &MyClass::mutex_) to refer to a mutex in some (unknown) object. + +#ifndef ABSL_BASE_THREAD_ANNOTATIONS_H_ +#define ABSL_BASE_THREAD_ANNOTATIONS_H_ + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +// TODO(mbonadei): Remove after the backward compatibility period. +#include "absl/base/internal/thread_annotations.h" // IWYU pragma: export + +// ABSL_GUARDED_BY() +// +// Documents if a shared field or global variable needs to be protected by a +// mutex. ABSL_GUARDED_BY() allows the user to specify a particular mutex that +// should be held when accessing the annotated variable. +// +// Although this annotation (and ABSL_PT_GUARDED_BY, below) cannot be applied to +// local variables, a local variable and its associated mutex can often be +// combined into a small class or struct, thereby allowing the annotation. +// +// Example: +// +// class Foo { +// Mutex mu_; +// int p1_ ABSL_GUARDED_BY(mu_); +// ... +// }; +#if ABSL_HAVE_ATTRIBUTE(guarded_by) +#define ABSL_GUARDED_BY(x) __attribute__((guarded_by(x))) +#else +#define ABSL_GUARDED_BY(x) +#endif + +// ABSL_PT_GUARDED_BY() +// +// Documents if the memory location pointed to by a pointer should be guarded +// by a mutex when dereferencing the pointer. +// +// Example: +// class Foo { +// Mutex mu_; +// int *p1_ ABSL_PT_GUARDED_BY(mu_); +// ... +// }; +// +// Note that a pointer variable to a shared memory location could itself be a +// shared variable. +// +// Example: +// +// // `q_`, guarded by `mu1_`, points to a shared memory location that is +// // guarded by `mu2_`: +// int *q_ ABSL_GUARDED_BY(mu1_) ABSL_PT_GUARDED_BY(mu2_); +#if ABSL_HAVE_ATTRIBUTE(pt_guarded_by) +#define ABSL_PT_GUARDED_BY(x) __attribute__((pt_guarded_by(x))) +#else +#define ABSL_PT_GUARDED_BY(x) +#endif + +// ABSL_ACQUIRED_AFTER() / ABSL_ACQUIRED_BEFORE() +// +// Documents the acquisition order between locks that can be held +// simultaneously by a thread. For any two locks that need to be annotated +// to establish an acquisition order, only one of them needs the annotation. +// (i.e. You don't have to annotate both locks with both ABSL_ACQUIRED_AFTER +// and ABSL_ACQUIRED_BEFORE.) +// +// As with ABSL_GUARDED_BY, this is only applicable to mutexes that are shared +// fields or global variables. +// +// Example: +// +// Mutex m1_; +// Mutex m2_ ABSL_ACQUIRED_AFTER(m1_); +#if ABSL_HAVE_ATTRIBUTE(acquired_after) +#define ABSL_ACQUIRED_AFTER(...) __attribute__((acquired_after(__VA_ARGS__))) +#else +#define ABSL_ACQUIRED_AFTER(...) +#endif + +#if ABSL_HAVE_ATTRIBUTE(acquired_before) +#define ABSL_ACQUIRED_BEFORE(...) __attribute__((acquired_before(__VA_ARGS__))) +#else +#define ABSL_ACQUIRED_BEFORE(...) +#endif + +// ABSL_EXCLUSIVE_LOCKS_REQUIRED() / ABSL_SHARED_LOCKS_REQUIRED() +// +// Documents a function that expects a mutex to be held prior to entry. +// The mutex is expected to be held both on entry to, and exit from, the +// function. +// +// An exclusive lock allows read-write access to the guarded data member(s), and +// only one thread can acquire a lock exclusively at any one time. A shared lock +// allows read-only access, and any number of threads can acquire a shared lock +// concurrently. +// +// Generally, non-const methods should be annotated with +// ABSL_EXCLUSIVE_LOCKS_REQUIRED, while const methods should be annotated with +// ABSL_SHARED_LOCKS_REQUIRED. +// +// Example: +// +// Mutex mu1, mu2; +// int a ABSL_GUARDED_BY(mu1); +// int b ABSL_GUARDED_BY(mu2); +// +// void foo() ABSL_EXCLUSIVE_LOCKS_REQUIRED(mu1, mu2) { ... } +// void bar() const ABSL_SHARED_LOCKS_REQUIRED(mu1, mu2) { ... } +#if ABSL_HAVE_ATTRIBUTE(exclusive_locks_required) +#define ABSL_EXCLUSIVE_LOCKS_REQUIRED(...) \ + __attribute__((exclusive_locks_required(__VA_ARGS__))) +#else +#define ABSL_EXCLUSIVE_LOCKS_REQUIRED(...) +#endif + +#if ABSL_HAVE_ATTRIBUTE(shared_locks_required) +#define ABSL_SHARED_LOCKS_REQUIRED(...) \ + __attribute__((shared_locks_required(__VA_ARGS__))) +#else +#define ABSL_SHARED_LOCKS_REQUIRED(...) +#endif + +// ABSL_LOCKS_EXCLUDED() +// +// Documents the locks that cannot be held by callers of this function, as they +// might be acquired by this function (Abseil's `Mutex` locks are +// non-reentrant). +#if ABSL_HAVE_ATTRIBUTE(locks_excluded) +#define ABSL_LOCKS_EXCLUDED(...) __attribute__((locks_excluded(__VA_ARGS__))) +#else +#define ABSL_LOCKS_EXCLUDED(...) +#endif + +// ABSL_LOCK_RETURNED() +// +// Documents a function that returns a mutex without acquiring it. For example, +// a public getter method that returns a pointer to a private mutex should +// be annotated with ABSL_LOCK_RETURNED. +#if ABSL_HAVE_ATTRIBUTE(lock_returned) +#define ABSL_LOCK_RETURNED(x) __attribute__((lock_returned(x))) +#else +#define ABSL_LOCK_RETURNED(x) +#endif + +// ABSL_LOCKABLE +// +// Documents if a class/type is a lockable type (such as the `Mutex` class). +#if ABSL_HAVE_ATTRIBUTE(lockable) +#define ABSL_LOCKABLE __attribute__((lockable)) +#else +#define ABSL_LOCKABLE +#endif + +// ABSL_SCOPED_LOCKABLE +// +// Documents if a class does RAII locking (such as the `MutexLock` class). +// The constructor should use `LOCK_FUNCTION()` to specify the mutex that is +// acquired, and the destructor should use `UNLOCK_FUNCTION()` with no +// arguments; the analysis will assume that the destructor unlocks whatever the +// constructor locked. +#if ABSL_HAVE_ATTRIBUTE(scoped_lockable) +#define ABSL_SCOPED_LOCKABLE __attribute__((scoped_lockable)) +#else +#define ABSL_SCOPED_LOCKABLE +#endif + +// ABSL_EXCLUSIVE_LOCK_FUNCTION() +// +// Documents functions that acquire a lock in the body of a function, and do +// not release it. +#if ABSL_HAVE_ATTRIBUTE(exclusive_lock_function) +#define ABSL_EXCLUSIVE_LOCK_FUNCTION(...) \ + __attribute__((exclusive_lock_function(__VA_ARGS__))) +#else +#define ABSL_EXCLUSIVE_LOCK_FUNCTION(...) +#endif + +// ABSL_SHARED_LOCK_FUNCTION() +// +// Documents functions that acquire a shared (reader) lock in the body of a +// function, and do not release it. +#if ABSL_HAVE_ATTRIBUTE(shared_lock_function) +#define ABSL_SHARED_LOCK_FUNCTION(...) \ + __attribute__((shared_lock_function(__VA_ARGS__))) +#else +#define ABSL_SHARED_LOCK_FUNCTION(...) +#endif + +// ABSL_UNLOCK_FUNCTION() +// +// Documents functions that expect a lock to be held on entry to the function, +// and release it in the body of the function. +#if ABSL_HAVE_ATTRIBUTE(unlock_function) +#define ABSL_UNLOCK_FUNCTION(...) __attribute__((unlock_function(__VA_ARGS__))) +#else +#define ABSL_UNLOCK_FUNCTION(...) +#endif + +// ABSL_EXCLUSIVE_TRYLOCK_FUNCTION() / ABSL_SHARED_TRYLOCK_FUNCTION() +// +// Documents functions that try to acquire a lock, and return success or failure +// (or a non-boolean value that can be interpreted as a boolean). +// The first argument should be `true` for functions that return `true` on +// success, or `false` for functions that return `false` on success. The second +// argument specifies the mutex that is locked on success. If unspecified, this +// mutex is assumed to be `this`. +#if ABSL_HAVE_ATTRIBUTE(exclusive_trylock_function) +#define ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(...) \ + __attribute__((exclusive_trylock_function(__VA_ARGS__))) +#else +#define ABSL_EXCLUSIVE_TRYLOCK_FUNCTION(...) +#endif + +#if ABSL_HAVE_ATTRIBUTE(shared_trylock_function) +#define ABSL_SHARED_TRYLOCK_FUNCTION(...) \ + __attribute__((shared_trylock_function(__VA_ARGS__))) +#else +#define ABSL_SHARED_TRYLOCK_FUNCTION(...) +#endif + +// ABSL_ASSERT_EXCLUSIVE_LOCK() / ABSL_ASSERT_SHARED_LOCK() +// +// Documents functions that dynamically check to see if a lock is held, and fail +// if it is not held. +#if ABSL_HAVE_ATTRIBUTE(assert_exclusive_lock) +#define ABSL_ASSERT_EXCLUSIVE_LOCK(...) \ + __attribute__((assert_exclusive_lock(__VA_ARGS__))) +#else +#define ABSL_ASSERT_EXCLUSIVE_LOCK(...) +#endif + +#if ABSL_HAVE_ATTRIBUTE(assert_shared_lock) +#define ABSL_ASSERT_SHARED_LOCK(...) \ + __attribute__((assert_shared_lock(__VA_ARGS__))) +#else +#define ABSL_ASSERT_SHARED_LOCK(...) +#endif + +// ABSL_NO_THREAD_SAFETY_ANALYSIS +// +// Turns off thread safety checking within the body of a particular function. +// This annotation is used to mark functions that are known to be correct, but +// the locking behavior is more complicated than the analyzer can handle. +#if ABSL_HAVE_ATTRIBUTE(no_thread_safety_analysis) +#define ABSL_NO_THREAD_SAFETY_ANALYSIS \ + __attribute__((no_thread_safety_analysis)) +#else +#define ABSL_NO_THREAD_SAFETY_ANALYSIS +#endif + +//------------------------------------------------------------------------------ +// Tool-Supplied Annotations +//------------------------------------------------------------------------------ + +// ABSL_TS_UNCHECKED should be placed around lock expressions that are not valid +// C++ syntax, but which are present for documentation purposes. These +// annotations will be ignored by the analysis. +#define ABSL_TS_UNCHECKED(x) "" + +// ABSL_TS_FIXME is used to mark lock expressions that are not valid C++ syntax. +// It is used by automated tools to mark and disable invalid expressions. +// The annotation should either be fixed, or changed to ABSL_TS_UNCHECKED. +#define ABSL_TS_FIXME(x) "" + +// Like ABSL_NO_THREAD_SAFETY_ANALYSIS, this turns off checking within the body +// of a particular function. However, this attribute is used to mark functions +// that are incorrect and need to be fixed. It is used by automated tools to +// avoid breaking the build when the analysis is updated. +// Code owners are expected to eventually fix the routine. +#define ABSL_NO_THREAD_SAFETY_ANALYSIS_FIXME ABSL_NO_THREAD_SAFETY_ANALYSIS + +// Similar to ABSL_NO_THREAD_SAFETY_ANALYSIS_FIXME, this macro marks a +// ABSL_GUARDED_BY annotation that needs to be fixed, because it is producing +// thread safety warning. It disables the ABSL_GUARDED_BY. +#define ABSL_GUARDED_BY_FIXME(x) + +// Disables warnings for a single read operation. This can be used to avoid +// warnings when it is known that the read is not actually involved in a race, +// but the compiler cannot confirm that. +#define ABSL_TS_UNCHECKED_READ(x) absl::base_internal::ts_unchecked_read(x) + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace base_internal { + +// Takes a reference to a guarded data member, and returns an unguarded +// reference. +// Do not use this function directly, use ABSL_TS_UNCHECKED_READ instead. +template +inline const T& ts_unchecked_read(const T& v) ABSL_NO_THREAD_SAFETY_ANALYSIS { + return v; +} + +template +inline T& ts_unchecked_read(T& v) ABSL_NO_THREAD_SAFETY_ANALYSIS { + return v; +} + +} // namespace base_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_BASE_THREAD_ANNOTATIONS_H_ diff --git a/src/absl/container/btree_map.h b/src/absl/container/btree_map.h new file mode 100644 index 0000000..286817f --- /dev/null +++ b/src/absl/container/btree_map.h @@ -0,0 +1,851 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: btree_map.h +// ----------------------------------------------------------------------------- +// +// This header file defines B-tree maps: sorted associative containers mapping +// keys to values. +// +// * `absl::btree_map<>` +// * `absl::btree_multimap<>` +// +// These B-tree types are similar to the corresponding types in the STL +// (`std::map` and `std::multimap`) and generally conform to the STL interfaces +// of those types. However, because they are implemented using B-trees, they +// are more efficient in most situations. +// +// Unlike `std::map` and `std::multimap`, which are commonly implemented using +// red-black tree nodes, B-tree maps use more generic B-tree nodes able to hold +// multiple values per node. Holding multiple values per node often makes +// B-tree maps perform better than their `std::map` counterparts, because +// multiple entries can be checked within the same cache hit. +// +// However, these types should not be considered drop-in replacements for +// `std::map` and `std::multimap` as there are some API differences, which are +// noted in this header file. The most consequential differences with respect to +// migrating to b-tree from the STL types are listed in the next paragraph. +// Other API differences are minor. +// +// Importantly, insertions and deletions may invalidate outstanding iterators, +// pointers, and references to elements. Such invalidations are typically only +// an issue if insertion and deletion operations are interleaved with the use of +// more than one iterator, pointer, or reference simultaneously. For this +// reason, `insert()` and `erase()` return a valid iterator at the current +// position. Another important difference is that key-types must be +// copy-constructible. + +#ifndef ABSL_CONTAINER_BTREE_MAP_H_ +#define ABSL_CONTAINER_BTREE_MAP_H_ + +#include "absl/container/internal/btree.h" // IWYU pragma: export +#include "absl/container/internal/btree_container.h" // IWYU pragma: export + +namespace absl { +ABSL_NAMESPACE_BEGIN + +namespace container_internal { + +template +struct map_params; + +} // namespace container_internal + +// absl::btree_map<> +// +// An `absl::btree_map` is an ordered associative container of +// unique keys and associated values designed to be a more efficient replacement +// for `std::map` (in most cases). +// +// Keys are sorted using an (optional) comparison function, which defaults to +// `std::less`. +// +// An `absl::btree_map` uses a default allocator of +// `std::allocator>` to allocate (and deallocate) +// nodes, and construct and destruct values within those nodes. You may +// instead specify a custom allocator `A` (which in turn requires specifying a +// custom comparator `C`) as in `absl::btree_map`. +// +template , + typename Alloc = std::allocator>> +class btree_map + : public container_internal::btree_map_container< + container_internal::btree>> { + using Base = typename btree_map::btree_map_container; + + public: + // Constructors and Assignment Operators + // + // A `btree_map` supports the same overload set as `std::map` + // for construction and assignment: + // + // * Default constructor + // + // absl::btree_map map1; + // + // * Initializer List constructor + // + // absl::btree_map map2 = + // {{1, "huey"}, {2, "dewey"}, {3, "louie"},}; + // + // * Copy constructor + // + // absl::btree_map map3(map2); + // + // * Copy assignment operator + // + // absl::btree_map map4; + // map4 = map3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::btree_map map5(std::move(map4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::btree_map map6; + // map6 = std::move(map5); + // + // * Range constructor + // + // std::vector> v = {{1, "a"}, {2, "b"}}; + // absl::btree_map map7(v.begin(), v.end()); + btree_map() {} + using Base::Base; + + // btree_map::begin() + // + // Returns an iterator to the beginning of the `btree_map`. + using Base::begin; + + // btree_map::cbegin() + // + // Returns a const iterator to the beginning of the `btree_map`. + using Base::cbegin; + + // btree_map::end() + // + // Returns an iterator to the end of the `btree_map`. + using Base::end; + + // btree_map::cend() + // + // Returns a const iterator to the end of the `btree_map`. + using Base::cend; + + // btree_map::empty() + // + // Returns whether or not the `btree_map` is empty. + using Base::empty; + + // btree_map::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `btree_map` under current memory constraints. This value can be thought + // of as the largest value of `std::distance(begin(), end())` for a + // `btree_map`. + using Base::max_size; + + // btree_map::size() + // + // Returns the number of elements currently within the `btree_map`. + using Base::size; + + // btree_map::clear() + // + // Removes all elements from the `btree_map`. Invalidates any references, + // pointers, or iterators referring to contained elements. + using Base::clear; + + // btree_map::erase() + // + // Erases elements within the `btree_map`. If an erase occurs, any references, + // pointers, or iterators are invalidated. + // Overloads are listed below. + // + // iterator erase(iterator position): + // iterator erase(const_iterator position): + // + // Erases the element at `position` of the `btree_map`, returning + // the iterator pointing to the element after the one that was erased + // (or end() if none exists). + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning + // the iterator pointing to the element after the interval that was erased + // (or end() if none exists). + // + // template size_type erase(const K& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // btree_map::insert() + // + // Inserts an element of the specified value into the `btree_map`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If an insertion + // occurs, any references, pointers, or iterators are invalidated. + // Overloads are listed below. + // + // std::pair insert(const value_type& value): + // + // Inserts a value into the `btree_map`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a bool denoting whether the insertion took place. + // + // std::pair insert(value_type&& value): + // + // Inserts a moveable value into the `btree_map`. Returns a pair + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a bool denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const value_type& value): + // iterator insert(const_iterator hint, value_type&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + using Base::insert; + + // btree_map::insert_or_assign() + // + // Inserts an element of the specified value into the `btree_map` provided + // that a value with the given key does not already exist, or replaces the + // corresponding mapped type with the forwarded `obj` argument if a key for + // that value already exists, returning an iterator pointing to the newly + // inserted element. Overloads are listed below. + // + // pair insert_or_assign(const key_type& k, M&& obj): + // pair insert_or_assign(key_type&& k, M&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `btree_map`. If the returned bool is true, insertion took place, and if + // it's false, assignment took place. + // + // iterator insert_or_assign(const_iterator hint, + // const key_type& k, M&& obj): + // iterator insert_or_assign(const_iterator hint, key_type&& k, M&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `btree_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + using Base::insert_or_assign; + + // btree_map::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_map`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. + using Base::emplace; + + // btree_map::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_map`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. + using Base::emplace_hint; + + // btree_map::try_emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_map`, provided that no element with the given key + // already exists. Unlike `emplace()`, if an element with the given key + // already exists, we guarantee that no element is constructed. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. + // + // Overloads are listed below. + // + // std::pair try_emplace(const key_type& k, Args&&... args): + // std::pair try_emplace(key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `btree_map`. + // + // iterator try_emplace(const_iterator hint, + // const key_type& k, Args&&... args): + // iterator try_emplace(const_iterator hint, key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `btree_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + using Base::try_emplace; + + // btree_map::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // template node_type extract(const K& k): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `btree_map` + // does not contain an element with a matching key, this function returns an + // empty node handle. + // + // NOTE: when compiled in an earlier version of C++ than C++17, + // `node_type::key()` returns a const reference to the key instead of a + // mutable reference. We cannot safely return a mutable reference without + // std::launder (which is not available before C++17). + // + // NOTE: In this context, `node_type` refers to the C++17 concept of a + // move-only type that owns and provides access to the elements in associative + // containers (https://en.cppreference.com/w/cpp/container/node_handle). + // It does NOT refer to the data layout of the underlying btree. + using Base::extract; + + // btree_map::merge() + // + // Extracts elements from a given `source` btree_map into this + // `btree_map`. If the destination `btree_map` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // btree_map::swap(btree_map& other) + // + // Exchanges the contents of this `btree_map` with those of the `other` + // btree_map, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `btree_map` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + using Base::swap; + + // btree_map::at() + // + // Returns a reference to the mapped value of the element with key equivalent + // to the passed key. + using Base::at; + + // btree_map::contains() + // + // template bool contains(const K& key) const: + // + // Determines whether an element comparing equal to the given `key` exists + // within the `btree_map`, returning `true` if so or `false` otherwise. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::contains; + + // btree_map::count() + // + // template size_type count(const K& key) const: + // + // Returns the number of elements comparing equal to the given `key` within + // the `btree_map`. Note that this function will return either `1` or `0` + // since duplicate elements are not allowed within a `btree_map`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::count; + + // btree_map::equal_range() + // + // Returns a half-open range [first, last), defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the `btree_map`. + using Base::equal_range; + + // btree_map::find() + // + // template iterator find(const K& key): + // template const_iterator find(const K& key) const: + // + // Finds an element with the passed `key` within the `btree_map`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::find; + + // btree_map::lower_bound() + // + // template iterator lower_bound(const K& key): + // template const_iterator lower_bound(const K& key) const: + // + // Finds the first element with a key that is not less than `key` within the + // `btree_map`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::lower_bound; + + // btree_map::upper_bound() + // + // template iterator upper_bound(const K& key): + // template const_iterator upper_bound(const K& key) const: + // + // Finds the first element with a key that is greater than `key` within the + // `btree_map`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::upper_bound; + + // btree_map::operator[]() + // + // Returns a reference to the value mapped to the passed key within the + // `btree_map`, performing an `insert()` if the key does not already + // exist. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. Otherwise iterators are not affected and references are not + // invalidated. Overloads are listed below. + // + // T& operator[](key_type&& key): + // T& operator[](const key_type& key): + // + // Inserts a value_type object constructed in-place if the element with the + // given key does not exist. + using Base::operator[]; + + // btree_map::get_allocator() + // + // Returns the allocator function associated with this `btree_map`. + using Base::get_allocator; + + // btree_map::key_comp(); + // + // Returns the key comparator associated with this `btree_map`. + using Base::key_comp; + + // btree_map::value_comp(); + // + // Returns the value comparator associated with this `btree_map`. + using Base::value_comp; +}; + +// absl::swap(absl::btree_map<>, absl::btree_map<>) +// +// Swaps the contents of two `absl::btree_map` containers. +template +void swap(btree_map &x, btree_map &y) { + return x.swap(y); +} + +// absl::erase_if(absl::btree_map<>, Pred) +// +// Erases all elements that satisfy the predicate pred from the container. +// Returns the number of erased elements. +template +typename btree_map::size_type erase_if( + btree_map &map, Pred pred) { + return container_internal::btree_access::erase_if(map, std::move(pred)); +} + +// absl::btree_multimap +// +// An `absl::btree_multimap` is an ordered associative container of +// keys and associated values designed to be a more efficient replacement for +// `std::multimap` (in most cases). Unlike `absl::btree_map`, a B-tree multimap +// allows multiple elements with equivalent keys. +// +// Keys are sorted using an (optional) comparison function, which defaults to +// `std::less`. +// +// An `absl::btree_multimap` uses a default allocator of +// `std::allocator>` to allocate (and deallocate) +// nodes, and construct and destruct values within those nodes. You may +// instead specify a custom allocator `A` (which in turn requires specifying a +// custom comparator `C`) as in `absl::btree_multimap`. +// +template , + typename Alloc = std::allocator>> +class btree_multimap + : public container_internal::btree_multimap_container< + container_internal::btree>> { + using Base = typename btree_multimap::btree_multimap_container; + + public: + // Constructors and Assignment Operators + // + // A `btree_multimap` supports the same overload set as `std::multimap` + // for construction and assignment: + // + // * Default constructor + // + // absl::btree_multimap map1; + // + // * Initializer List constructor + // + // absl::btree_multimap map2 = + // {{1, "huey"}, {2, "dewey"}, {3, "louie"},}; + // + // * Copy constructor + // + // absl::btree_multimap map3(map2); + // + // * Copy assignment operator + // + // absl::btree_multimap map4; + // map4 = map3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::btree_multimap map5(std::move(map4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::btree_multimap map6; + // map6 = std::move(map5); + // + // * Range constructor + // + // std::vector> v = {{1, "a"}, {2, "b"}}; + // absl::btree_multimap map7(v.begin(), v.end()); + btree_multimap() {} + using Base::Base; + + // btree_multimap::begin() + // + // Returns an iterator to the beginning of the `btree_multimap`. + using Base::begin; + + // btree_multimap::cbegin() + // + // Returns a const iterator to the beginning of the `btree_multimap`. + using Base::cbegin; + + // btree_multimap::end() + // + // Returns an iterator to the end of the `btree_multimap`. + using Base::end; + + // btree_multimap::cend() + // + // Returns a const iterator to the end of the `btree_multimap`. + using Base::cend; + + // btree_multimap::empty() + // + // Returns whether or not the `btree_multimap` is empty. + using Base::empty; + + // btree_multimap::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `btree_multimap` under current memory constraints. This value can be + // thought of as the largest value of `std::distance(begin(), end())` for a + // `btree_multimap`. + using Base::max_size; + + // btree_multimap::size() + // + // Returns the number of elements currently within the `btree_multimap`. + using Base::size; + + // btree_multimap::clear() + // + // Removes all elements from the `btree_multimap`. Invalidates any references, + // pointers, or iterators referring to contained elements. + using Base::clear; + + // btree_multimap::erase() + // + // Erases elements within the `btree_multimap`. If an erase occurs, any + // references, pointers, or iterators are invalidated. + // Overloads are listed below. + // + // iterator erase(iterator position): + // iterator erase(const_iterator position): + // + // Erases the element at `position` of the `btree_multimap`, returning + // the iterator pointing to the element after the one that was erased + // (or end() if none exists). + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning + // the iterator pointing to the element after the interval that was erased + // (or end() if none exists). + // + // template size_type erase(const K& key): + // + // Erases the elements matching the key, if any exist, returning the + // number of elements erased. + using Base::erase; + + // btree_multimap::insert() + // + // Inserts an element of the specified value into the `btree_multimap`, + // returning an iterator pointing to the newly inserted element. + // Any references, pointers, or iterators are invalidated. Overloads are + // listed below. + // + // iterator insert(const value_type& value): + // + // Inserts a value into the `btree_multimap`, returning an iterator to the + // inserted element. + // + // iterator insert(value_type&& value): + // + // Inserts a moveable value into the `btree_multimap`, returning an iterator + // to the inserted element. + // + // iterator insert(const_iterator hint, const value_type& value): + // iterator insert(const_iterator hint, value_type&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + using Base::insert; + + // btree_multimap::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_multimap`. Any references, pointers, or iterators are + // invalidated. + using Base::emplace; + + // btree_multimap::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_multimap`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search. + // + // Any references, pointers, or iterators are invalidated. + using Base::emplace_hint; + + // btree_multimap::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // template node_type extract(const K& k): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `btree_multimap` + // does not contain an element with a matching key, this function returns an + // empty node handle. + // + // NOTE: when compiled in an earlier version of C++ than C++17, + // `node_type::key()` returns a const reference to the key instead of a + // mutable reference. We cannot safely return a mutable reference without + // std::launder (which is not available before C++17). + // + // NOTE: In this context, `node_type` refers to the C++17 concept of a + // move-only type that owns and provides access to the elements in associative + // containers (https://en.cppreference.com/w/cpp/container/node_handle). + // It does NOT refer to the data layout of the underlying btree. + using Base::extract; + + // btree_multimap::merge() + // + // Extracts all elements from a given `source` btree_multimap into this + // `btree_multimap`. + using Base::merge; + + // btree_multimap::swap(btree_multimap& other) + // + // Exchanges the contents of this `btree_multimap` with those of the `other` + // btree_multimap, avoiding invocation of any move, copy, or swap operations + // on individual elements. + // + // All iterators and references on the `btree_multimap` remain valid, + // excepting for the past-the-end iterator, which is invalidated. + using Base::swap; + + // btree_multimap::contains() + // + // template bool contains(const K& key) const: + // + // Determines whether an element comparing equal to the given `key` exists + // within the `btree_multimap`, returning `true` if so or `false` otherwise. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::contains; + + // btree_multimap::count() + // + // template size_type count(const K& key) const: + // + // Returns the number of elements comparing equal to the given `key` within + // the `btree_multimap`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::count; + + // btree_multimap::equal_range() + // + // Returns a half-open range [first, last), defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `btree_multimap`. + using Base::equal_range; + + // btree_multimap::find() + // + // template iterator find(const K& key): + // template const_iterator find(const K& key) const: + // + // Finds an element with the passed `key` within the `btree_multimap`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::find; + + // btree_multimap::lower_bound() + // + // template iterator lower_bound(const K& key): + // template const_iterator lower_bound(const K& key) const: + // + // Finds the first element with a key that is not less than `key` within the + // `btree_multimap`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::lower_bound; + + // btree_multimap::upper_bound() + // + // template iterator upper_bound(const K& key): + // template const_iterator upper_bound(const K& key) const: + // + // Finds the first element with a key that is greater than `key` within the + // `btree_multimap`. + // + // Supports heterogeneous lookup, provided that the map has a compatible + // heterogeneous comparator. + using Base::upper_bound; + + // btree_multimap::get_allocator() + // + // Returns the allocator function associated with this `btree_multimap`. + using Base::get_allocator; + + // btree_multimap::key_comp(); + // + // Returns the key comparator associated with this `btree_multimap`. + using Base::key_comp; + + // btree_multimap::value_comp(); + // + // Returns the value comparator associated with this `btree_multimap`. + using Base::value_comp; +}; + +// absl::swap(absl::btree_multimap<>, absl::btree_multimap<>) +// +// Swaps the contents of two `absl::btree_multimap` containers. +template +void swap(btree_multimap &x, btree_multimap &y) { + return x.swap(y); +} + +// absl::erase_if(absl::btree_multimap<>, Pred) +// +// Erases all elements that satisfy the predicate pred from the container. +// Returns the number of erased elements. +template +typename btree_multimap::size_type erase_if( + btree_multimap &map, Pred pred) { + return container_internal::btree_access::erase_if(map, std::move(pred)); +} + +namespace container_internal { + +// A parameters structure for holding the type parameters for a btree_map. +// Compare and Alloc should be nothrow copy-constructible. +template +struct map_params : common_params> { + using super_type = typename map_params::common_params; + using mapped_type = Data; + // This type allows us to move keys when it is safe to do so. It is safe + // for maps in which value_type and mutable_value_type are layout compatible. + using slot_policy = typename super_type::slot_policy; + using slot_type = typename super_type::slot_type; + using value_type = typename super_type::value_type; + using init_type = typename super_type::init_type; + + template + static auto key(const V &value) -> decltype(value.first) { + return value.first; + } + static const Key &key(const slot_type *s) { return slot_policy::key(s); } + static const Key &key(slot_type *s) { return slot_policy::key(s); } + // For use in node handle. + static auto mutable_key(slot_type *s) + -> decltype(slot_policy::mutable_key(s)) { + return slot_policy::mutable_key(s); + } + static mapped_type &value(value_type *value) { return value->second; } +}; + +} // namespace container_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_BTREE_MAP_H_ diff --git a/src/absl/container/btree_set.h b/src/absl/container/btree_set.h new file mode 100644 index 0000000..695b09f --- /dev/null +++ b/src/absl/container/btree_set.h @@ -0,0 +1,793 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: btree_set.h +// ----------------------------------------------------------------------------- +// +// This header file defines B-tree sets: sorted associative containers of +// values. +// +// * `absl::btree_set<>` +// * `absl::btree_multiset<>` +// +// These B-tree types are similar to the corresponding types in the STL +// (`std::set` and `std::multiset`) and generally conform to the STL interfaces +// of those types. However, because they are implemented using B-trees, they +// are more efficient in most situations. +// +// Unlike `std::set` and `std::multiset`, which are commonly implemented using +// red-black tree nodes, B-tree sets use more generic B-tree nodes able to hold +// multiple values per node. Holding multiple values per node often makes +// B-tree sets perform better than their `std::set` counterparts, because +// multiple entries can be checked within the same cache hit. +// +// However, these types should not be considered drop-in replacements for +// `std::set` and `std::multiset` as there are some API differences, which are +// noted in this header file. The most consequential differences with respect to +// migrating to b-tree from the STL types are listed in the next paragraph. +// Other API differences are minor. +// +// Importantly, insertions and deletions may invalidate outstanding iterators, +// pointers, and references to elements. Such invalidations are typically only +// an issue if insertion and deletion operations are interleaved with the use of +// more than one iterator, pointer, or reference simultaneously. For this +// reason, `insert()` and `erase()` return a valid iterator at the current +// position. + +#ifndef ABSL_CONTAINER_BTREE_SET_H_ +#define ABSL_CONTAINER_BTREE_SET_H_ + +#include "absl/container/internal/btree.h" // IWYU pragma: export +#include "absl/container/internal/btree_container.h" // IWYU pragma: export + +namespace absl { +ABSL_NAMESPACE_BEGIN + +namespace container_internal { + +template +struct set_slot_policy; + +template +struct set_params; + +} // namespace container_internal + +// absl::btree_set<> +// +// An `absl::btree_set` is an ordered associative container of unique key +// values designed to be a more efficient replacement for `std::set` (in most +// cases). +// +// Keys are sorted using an (optional) comparison function, which defaults to +// `std::less`. +// +// An `absl::btree_set` uses a default allocator of `std::allocator` to +// allocate (and deallocate) nodes, and construct and destruct values within +// those nodes. You may instead specify a custom allocator `A` (which in turn +// requires specifying a custom comparator `C`) as in +// `absl::btree_set`. +// +template , + typename Alloc = std::allocator> +class btree_set + : public container_internal::btree_set_container< + container_internal::btree>> { + using Base = typename btree_set::btree_set_container; + + public: + // Constructors and Assignment Operators + // + // A `btree_set` supports the same overload set as `std::set` + // for construction and assignment: + // + // * Default constructor + // + // absl::btree_set set1; + // + // * Initializer List constructor + // + // absl::btree_set set2 = + // {{"huey"}, {"dewey"}, {"louie"},}; + // + // * Copy constructor + // + // absl::btree_set set3(set2); + // + // * Copy assignment operator + // + // absl::btree_set set4; + // set4 = set3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::btree_set set5(std::move(set4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::btree_set set6; + // set6 = std::move(set5); + // + // * Range constructor + // + // std::vector v = {"a", "b"}; + // absl::btree_set set7(v.begin(), v.end()); + btree_set() {} + using Base::Base; + + // btree_set::begin() + // + // Returns an iterator to the beginning of the `btree_set`. + using Base::begin; + + // btree_set::cbegin() + // + // Returns a const iterator to the beginning of the `btree_set`. + using Base::cbegin; + + // btree_set::end() + // + // Returns an iterator to the end of the `btree_set`. + using Base::end; + + // btree_set::cend() + // + // Returns a const iterator to the end of the `btree_set`. + using Base::cend; + + // btree_set::empty() + // + // Returns whether or not the `btree_set` is empty. + using Base::empty; + + // btree_set::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `btree_set` under current memory constraints. This value can be thought + // of as the largest value of `std::distance(begin(), end())` for a + // `btree_set`. + using Base::max_size; + + // btree_set::size() + // + // Returns the number of elements currently within the `btree_set`. + using Base::size; + + // btree_set::clear() + // + // Removes all elements from the `btree_set`. Invalidates any references, + // pointers, or iterators referring to contained elements. + using Base::clear; + + // btree_set::erase() + // + // Erases elements within the `btree_set`. Overloads are listed below. + // + // iterator erase(iterator position): + // iterator erase(const_iterator position): + // + // Erases the element at `position` of the `btree_set`, returning + // the iterator pointing to the element after the one that was erased + // (or end() if none exists). + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning + // the iterator pointing to the element after the interval that was erased + // (or end() if none exists). + // + // template size_type erase(const K& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // btree_set::insert() + // + // Inserts an element of the specified value into the `btree_set`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If an insertion + // occurs, any references, pointers, or iterators are invalidated. + // Overloads are listed below. + // + // std::pair insert(const value_type& value): + // + // Inserts a value into the `btree_set`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a bool denoting whether the insertion took place. + // + // std::pair insert(value_type&& value): + // + // Inserts a moveable value into the `btree_set`. Returns a pair + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a bool denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const value_type& value): + // iterator insert(const_iterator hint, value_type&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + using Base::insert; + + // btree_set::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_set`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. + using Base::emplace; + + // btree_set::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_set`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If an insertion occurs, any references, pointers, or iterators are + // invalidated. + using Base::emplace_hint; + + // btree_set::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // template node_type extract(const K& k): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `btree_set` + // does not contain an element with a matching key, this function returns an + // empty node handle. + // + // NOTE: In this context, `node_type` refers to the C++17 concept of a + // move-only type that owns and provides access to the elements in associative + // containers (https://en.cppreference.com/w/cpp/container/node_handle). + // It does NOT refer to the data layout of the underlying btree. + using Base::extract; + + // btree_set::merge() + // + // Extracts elements from a given `source` btree_set into this + // `btree_set`. If the destination `btree_set` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // btree_set::swap(btree_set& other) + // + // Exchanges the contents of this `btree_set` with those of the `other` + // btree_set, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `btree_set` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + using Base::swap; + + // btree_set::contains() + // + // template bool contains(const K& key) const: + // + // Determines whether an element comparing equal to the given `key` exists + // within the `btree_set`, returning `true` if so or `false` otherwise. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::contains; + + // btree_set::count() + // + // template size_type count(const K& key) const: + // + // Returns the number of elements comparing equal to the given `key` within + // the `btree_set`. Note that this function will return either `1` or `0` + // since duplicate elements are not allowed within a `btree_set`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::count; + + // btree_set::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `btree_set`. + using Base::equal_range; + + // btree_set::find() + // + // template iterator find(const K& key): + // template const_iterator find(const K& key) const: + // + // Finds an element with the passed `key` within the `btree_set`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::find; + + // btree_set::lower_bound() + // + // template iterator lower_bound(const K& key): + // template const_iterator lower_bound(const K& key) const: + // + // Finds the first element that is not less than `key` within the `btree_set`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::lower_bound; + + // btree_set::upper_bound() + // + // template iterator upper_bound(const K& key): + // template const_iterator upper_bound(const K& key) const: + // + // Finds the first element that is greater than `key` within the `btree_set`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::upper_bound; + + // btree_set::get_allocator() + // + // Returns the allocator function associated with this `btree_set`. + using Base::get_allocator; + + // btree_set::key_comp(); + // + // Returns the key comparator associated with this `btree_set`. + using Base::key_comp; + + // btree_set::value_comp(); + // + // Returns the value comparator associated with this `btree_set`. The keys to + // sort the elements are the values themselves, therefore `value_comp` and its + // sibling member function `key_comp` are equivalent. + using Base::value_comp; +}; + +// absl::swap(absl::btree_set<>, absl::btree_set<>) +// +// Swaps the contents of two `absl::btree_set` containers. +template +void swap(btree_set &x, btree_set &y) { + return x.swap(y); +} + +// absl::erase_if(absl::btree_set<>, Pred) +// +// Erases all elements that satisfy the predicate pred from the container. +// Returns the number of erased elements. +template +typename btree_set::size_type erase_if(btree_set &set, + Pred pred) { + return container_internal::btree_access::erase_if(set, std::move(pred)); +} + +// absl::btree_multiset<> +// +// An `absl::btree_multiset` is an ordered associative container of +// keys and associated values designed to be a more efficient replacement +// for `std::multiset` (in most cases). Unlike `absl::btree_set`, a B-tree +// multiset allows equivalent elements. +// +// Keys are sorted using an (optional) comparison function, which defaults to +// `std::less`. +// +// An `absl::btree_multiset` uses a default allocator of `std::allocator` +// to allocate (and deallocate) nodes, and construct and destruct values within +// those nodes. You may instead specify a custom allocator `A` (which in turn +// requires specifying a custom comparator `C`) as in +// `absl::btree_multiset`. +// +template , + typename Alloc = std::allocator> +class btree_multiset + : public container_internal::btree_multiset_container< + container_internal::btree>> { + using Base = typename btree_multiset::btree_multiset_container; + + public: + // Constructors and Assignment Operators + // + // A `btree_multiset` supports the same overload set as `std::set` + // for construction and assignment: + // + // * Default constructor + // + // absl::btree_multiset set1; + // + // * Initializer List constructor + // + // absl::btree_multiset set2 = + // {{"huey"}, {"dewey"}, {"louie"},}; + // + // * Copy constructor + // + // absl::btree_multiset set3(set2); + // + // * Copy assignment operator + // + // absl::btree_multiset set4; + // set4 = set3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::btree_multiset set5(std::move(set4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::btree_multiset set6; + // set6 = std::move(set5); + // + // * Range constructor + // + // std::vector v = {"a", "b"}; + // absl::btree_multiset set7(v.begin(), v.end()); + btree_multiset() {} + using Base::Base; + + // btree_multiset::begin() + // + // Returns an iterator to the beginning of the `btree_multiset`. + using Base::begin; + + // btree_multiset::cbegin() + // + // Returns a const iterator to the beginning of the `btree_multiset`. + using Base::cbegin; + + // btree_multiset::end() + // + // Returns an iterator to the end of the `btree_multiset`. + using Base::end; + + // btree_multiset::cend() + // + // Returns a const iterator to the end of the `btree_multiset`. + using Base::cend; + + // btree_multiset::empty() + // + // Returns whether or not the `btree_multiset` is empty. + using Base::empty; + + // btree_multiset::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `btree_multiset` under current memory constraints. This value can be + // thought of as the largest value of `std::distance(begin(), end())` for a + // `btree_multiset`. + using Base::max_size; + + // btree_multiset::size() + // + // Returns the number of elements currently within the `btree_multiset`. + using Base::size; + + // btree_multiset::clear() + // + // Removes all elements from the `btree_multiset`. Invalidates any references, + // pointers, or iterators referring to contained elements. + using Base::clear; + + // btree_multiset::erase() + // + // Erases elements within the `btree_multiset`. Overloads are listed below. + // + // iterator erase(iterator position): + // iterator erase(const_iterator position): + // + // Erases the element at `position` of the `btree_multiset`, returning + // the iterator pointing to the element after the one that was erased + // (or end() if none exists). + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning + // the iterator pointing to the element after the interval that was erased + // (or end() if none exists). + // + // template size_type erase(const K& key): + // + // Erases the elements matching the key, if any exist, returning the + // number of elements erased. + using Base::erase; + + // btree_multiset::insert() + // + // Inserts an element of the specified value into the `btree_multiset`, + // returning an iterator pointing to the newly inserted element. + // Any references, pointers, or iterators are invalidated. Overloads are + // listed below. + // + // iterator insert(const value_type& value): + // + // Inserts a value into the `btree_multiset`, returning an iterator to the + // inserted element. + // + // iterator insert(value_type&& value): + // + // Inserts a moveable value into the `btree_multiset`, returning an iterator + // to the inserted element. + // + // iterator insert(const_iterator hint, const value_type& value): + // iterator insert(const_iterator hint, value_type&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + using Base::insert; + + // btree_multiset::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_multiset`. Any references, pointers, or iterators are + // invalidated. + using Base::emplace; + + // btree_multiset::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `btree_multiset`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search. + // + // Any references, pointers, or iterators are invalidated. + using Base::emplace_hint; + + // btree_multiset::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // template node_type extract(const K& k): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `btree_multiset` + // does not contain an element with a matching key, this function returns an + // empty node handle. + // + // NOTE: In this context, `node_type` refers to the C++17 concept of a + // move-only type that owns and provides access to the elements in associative + // containers (https://en.cppreference.com/w/cpp/container/node_handle). + // It does NOT refer to the data layout of the underlying btree. + using Base::extract; + + // btree_multiset::merge() + // + // Extracts all elements from a given `source` btree_multiset into this + // `btree_multiset`. + using Base::merge; + + // btree_multiset::swap(btree_multiset& other) + // + // Exchanges the contents of this `btree_multiset` with those of the `other` + // btree_multiset, avoiding invocation of any move, copy, or swap operations + // on individual elements. + // + // All iterators and references on the `btree_multiset` remain valid, + // excepting for the past-the-end iterator, which is invalidated. + using Base::swap; + + // btree_multiset::contains() + // + // template bool contains(const K& key) const: + // + // Determines whether an element comparing equal to the given `key` exists + // within the `btree_multiset`, returning `true` if so or `false` otherwise. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::contains; + + // btree_multiset::count() + // + // template size_type count(const K& key) const: + // + // Returns the number of elements comparing equal to the given `key` within + // the `btree_multiset`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::count; + + // btree_multiset::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `btree_multiset`. + using Base::equal_range; + + // btree_multiset::find() + // + // template iterator find(const K& key): + // template const_iterator find(const K& key) const: + // + // Finds an element with the passed `key` within the `btree_multiset`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::find; + + // btree_multiset::lower_bound() + // + // template iterator lower_bound(const K& key): + // template const_iterator lower_bound(const K& key) const: + // + // Finds the first element that is not less than `key` within the + // `btree_multiset`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::lower_bound; + + // btree_multiset::upper_bound() + // + // template iterator upper_bound(const K& key): + // template const_iterator upper_bound(const K& key) const: + // + // Finds the first element that is greater than `key` within the + // `btree_multiset`. + // + // Supports heterogeneous lookup, provided that the set has a compatible + // heterogeneous comparator. + using Base::upper_bound; + + // btree_multiset::get_allocator() + // + // Returns the allocator function associated with this `btree_multiset`. + using Base::get_allocator; + + // btree_multiset::key_comp(); + // + // Returns the key comparator associated with this `btree_multiset`. + using Base::key_comp; + + // btree_multiset::value_comp(); + // + // Returns the value comparator associated with this `btree_multiset`. The + // keys to sort the elements are the values themselves, therefore `value_comp` + // and its sibling member function `key_comp` are equivalent. + using Base::value_comp; +}; + +// absl::swap(absl::btree_multiset<>, absl::btree_multiset<>) +// +// Swaps the contents of two `absl::btree_multiset` containers. +template +void swap(btree_multiset &x, btree_multiset &y) { + return x.swap(y); +} + +// absl::erase_if(absl::btree_multiset<>, Pred) +// +// Erases all elements that satisfy the predicate pred from the container. +// Returns the number of erased elements. +template +typename btree_multiset::size_type erase_if( + btree_multiset & set, Pred pred) { + return container_internal::btree_access::erase_if(set, std::move(pred)); +} + +namespace container_internal { + +// This type implements the necessary functions from the +// absl::container_internal::slot_type interface for btree_(multi)set. +template +struct set_slot_policy { + using slot_type = Key; + using value_type = Key; + using mutable_value_type = Key; + + static value_type &element(slot_type *slot) { return *slot; } + static const value_type &element(const slot_type *slot) { return *slot; } + + template + static void construct(Alloc *alloc, slot_type *slot, Args &&...args) { + absl::allocator_traits::construct(*alloc, slot, + std::forward(args)...); + } + + template + static void construct(Alloc *alloc, slot_type *slot, slot_type *other) { + absl::allocator_traits::construct(*alloc, slot, std::move(*other)); + } + + template + static void construct(Alloc *alloc, slot_type *slot, const slot_type *other) { + absl::allocator_traits::construct(*alloc, slot, *other); + } + + template + static void destroy(Alloc *alloc, slot_type *slot) { + absl::allocator_traits::destroy(*alloc, slot); + } + + template + static void transfer(Alloc *alloc, slot_type *new_slot, slot_type *old_slot) { + construct(alloc, new_slot, old_slot); + destroy(alloc, old_slot); + } +}; + +// A parameters structure for holding the type parameters for a btree_set. +// Compare and Alloc should be nothrow copy-constructible. +template +struct set_params : common_params> { + using value_type = Key; + using slot_type = typename set_params::common_params::slot_type; + + template + static const V &key(const V &value) { + return value; + } + static const Key &key(const slot_type *slot) { return *slot; } + static const Key &key(slot_type *slot) { return *slot; } +}; + +} // namespace container_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_BTREE_SET_H_ diff --git a/src/absl/container/fixed_array.h b/src/absl/container/fixed_array.h new file mode 100644 index 0000000..2aefae3 --- /dev/null +++ b/src/absl/container/fixed_array.h @@ -0,0 +1,529 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: fixed_array.h +// ----------------------------------------------------------------------------- +// +// A `FixedArray` represents a non-resizable array of `T` where the length of +// the array can be determined at run-time. It is a good replacement for +// non-standard and deprecated uses of `alloca()` and variable length arrays +// within the GCC extension. (See +// https://gcc.gnu.org/onlinedocs/gcc/Variable-Length.html). +// +// `FixedArray` allocates small arrays inline, keeping performance fast by +// avoiding heap operations. It also helps reduce the chances of +// accidentally overflowing your stack if large input is passed to +// your function. + +#ifndef ABSL_CONTAINER_FIXED_ARRAY_H_ +#define ABSL_CONTAINER_FIXED_ARRAY_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/algorithm/algorithm.h" +#include "absl/base/config.h" +#include "absl/base/dynamic_annotations.h" +#include "absl/base/internal/throw_delegate.h" +#include "absl/base/macros.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" +#include "absl/container/internal/compressed_tuple.h" +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +constexpr static auto kFixedArrayUseDefault = static_cast(-1); + +// ----------------------------------------------------------------------------- +// FixedArray +// ----------------------------------------------------------------------------- +// +// A `FixedArray` provides a run-time fixed-size array, allocating a small array +// inline for efficiency. +// +// Most users should not specify an `inline_elements` argument and let +// `FixedArray` automatically determine the number of elements +// to store inline based on `sizeof(T)`. If `inline_elements` is specified, the +// `FixedArray` implementation will use inline storage for arrays with a +// length <= `inline_elements`. +// +// Note that a `FixedArray` constructed with a `size_type` argument will +// default-initialize its values by leaving trivially constructible types +// uninitialized (e.g. int, int[4], double), and others default-constructed. +// This matches the behavior of c-style arrays and `std::array`, but not +// `std::vector`. +template > +class FixedArray { + static_assert(!std::is_array::value || std::extent::value > 0, + "Arrays with unknown bounds cannot be used with FixedArray."); + + static constexpr size_t kInlineBytesDefault = 256; + + using AllocatorTraits = std::allocator_traits; + // std::iterator_traits isn't guaranteed to be SFINAE-friendly until C++17, + // but this seems to be mostly pedantic. + template + using EnableIfForwardIterator = absl::enable_if_t::iterator_category, + std::forward_iterator_tag>::value>; + static constexpr bool NoexceptCopyable() { + return std::is_nothrow_copy_constructible::value && + absl::allocator_is_nothrow::value; + } + static constexpr bool NoexceptMovable() { + return std::is_nothrow_move_constructible::value && + absl::allocator_is_nothrow::value; + } + static constexpr bool DefaultConstructorIsNonTrivial() { + return !absl::is_trivially_default_constructible::value; + } + + public: + using allocator_type = typename AllocatorTraits::allocator_type; + using value_type = typename AllocatorTraits::value_type; + using pointer = typename AllocatorTraits::pointer; + using const_pointer = typename AllocatorTraits::const_pointer; + using reference = value_type&; + using const_reference = const value_type&; + using size_type = typename AllocatorTraits::size_type; + using difference_type = typename AllocatorTraits::difference_type; + using iterator = pointer; + using const_iterator = const_pointer; + using reverse_iterator = std::reverse_iterator; + using const_reverse_iterator = std::reverse_iterator; + + static constexpr size_type inline_elements = + (N == kFixedArrayUseDefault ? kInlineBytesDefault / sizeof(value_type) + : static_cast(N)); + + FixedArray( + const FixedArray& other, + const allocator_type& a = allocator_type()) noexcept(NoexceptCopyable()) + : FixedArray(other.begin(), other.end(), a) {} + + FixedArray( + FixedArray&& other, + const allocator_type& a = allocator_type()) noexcept(NoexceptMovable()) + : FixedArray(std::make_move_iterator(other.begin()), + std::make_move_iterator(other.end()), a) {} + + // Creates an array object that can store `n` elements. + // Note that trivially constructible elements will be uninitialized. + explicit FixedArray(size_type n, const allocator_type& a = allocator_type()) + : storage_(n, a) { + if (DefaultConstructorIsNonTrivial()) { + memory_internal::ConstructRange(storage_.alloc(), storage_.begin(), + storage_.end()); + } + } + + // Creates an array initialized with `n` copies of `val`. + FixedArray(size_type n, const value_type& val, + const allocator_type& a = allocator_type()) + : storage_(n, a) { + memory_internal::ConstructRange(storage_.alloc(), storage_.begin(), + storage_.end(), val); + } + + // Creates an array initialized with the size and contents of `init_list`. + FixedArray(std::initializer_list init_list, + const allocator_type& a = allocator_type()) + : FixedArray(init_list.begin(), init_list.end(), a) {} + + // Creates an array initialized with the elements from the input + // range. The array's size will always be `std::distance(first, last)`. + // REQUIRES: Iterator must be a forward_iterator or better. + template * = nullptr> + FixedArray(Iterator first, Iterator last, + const allocator_type& a = allocator_type()) + : storage_(std::distance(first, last), a) { + memory_internal::CopyRange(storage_.alloc(), storage_.begin(), first, last); + } + + ~FixedArray() noexcept { + for (auto* cur = storage_.begin(); cur != storage_.end(); ++cur) { + AllocatorTraits::destroy(storage_.alloc(), cur); + } + } + + // Assignments are deleted because they break the invariant that the size of a + // `FixedArray` never changes. + void operator=(FixedArray&&) = delete; + void operator=(const FixedArray&) = delete; + + // FixedArray::size() + // + // Returns the length of the fixed array. + size_type size() const { return storage_.size(); } + + // FixedArray::max_size() + // + // Returns the largest possible value of `std::distance(begin(), end())` for a + // `FixedArray`. This is equivalent to the most possible addressable bytes + // over the number of bytes taken by T. + constexpr size_type max_size() const { + return (std::numeric_limits::max)() / sizeof(value_type); + } + + // FixedArray::empty() + // + // Returns whether or not the fixed array is empty. + bool empty() const { return size() == 0; } + + // FixedArray::memsize() + // + // Returns the memory size of the fixed array in bytes. + size_t memsize() const { return size() * sizeof(value_type); } + + // FixedArray::data() + // + // Returns a const T* pointer to elements of the `FixedArray`. This pointer + // can be used to access (but not modify) the contained elements. + const_pointer data() const { return AsValueType(storage_.begin()); } + + // Overload of FixedArray::data() to return a T* pointer to elements of the + // fixed array. This pointer can be used to access and modify the contained + // elements. + pointer data() { return AsValueType(storage_.begin()); } + + // FixedArray::operator[] + // + // Returns a reference the ith element of the fixed array. + // REQUIRES: 0 <= i < size() + reference operator[](size_type i) { + ABSL_HARDENING_ASSERT(i < size()); + return data()[i]; + } + + // Overload of FixedArray::operator()[] to return a const reference to the + // ith element of the fixed array. + // REQUIRES: 0 <= i < size() + const_reference operator[](size_type i) const { + ABSL_HARDENING_ASSERT(i < size()); + return data()[i]; + } + + // FixedArray::at + // + // Bounds-checked access. Returns a reference to the ith element of the fixed + // array, or throws std::out_of_range + reference at(size_type i) { + if (ABSL_PREDICT_FALSE(i >= size())) { + base_internal::ThrowStdOutOfRange("FixedArray::at failed bounds check"); + } + return data()[i]; + } + + // Overload of FixedArray::at() to return a const reference to the ith element + // of the fixed array. + const_reference at(size_type i) const { + if (ABSL_PREDICT_FALSE(i >= size())) { + base_internal::ThrowStdOutOfRange("FixedArray::at failed bounds check"); + } + return data()[i]; + } + + // FixedArray::front() + // + // Returns a reference to the first element of the fixed array. + reference front() { + ABSL_HARDENING_ASSERT(!empty()); + return data()[0]; + } + + // Overload of FixedArray::front() to return a reference to the first element + // of a fixed array of const values. + const_reference front() const { + ABSL_HARDENING_ASSERT(!empty()); + return data()[0]; + } + + // FixedArray::back() + // + // Returns a reference to the last element of the fixed array. + reference back() { + ABSL_HARDENING_ASSERT(!empty()); + return data()[size() - 1]; + } + + // Overload of FixedArray::back() to return a reference to the last element + // of a fixed array of const values. + const_reference back() const { + ABSL_HARDENING_ASSERT(!empty()); + return data()[size() - 1]; + } + + // FixedArray::begin() + // + // Returns an iterator to the beginning of the fixed array. + iterator begin() { return data(); } + + // Overload of FixedArray::begin() to return a const iterator to the + // beginning of the fixed array. + const_iterator begin() const { return data(); } + + // FixedArray::cbegin() + // + // Returns a const iterator to the beginning of the fixed array. + const_iterator cbegin() const { return begin(); } + + // FixedArray::end() + // + // Returns an iterator to the end of the fixed array. + iterator end() { return data() + size(); } + + // Overload of FixedArray::end() to return a const iterator to the end of the + // fixed array. + const_iterator end() const { return data() + size(); } + + // FixedArray::cend() + // + // Returns a const iterator to the end of the fixed array. + const_iterator cend() const { return end(); } + + // FixedArray::rbegin() + // + // Returns a reverse iterator from the end of the fixed array. + reverse_iterator rbegin() { return reverse_iterator(end()); } + + // Overload of FixedArray::rbegin() to return a const reverse iterator from + // the end of the fixed array. + const_reverse_iterator rbegin() const { + return const_reverse_iterator(end()); + } + + // FixedArray::crbegin() + // + // Returns a const reverse iterator from the end of the fixed array. + const_reverse_iterator crbegin() const { return rbegin(); } + + // FixedArray::rend() + // + // Returns a reverse iterator from the beginning of the fixed array. + reverse_iterator rend() { return reverse_iterator(begin()); } + + // Overload of FixedArray::rend() for returning a const reverse iterator + // from the beginning of the fixed array. + const_reverse_iterator rend() const { + return const_reverse_iterator(begin()); + } + + // FixedArray::crend() + // + // Returns a reverse iterator from the beginning of the fixed array. + const_reverse_iterator crend() const { return rend(); } + + // FixedArray::fill() + // + // Assigns the given `value` to all elements in the fixed array. + void fill(const value_type& val) { std::fill(begin(), end(), val); } + + // Relational operators. Equality operators are elementwise using + // `operator==`, while order operators order FixedArrays lexicographically. + friend bool operator==(const FixedArray& lhs, const FixedArray& rhs) { + return absl::equal(lhs.begin(), lhs.end(), rhs.begin(), rhs.end()); + } + + friend bool operator!=(const FixedArray& lhs, const FixedArray& rhs) { + return !(lhs == rhs); + } + + friend bool operator<(const FixedArray& lhs, const FixedArray& rhs) { + return std::lexicographical_compare(lhs.begin(), lhs.end(), rhs.begin(), + rhs.end()); + } + + friend bool operator>(const FixedArray& lhs, const FixedArray& rhs) { + return rhs < lhs; + } + + friend bool operator<=(const FixedArray& lhs, const FixedArray& rhs) { + return !(rhs < lhs); + } + + friend bool operator>=(const FixedArray& lhs, const FixedArray& rhs) { + return !(lhs < rhs); + } + + template + friend H AbslHashValue(H h, const FixedArray& v) { + return H::combine(H::combine_contiguous(std::move(h), v.data(), v.size()), + v.size()); + } + + private: + // StorageElement + // + // For FixedArrays with a C-style-array value_type, StorageElement is a POD + // wrapper struct called StorageElementWrapper that holds the value_type + // instance inside. This is needed for construction and destruction of the + // entire array regardless of how many dimensions it has. For all other cases, + // StorageElement is just an alias of value_type. + // + // Maintainer's Note: The simpler solution would be to simply wrap value_type + // in a struct whether it's an array or not. That causes some paranoid + // diagnostics to misfire, believing that 'data()' returns a pointer to a + // single element, rather than the packed array that it really is. + // e.g.: + // + // FixedArray buf(1); + // sprintf(buf.data(), "foo"); + // + // error: call to int __builtin___sprintf_chk(etc...) + // will always overflow destination buffer [-Werror] + // + template , + size_t InnerN = std::extent::value> + struct StorageElementWrapper { + InnerT array[InnerN]; + }; + + using StorageElement = + absl::conditional_t::value, + StorageElementWrapper, value_type>; + + static pointer AsValueType(pointer ptr) { return ptr; } + static pointer AsValueType(StorageElementWrapper* ptr) { + return std::addressof(ptr->array); + } + + static_assert(sizeof(StorageElement) == sizeof(value_type), ""); + static_assert(alignof(StorageElement) == alignof(value_type), ""); + + class NonEmptyInlinedStorage { + public: + StorageElement* data() { return reinterpret_cast(buff_); } + void AnnotateConstruct(size_type n); + void AnnotateDestruct(size_type n); + +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + void* RedzoneBegin() { return &redzone_begin_; } + void* RedzoneEnd() { return &redzone_end_ + 1; } +#endif // ABSL_HAVE_ADDRESS_SANITIZER + + private: + ABSL_ADDRESS_SANITIZER_REDZONE(redzone_begin_); + alignas(StorageElement) char buff_[sizeof(StorageElement[inline_elements])]; + ABSL_ADDRESS_SANITIZER_REDZONE(redzone_end_); + }; + + class EmptyInlinedStorage { + public: + StorageElement* data() { return nullptr; } + void AnnotateConstruct(size_type) {} + void AnnotateDestruct(size_type) {} + }; + + using InlinedStorage = + absl::conditional_t; + + // Storage + // + // An instance of Storage manages the inline and out-of-line memory for + // instances of FixedArray. This guarantees that even when construction of + // individual elements fails in the FixedArray constructor body, the + // destructor for Storage will still be called and out-of-line memory will be + // properly deallocated. + // + class Storage : public InlinedStorage { + public: + Storage(size_type n, const allocator_type& a) + : size_alloc_(n, a), data_(InitializeData()) {} + + ~Storage() noexcept { + if (UsingInlinedStorage(size())) { + InlinedStorage::AnnotateDestruct(size()); + } else { + AllocatorTraits::deallocate(alloc(), AsValueType(begin()), size()); + } + } + + size_type size() const { return size_alloc_.template get<0>(); } + StorageElement* begin() const { return data_; } + StorageElement* end() const { return begin() + size(); } + allocator_type& alloc() { return size_alloc_.template get<1>(); } + + private: + static bool UsingInlinedStorage(size_type n) { + return n <= inline_elements; + } + + StorageElement* InitializeData() { + if (UsingInlinedStorage(size())) { + InlinedStorage::AnnotateConstruct(size()); + return InlinedStorage::data(); + } else { + return reinterpret_cast( + AllocatorTraits::allocate(alloc(), size())); + } + } + + // `CompressedTuple` takes advantage of EBCO for stateless `allocator_type`s + container_internal::CompressedTuple size_alloc_; + StorageElement* data_; + }; + + Storage storage_; +}; + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +template +constexpr size_t FixedArray::kInlineBytesDefault; + +template +constexpr typename FixedArray::size_type + FixedArray::inline_elements; +#endif + +template +void FixedArray::NonEmptyInlinedStorage::AnnotateConstruct( + typename FixedArray::size_type n) { +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + if (!n) return; + ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(data(), RedzoneEnd(), RedzoneEnd(), + data() + n); + ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(RedzoneBegin(), data(), data(), + RedzoneBegin()); +#endif // ABSL_HAVE_ADDRESS_SANITIZER + static_cast(n); // Mark used when not in asan mode +} + +template +void FixedArray::NonEmptyInlinedStorage::AnnotateDestruct( + typename FixedArray::size_type n) { +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + if (!n) return; + ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(data(), RedzoneEnd(), data() + n, + RedzoneEnd()); + ABSL_ANNOTATE_CONTIGUOUS_CONTAINER(RedzoneBegin(), data(), RedzoneBegin(), + data()); +#endif // ABSL_HAVE_ADDRESS_SANITIZER + static_cast(n); // Mark used when not in asan mode +} +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_FIXED_ARRAY_H_ diff --git a/src/absl/container/flat_hash_map.h b/src/absl/container/flat_hash_map.h new file mode 100644 index 0000000..e6bdbd9 --- /dev/null +++ b/src/absl/container/flat_hash_map.h @@ -0,0 +1,613 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: flat_hash_map.h +// ----------------------------------------------------------------------------- +// +// An `absl::flat_hash_map` is an unordered associative container of +// unique keys and associated values designed to be a more efficient replacement +// for `std::unordered_map`. Like `unordered_map`, search, insertion, and +// deletion of map elements can be done as an `O(1)` operation. However, +// `flat_hash_map` (and other unordered associative containers known as the +// collection of Abseil "Swiss tables") contain other optimizations that result +// in both memory and computation advantages. +// +// In most cases, your default choice for a hash map should be a map of type +// `flat_hash_map`. + +#ifndef ABSL_CONTAINER_FLAT_HASH_MAP_H_ +#define ABSL_CONTAINER_FLAT_HASH_MAP_H_ + +#include +#include +#include +#include + +#include "absl/algorithm/container.h" +#include "absl/base/macros.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/hash_function_defaults.h" // IWYU pragma: export +#include "absl/container/internal/raw_hash_map.h" // IWYU pragma: export +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { +template +struct FlatHashMapPolicy; +} // namespace container_internal + +// ----------------------------------------------------------------------------- +// absl::flat_hash_map +// ----------------------------------------------------------------------------- +// +// An `absl::flat_hash_map` is an unordered associative container which +// has been optimized for both speed and memory footprint in most common use +// cases. Its interface is similar to that of `std::unordered_map` with +// the following notable differences: +// +// * Requires keys that are CopyConstructible +// * Requires values that are MoveConstructible +// * Supports heterogeneous lookup, through `find()`, `operator[]()` and +// `insert()`, provided that the map is provided a compatible heterogeneous +// hashing function and equality operator. +// * Invalidates any references and pointers to elements within the table after +// `rehash()`. +// * Contains a `capacity()` member function indicating the number of element +// slots (open, deleted, and empty) within the hash map. +// * Returns `void` from the `erase(iterator)` overload. +// +// By default, `flat_hash_map` uses the `absl::Hash` hashing framework. +// All fundamental and Abseil types that support the `absl::Hash` framework have +// a compatible equality operator for comparing insertions into `flat_hash_map`. +// If your type is not yet supported by the `absl::Hash` framework, see +// absl/hash/hash.h for information on extending Abseil hashing to user-defined +// types. +// +// Using `absl::flat_hash_map` at interface boundaries in dynamically loaded +// libraries (e.g. .dll, .so) is unsupported due to way `absl::Hash` values may +// be randomized across dynamically loaded libraries. +// +// NOTE: A `flat_hash_map` stores its value types directly inside its +// implementation array to avoid memory indirection. Because a `flat_hash_map` +// is designed to move data when rehashed, map values will not retain pointer +// stability. If you require pointer stability, or if your values are large, +// consider using `absl::flat_hash_map>` instead. +// If your types are not moveable or you require pointer stability for keys, +// consider `absl::node_hash_map`. +// +// Example: +// +// // Create a flat hash map of three strings (that map to strings) +// absl::flat_hash_map ducks = +// {{"a", "huey"}, {"b", "dewey"}, {"c", "louie"}}; +// +// // Insert a new element into the flat hash map +// ducks.insert({"d", "donald"}); +// +// // Force a rehash of the flat hash map +// ducks.rehash(0); +// +// // Find the element with the key "b" +// std::string search_key = "b"; +// auto result = ducks.find(search_key); +// if (result != ducks.end()) { +// std::cout << "Result: " << result->second << std::endl; +// } +template , + class Eq = absl::container_internal::hash_default_eq, + class Allocator = std::allocator>> +class flat_hash_map : public absl::container_internal::raw_hash_map< + absl::container_internal::FlatHashMapPolicy, + Hash, Eq, Allocator> { + using Base = typename flat_hash_map::raw_hash_map; + + public: + // Constructors and Assignment Operators + // + // A flat_hash_map supports the same overload set as `std::unordered_map` + // for construction and assignment: + // + // * Default constructor + // + // // No allocation for the table's elements is made. + // absl::flat_hash_map map1; + // + // * Initializer List constructor + // + // absl::flat_hash_map map2 = + // {{1, "huey"}, {2, "dewey"}, {3, "louie"},}; + // + // * Copy constructor + // + // absl::flat_hash_map map3(map2); + // + // * Copy assignment operator + // + // // Hash functor and Comparator are copied as well + // absl::flat_hash_map map4; + // map4 = map3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::flat_hash_map map5(std::move(map4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::flat_hash_map map6; + // map6 = std::move(map5); + // + // * Range constructor + // + // std::vector> v = {{1, "a"}, {2, "b"}}; + // absl::flat_hash_map map7(v.begin(), v.end()); + flat_hash_map() {} + using Base::Base; + + // flat_hash_map::begin() + // + // Returns an iterator to the beginning of the `flat_hash_map`. + using Base::begin; + + // flat_hash_map::cbegin() + // + // Returns a const iterator to the beginning of the `flat_hash_map`. + using Base::cbegin; + + // flat_hash_map::cend() + // + // Returns a const iterator to the end of the `flat_hash_map`. + using Base::cend; + + // flat_hash_map::end() + // + // Returns an iterator to the end of the `flat_hash_map`. + using Base::end; + + // flat_hash_map::capacity() + // + // Returns the number of element slots (assigned, deleted, and empty) + // available within the `flat_hash_map`. + // + // NOTE: this member function is particular to `absl::flat_hash_map` and is + // not provided in the `std::unordered_map` API. + using Base::capacity; + + // flat_hash_map::empty() + // + // Returns whether or not the `flat_hash_map` is empty. + using Base::empty; + + // flat_hash_map::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `flat_hash_map` under current memory constraints. This value can be thought + // of the largest value of `std::distance(begin(), end())` for a + // `flat_hash_map`. + using Base::max_size; + + // flat_hash_map::size() + // + // Returns the number of elements currently within the `flat_hash_map`. + using Base::size; + + // flat_hash_map::clear() + // + // Removes all elements from the `flat_hash_map`. Invalidates any references, + // pointers, or iterators referring to contained elements. + // + // NOTE: this operation may shrink the underlying buffer. To avoid shrinking + // the underlying buffer call `erase(begin(), end())`. + using Base::clear; + + // flat_hash_map::erase() + // + // Erases elements within the `flat_hash_map`. Erasing does not trigger a + // rehash. Overloads are listed below. + // + // void erase(const_iterator pos): + // + // Erases the element at `position` of the `flat_hash_map`, returning + // `void`. + // + // NOTE: returning `void` in this case is different than that of STL + // containers in general and `std::unordered_map` in particular (which + // return an iterator to the element following the erased element). If that + // iterator is needed, simply post increment the iterator: + // + // map.erase(it++); + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning an + // iterator pointing to `last`. + // + // size_type erase(const key_type& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // flat_hash_map::insert() + // + // Inserts an element of the specified value into the `flat_hash_map`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If rehashing occurs + // due to the insertion, all iterators are invalidated. Overloads are listed + // below. + // + // std::pair insert(const init_type& value): + // + // Inserts a value into the `flat_hash_map`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a bool denoting whether the insertion took place. + // + // std::pair insert(T&& value): + // std::pair insert(init_type&& value): + // + // Inserts a moveable value into the `flat_hash_map`. Returns a pair + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a bool denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const init_type& value): + // iterator insert(const_iterator hint, T&& value): + // iterator insert(const_iterator hint, init_type&& value); + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently, for `flat_hash_map` we guarantee the + // first match is inserted. + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently within the initializer list, for + // `flat_hash_map` we guarantee the first match is inserted. + using Base::insert; + + // flat_hash_map::insert_or_assign() + // + // Inserts an element of the specified value into the `flat_hash_map` provided + // that a value with the given key does not already exist, or replaces it with + // the element value if a key for that value already exists, returning an + // iterator pointing to the newly inserted element. If rehashing occurs due + // to the insertion, all existing iterators are invalidated. Overloads are + // listed below. + // + // pair insert_or_assign(const init_type& k, T&& obj): + // pair insert_or_assign(init_type&& k, T&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `flat_hash_map`. + // + // iterator insert_or_assign(const_iterator hint, + // const init_type& k, T&& obj): + // iterator insert_or_assign(const_iterator hint, init_type&& k, T&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `flat_hash_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + using Base::insert_or_assign; + + // flat_hash_map::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `flat_hash_map`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace; + + // flat_hash_map::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `flat_hash_map`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace_hint; + + // flat_hash_map::try_emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `flat_hash_map`, provided that no element with the given key + // already exists. Unlike `emplace()`, if an element with the given key + // already exists, we guarantee that no element is constructed. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + // Overloads are listed below. + // + // pair try_emplace(const key_type& k, Args&&... args): + // pair try_emplace(key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `flat_hash_map`. + // + // iterator try_emplace(const_iterator hint, + // const key_type& k, Args&&... args): + // iterator try_emplace(const_iterator hint, key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `flat_hash_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + // + // All `try_emplace()` overloads make the same guarantees regarding rvalue + // arguments as `std::unordered_map::try_emplace()`, namely that these + // functions will not move from rvalue arguments if insertions do not happen. + using Base::try_emplace; + + // flat_hash_map::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the key,value pair of the element at the indicated position and + // returns a node handle owning that extracted data. + // + // node_type extract(const key_type& x): + // + // Extracts the key,value pair of the element with a key matching the passed + // key value and returns a node handle owning that extracted data. If the + // `flat_hash_map` does not contain an element with a matching key, this + // function returns an empty node handle. + // + // NOTE: when compiled in an earlier version of C++ than C++17, + // `node_type::key()` returns a const reference to the key instead of a + // mutable reference. We cannot safely return a mutable reference without + // std::launder (which is not available before C++17). + using Base::extract; + + // flat_hash_map::merge() + // + // Extracts elements from a given `source` flat hash map into this + // `flat_hash_map`. If the destination `flat_hash_map` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // flat_hash_map::swap(flat_hash_map& other) + // + // Exchanges the contents of this `flat_hash_map` with those of the `other` + // flat hash map, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `flat_hash_map` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + // + // `swap()` requires that the flat hash map's hashing and key equivalence + // functions be Swappable, and are exchanged using unqualified calls to + // non-member `swap()`. If the map's allocator has + // `std::allocator_traits::propagate_on_container_swap::value` + // set to `true`, the allocators are also exchanged using an unqualified call + // to non-member `swap()`; otherwise, the allocators are not swapped. + using Base::swap; + + // flat_hash_map::rehash(count) + // + // Rehashes the `flat_hash_map`, setting the number of slots to be at least + // the passed value. If the new number of slots increases the load factor more + // than the current maximum load factor + // (`count` < `size()` / `max_load_factor()`), then the new number of slots + // will be at least `size()` / `max_load_factor()`. + // + // To force a rehash, pass rehash(0). + // + // NOTE: unlike behavior in `std::unordered_map`, references are also + // invalidated upon a `rehash()`. + using Base::rehash; + + // flat_hash_map::reserve(count) + // + // Sets the number of slots in the `flat_hash_map` to the number needed to + // accommodate at least `count` total elements without exceeding the current + // maximum load factor, and may rehash the container if needed. + using Base::reserve; + + // flat_hash_map::at() + // + // Returns a reference to the mapped value of the element with key equivalent + // to the passed key. + using Base::at; + + // flat_hash_map::contains() + // + // Determines whether an element with a key comparing equal to the given `key` + // exists within the `flat_hash_map`, returning `true` if so or `false` + // otherwise. + using Base::contains; + + // flat_hash_map::count(const Key& key) const + // + // Returns the number of elements with a key comparing equal to the given + // `key` within the `flat_hash_map`. note that this function will return + // either `1` or `0` since duplicate keys are not allowed within a + // `flat_hash_map`. + using Base::count; + + // flat_hash_map::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `flat_hash_map`. + using Base::equal_range; + + // flat_hash_map::find() + // + // Finds an element with the passed `key` within the `flat_hash_map`. + using Base::find; + + // flat_hash_map::operator[]() + // + // Returns a reference to the value mapped to the passed key within the + // `flat_hash_map`, performing an `insert()` if the key does not already + // exist. + // + // If an insertion occurs and results in a rehashing of the container, all + // iterators are invalidated. Otherwise iterators are not affected and + // references are not invalidated. Overloads are listed below. + // + // T& operator[](const Key& key): + // + // Inserts an init_type object constructed in-place if the element with the + // given key does not exist. + // + // T& operator[](Key&& key): + // + // Inserts an init_type object constructed in-place provided that an element + // with the given key does not exist. + using Base::operator[]; + + // flat_hash_map::bucket_count() + // + // Returns the number of "buckets" within the `flat_hash_map`. Note that + // because a flat hash map contains all elements within its internal storage, + // this value simply equals the current capacity of the `flat_hash_map`. + using Base::bucket_count; + + // flat_hash_map::load_factor() + // + // Returns the current load factor of the `flat_hash_map` (the average number + // of slots occupied with a value within the hash map). + using Base::load_factor; + + // flat_hash_map::max_load_factor() + // + // Manages the maximum load factor of the `flat_hash_map`. Overloads are + // listed below. + // + // float flat_hash_map::max_load_factor() + // + // Returns the current maximum load factor of the `flat_hash_map`. + // + // void flat_hash_map::max_load_factor(float ml) + // + // Sets the maximum load factor of the `flat_hash_map` to the passed value. + // + // NOTE: This overload is provided only for API compatibility with the STL; + // `flat_hash_map` will ignore any set load factor and manage its rehashing + // internally as an implementation detail. + using Base::max_load_factor; + + // flat_hash_map::get_allocator() + // + // Returns the allocator function associated with this `flat_hash_map`. + using Base::get_allocator; + + // flat_hash_map::hash_function() + // + // Returns the hashing function used to hash the keys within this + // `flat_hash_map`. + using Base::hash_function; + + // flat_hash_map::key_eq() + // + // Returns the function used for comparing keys equality. + using Base::key_eq; +}; + +// erase_if(flat_hash_map<>, Pred) +// +// Erases all elements that satisfy the predicate `pred` from the container `c`. +// Returns the number of erased elements. +template +typename flat_hash_map::size_type erase_if( + flat_hash_map& c, Predicate pred) { + return container_internal::EraseIf(pred, &c); +} + +namespace container_internal { + +template +struct FlatHashMapPolicy { + using slot_policy = container_internal::map_slot_policy; + using slot_type = typename slot_policy::slot_type; + using key_type = K; + using mapped_type = V; + using init_type = std::pair; + + template + static void construct(Allocator* alloc, slot_type* slot, Args&&... args) { + slot_policy::construct(alloc, slot, std::forward(args)...); + } + + template + static void destroy(Allocator* alloc, slot_type* slot) { + slot_policy::destroy(alloc, slot); + } + + template + static void transfer(Allocator* alloc, slot_type* new_slot, + slot_type* old_slot) { + slot_policy::transfer(alloc, new_slot, old_slot); + } + + template + static decltype(absl::container_internal::DecomposePair( + std::declval(), std::declval()...)) + apply(F&& f, Args&&... args) { + return absl::container_internal::DecomposePair(std::forward(f), + std::forward(args)...); + } + + static size_t space_used(const slot_type*) { return 0; } + + static std::pair& element(slot_type* slot) { return slot->value; } + + static V& value(std::pair* kv) { return kv->second; } + static const V& value(const std::pair* kv) { return kv->second; } +}; + +} // namespace container_internal + +namespace container_algorithm_internal { + +// Specialization of trait in absl/algorithm/container.h +template +struct IsUnorderedContainer< + absl::flat_hash_map> : std::true_type {}; + +} // namespace container_algorithm_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_FLAT_HASH_MAP_H_ diff --git a/src/absl/container/flat_hash_set.h b/src/absl/container/flat_hash_set.h new file mode 100644 index 0000000..4938c70 --- /dev/null +++ b/src/absl/container/flat_hash_set.h @@ -0,0 +1,510 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: flat_hash_set.h +// ----------------------------------------------------------------------------- +// +// An `absl::flat_hash_set` is an unordered associative container designed to +// be a more efficient replacement for `std::unordered_set`. Like +// `unordered_set`, search, insertion, and deletion of set elements can be done +// as an `O(1)` operation. However, `flat_hash_set` (and other unordered +// associative containers known as the collection of Abseil "Swiss tables") +// contain other optimizations that result in both memory and computation +// advantages. +// +// In most cases, your default choice for a hash set should be a set of type +// `flat_hash_set`. +#ifndef ABSL_CONTAINER_FLAT_HASH_SET_H_ +#define ABSL_CONTAINER_FLAT_HASH_SET_H_ + +#include +#include + +#include "absl/algorithm/container.h" +#include "absl/base/macros.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/hash_function_defaults.h" // IWYU pragma: export +#include "absl/container/internal/raw_hash_set.h" // IWYU pragma: export +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { +template +struct FlatHashSetPolicy; +} // namespace container_internal + +// ----------------------------------------------------------------------------- +// absl::flat_hash_set +// ----------------------------------------------------------------------------- +// +// An `absl::flat_hash_set` is an unordered associative container which has +// been optimized for both speed and memory footprint in most common use cases. +// Its interface is similar to that of `std::unordered_set` with the +// following notable differences: +// +// * Requires keys that are CopyConstructible +// * Supports heterogeneous lookup, through `find()` and `insert()`, provided +// that the set is provided a compatible heterogeneous hashing function and +// equality operator. +// * Invalidates any references and pointers to elements within the table after +// `rehash()`. +// * Contains a `capacity()` member function indicating the number of element +// slots (open, deleted, and empty) within the hash set. +// * Returns `void` from the `erase(iterator)` overload. +// +// By default, `flat_hash_set` uses the `absl::Hash` hashing framework. All +// fundamental and Abseil types that support the `absl::Hash` framework have a +// compatible equality operator for comparing insertions into `flat_hash_set`. +// If your type is not yet supported by the `absl::Hash` framework, see +// absl/hash/hash.h for information on extending Abseil hashing to user-defined +// types. +// +// Using `absl::flat_hash_set` at interface boundaries in dynamically loaded +// libraries (e.g. .dll, .so) is unsupported due to way `absl::Hash` values may +// be randomized across dynamically loaded libraries. +// +// NOTE: A `flat_hash_set` stores its keys directly inside its implementation +// array to avoid memory indirection. Because a `flat_hash_set` is designed to +// move data when rehashed, set keys will not retain pointer stability. If you +// require pointer stability, consider using +// `absl::flat_hash_set>`. If your type is not moveable and +// you require pointer stability, consider `absl::node_hash_set` instead. +// +// Example: +// +// // Create a flat hash set of three strings +// absl::flat_hash_set ducks = +// {"huey", "dewey", "louie"}; +// +// // Insert a new element into the flat hash set +// ducks.insert("donald"); +// +// // Force a rehash of the flat hash set +// ducks.rehash(0); +// +// // See if "dewey" is present +// if (ducks.contains("dewey")) { +// std::cout << "We found dewey!" << std::endl; +// } +template , + class Eq = absl::container_internal::hash_default_eq, + class Allocator = std::allocator> +class flat_hash_set + : public absl::container_internal::raw_hash_set< + absl::container_internal::FlatHashSetPolicy, Hash, Eq, Allocator> { + using Base = typename flat_hash_set::raw_hash_set; + + public: + // Constructors and Assignment Operators + // + // A flat_hash_set supports the same overload set as `std::unordered_set` + // for construction and assignment: + // + // * Default constructor + // + // // No allocation for the table's elements is made. + // absl::flat_hash_set set1; + // + // * Initializer List constructor + // + // absl::flat_hash_set set2 = + // {{"huey"}, {"dewey"}, {"louie"},}; + // + // * Copy constructor + // + // absl::flat_hash_set set3(set2); + // + // * Copy assignment operator + // + // // Hash functor and Comparator are copied as well + // absl::flat_hash_set set4; + // set4 = set3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::flat_hash_set set5(std::move(set4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::flat_hash_set set6; + // set6 = std::move(set5); + // + // * Range constructor + // + // std::vector v = {"a", "b"}; + // absl::flat_hash_set set7(v.begin(), v.end()); + flat_hash_set() {} + using Base::Base; + + // flat_hash_set::begin() + // + // Returns an iterator to the beginning of the `flat_hash_set`. + using Base::begin; + + // flat_hash_set::cbegin() + // + // Returns a const iterator to the beginning of the `flat_hash_set`. + using Base::cbegin; + + // flat_hash_set::cend() + // + // Returns a const iterator to the end of the `flat_hash_set`. + using Base::cend; + + // flat_hash_set::end() + // + // Returns an iterator to the end of the `flat_hash_set`. + using Base::end; + + // flat_hash_set::capacity() + // + // Returns the number of element slots (assigned, deleted, and empty) + // available within the `flat_hash_set`. + // + // NOTE: this member function is particular to `absl::flat_hash_set` and is + // not provided in the `std::unordered_set` API. + using Base::capacity; + + // flat_hash_set::empty() + // + // Returns whether or not the `flat_hash_set` is empty. + using Base::empty; + + // flat_hash_set::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `flat_hash_set` under current memory constraints. This value can be thought + // of the largest value of `std::distance(begin(), end())` for a + // `flat_hash_set`. + using Base::max_size; + + // flat_hash_set::size() + // + // Returns the number of elements currently within the `flat_hash_set`. + using Base::size; + + // flat_hash_set::clear() + // + // Removes all elements from the `flat_hash_set`. Invalidates any references, + // pointers, or iterators referring to contained elements. + // + // NOTE: this operation may shrink the underlying buffer. To avoid shrinking + // the underlying buffer call `erase(begin(), end())`. + using Base::clear; + + // flat_hash_set::erase() + // + // Erases elements within the `flat_hash_set`. Erasing does not trigger a + // rehash. Overloads are listed below. + // + // void erase(const_iterator pos): + // + // Erases the element at `position` of the `flat_hash_set`, returning + // `void`. + // + // NOTE: returning `void` in this case is different than that of STL + // containers in general and `std::unordered_set` in particular (which + // return an iterator to the element following the erased element). If that + // iterator is needed, simply post increment the iterator: + // + // set.erase(it++); + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning an + // iterator pointing to `last`. + // + // size_type erase(const key_type& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // flat_hash_set::insert() + // + // Inserts an element of the specified value into the `flat_hash_set`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If rehashing occurs + // due to the insertion, all iterators are invalidated. Overloads are listed + // below. + // + // std::pair insert(const T& value): + // + // Inserts a value into the `flat_hash_set`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a bool denoting whether the insertion took place. + // + // std::pair insert(T&& value): + // + // Inserts a moveable value into the `flat_hash_set`. Returns a pair + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a bool denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const T& value): + // iterator insert(const_iterator hint, T&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently, for `flat_hash_set` we guarantee the + // first match is inserted. + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently within the initializer list, for + // `flat_hash_set` we guarantee the first match is inserted. + using Base::insert; + + // flat_hash_set::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `flat_hash_set`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace; + + // flat_hash_set::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `flat_hash_set`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace_hint; + + // flat_hash_set::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // node_type extract(const key_type& x): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `flat_hash_set` + // does not contain an element with a matching key, this function returns an + // empty node handle. + using Base::extract; + + // flat_hash_set::merge() + // + // Extracts elements from a given `source` flat hash set into this + // `flat_hash_set`. If the destination `flat_hash_set` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // flat_hash_set::swap(flat_hash_set& other) + // + // Exchanges the contents of this `flat_hash_set` with those of the `other` + // flat hash set, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `flat_hash_set` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + // + // `swap()` requires that the flat hash set's hashing and key equivalence + // functions be Swappable, and are exchaged using unqualified calls to + // non-member `swap()`. If the set's allocator has + // `std::allocator_traits::propagate_on_container_swap::value` + // set to `true`, the allocators are also exchanged using an unqualified call + // to non-member `swap()`; otherwise, the allocators are not swapped. + using Base::swap; + + // flat_hash_set::rehash(count) + // + // Rehashes the `flat_hash_set`, setting the number of slots to be at least + // the passed value. If the new number of slots increases the load factor more + // than the current maximum load factor + // (`count` < `size()` / `max_load_factor()`), then the new number of slots + // will be at least `size()` / `max_load_factor()`. + // + // To force a rehash, pass rehash(0). + // + // NOTE: unlike behavior in `std::unordered_set`, references are also + // invalidated upon a `rehash()`. + using Base::rehash; + + // flat_hash_set::reserve(count) + // + // Sets the number of slots in the `flat_hash_set` to the number needed to + // accommodate at least `count` total elements without exceeding the current + // maximum load factor, and may rehash the container if needed. + using Base::reserve; + + // flat_hash_set::contains() + // + // Determines whether an element comparing equal to the given `key` exists + // within the `flat_hash_set`, returning `true` if so or `false` otherwise. + using Base::contains; + + // flat_hash_set::count(const Key& key) const + // + // Returns the number of elements comparing equal to the given `key` within + // the `flat_hash_set`. note that this function will return either `1` or `0` + // since duplicate elements are not allowed within a `flat_hash_set`. + using Base::count; + + // flat_hash_set::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `flat_hash_set`. + using Base::equal_range; + + // flat_hash_set::find() + // + // Finds an element with the passed `key` within the `flat_hash_set`. + using Base::find; + + // flat_hash_set::bucket_count() + // + // Returns the number of "buckets" within the `flat_hash_set`. Note that + // because a flat hash set contains all elements within its internal storage, + // this value simply equals the current capacity of the `flat_hash_set`. + using Base::bucket_count; + + // flat_hash_set::load_factor() + // + // Returns the current load factor of the `flat_hash_set` (the average number + // of slots occupied with a value within the hash set). + using Base::load_factor; + + // flat_hash_set::max_load_factor() + // + // Manages the maximum load factor of the `flat_hash_set`. Overloads are + // listed below. + // + // float flat_hash_set::max_load_factor() + // + // Returns the current maximum load factor of the `flat_hash_set`. + // + // void flat_hash_set::max_load_factor(float ml) + // + // Sets the maximum load factor of the `flat_hash_set` to the passed value. + // + // NOTE: This overload is provided only for API compatibility with the STL; + // `flat_hash_set` will ignore any set load factor and manage its rehashing + // internally as an implementation detail. + using Base::max_load_factor; + + // flat_hash_set::get_allocator() + // + // Returns the allocator function associated with this `flat_hash_set`. + using Base::get_allocator; + + // flat_hash_set::hash_function() + // + // Returns the hashing function used to hash the keys within this + // `flat_hash_set`. + using Base::hash_function; + + // flat_hash_set::key_eq() + // + // Returns the function used for comparing keys equality. + using Base::key_eq; +}; + +// erase_if(flat_hash_set<>, Pred) +// +// Erases all elements that satisfy the predicate `pred` from the container `c`. +// Returns the number of erased elements. +template +typename flat_hash_set::size_type erase_if( + flat_hash_set& c, Predicate pred) { + return container_internal::EraseIf(pred, &c); +} + +namespace container_internal { + +template +struct FlatHashSetPolicy { + using slot_type = T; + using key_type = T; + using init_type = T; + using constant_iterators = std::true_type; + + template + static void construct(Allocator* alloc, slot_type* slot, Args&&... args) { + absl::allocator_traits::construct(*alloc, slot, + std::forward(args)...); + } + + template + static void destroy(Allocator* alloc, slot_type* slot) { + absl::allocator_traits::destroy(*alloc, slot); + } + + template + static void transfer(Allocator* alloc, slot_type* new_slot, + slot_type* old_slot) { + construct(alloc, new_slot, std::move(*old_slot)); + destroy(alloc, old_slot); + } + + static T& element(slot_type* slot) { return *slot; } + + template + static decltype(absl::container_internal::DecomposeValue( + std::declval(), std::declval()...)) + apply(F&& f, Args&&... args) { + return absl::container_internal::DecomposeValue( + std::forward(f), std::forward(args)...); + } + + static size_t space_used(const T*) { return 0; } +}; +} // namespace container_internal + +namespace container_algorithm_internal { + +// Specialization of trait in absl/algorithm/container.h +template +struct IsUnorderedContainer> + : std::true_type {}; + +} // namespace container_algorithm_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_FLAT_HASH_SET_H_ diff --git a/src/absl/container/inlined_vector.h b/src/absl/container/inlined_vector.h new file mode 100644 index 0000000..e8d1019 --- /dev/null +++ b/src/absl/container/inlined_vector.h @@ -0,0 +1,866 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: inlined_vector.h +// ----------------------------------------------------------------------------- +// +// This header file contains the declaration and definition of an "inlined +// vector" which behaves in an equivalent fashion to a `std::vector`, except +// that storage for small sequences of the vector are provided inline without +// requiring any heap allocation. +// +// An `absl::InlinedVector` specifies the default capacity `N` as one of +// its template parameters. Instances where `size() <= N` hold contained +// elements in inline space. Typically `N` is very small so that sequences that +// are expected to be short do not require allocations. +// +// An `absl::InlinedVector` does not usually require a specific allocator. If +// the inlined vector grows beyond its initial constraints, it will need to +// allocate (as any normal `std::vector` would). This is usually performed with +// the default allocator (defined as `std::allocator`). Optionally, a custom +// allocator type may be specified as `A` in `absl::InlinedVector`. + +#ifndef ABSL_CONTAINER_INLINED_VECTOR_H_ +#define ABSL_CONTAINER_INLINED_VECTOR_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/algorithm/algorithm.h" +#include "absl/base/internal/throw_delegate.h" +#include "absl/base/macros.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" +#include "absl/container/internal/inlined_vector.h" +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +// ----------------------------------------------------------------------------- +// InlinedVector +// ----------------------------------------------------------------------------- +// +// An `absl::InlinedVector` is designed to be a drop-in replacement for +// `std::vector` for use cases where the vector's size is sufficiently small +// that it can be inlined. If the inlined vector does grow beyond its estimated +// capacity, it will trigger an initial allocation on the heap, and will behave +// as a `std::vector`. The API of the `absl::InlinedVector` within this file is +// designed to cover the same API footprint as covered by `std::vector`. +template > +class InlinedVector { + static_assert(N > 0, "`absl::InlinedVector` requires an inlined capacity."); + + using Storage = inlined_vector_internal::Storage; + + template + using AllocatorTraits = inlined_vector_internal::AllocatorTraits; + template + using MoveIterator = inlined_vector_internal::MoveIterator; + template + using IsMemcpyOk = inlined_vector_internal::IsMemcpyOk; + + template + using IteratorValueAdapter = + inlined_vector_internal::IteratorValueAdapter; + template + using CopyValueAdapter = inlined_vector_internal::CopyValueAdapter; + template + using DefaultValueAdapter = + inlined_vector_internal::DefaultValueAdapter; + + template + using EnableIfAtLeastForwardIterator = absl::enable_if_t< + inlined_vector_internal::IsAtLeastForwardIterator::value, int>; + template + using DisableIfAtLeastForwardIterator = absl::enable_if_t< + !inlined_vector_internal::IsAtLeastForwardIterator::value, int>; + + public: + using allocator_type = A; + using value_type = inlined_vector_internal::ValueType; + using pointer = inlined_vector_internal::Pointer; + using const_pointer = inlined_vector_internal::ConstPointer; + using size_type = inlined_vector_internal::SizeType; + using difference_type = inlined_vector_internal::DifferenceType; + using reference = inlined_vector_internal::Reference; + using const_reference = inlined_vector_internal::ConstReference; + using iterator = inlined_vector_internal::Iterator; + using const_iterator = inlined_vector_internal::ConstIterator; + using reverse_iterator = inlined_vector_internal::ReverseIterator; + using const_reverse_iterator = + inlined_vector_internal::ConstReverseIterator; + + // --------------------------------------------------------------------------- + // InlinedVector Constructors and Destructor + // --------------------------------------------------------------------------- + + // Creates an empty inlined vector with a value-initialized allocator. + InlinedVector() noexcept(noexcept(allocator_type())) : storage_() {} + + // Creates an empty inlined vector with a copy of `allocator`. + explicit InlinedVector(const allocator_type& allocator) noexcept + : storage_(allocator) {} + + // Creates an inlined vector with `n` copies of `value_type()`. + explicit InlinedVector(size_type n, + const allocator_type& allocator = allocator_type()) + : storage_(allocator) { + storage_.Initialize(DefaultValueAdapter(), n); + } + + // Creates an inlined vector with `n` copies of `v`. + InlinedVector(size_type n, const_reference v, + const allocator_type& allocator = allocator_type()) + : storage_(allocator) { + storage_.Initialize(CopyValueAdapter(std::addressof(v)), n); + } + + // Creates an inlined vector with copies of the elements of `list`. + InlinedVector(std::initializer_list list, + const allocator_type& allocator = allocator_type()) + : InlinedVector(list.begin(), list.end(), allocator) {} + + // Creates an inlined vector with elements constructed from the provided + // forward iterator range [`first`, `last`). + // + // NOTE: the `enable_if` prevents ambiguous interpretation between a call to + // this constructor with two integral arguments and a call to the above + // `InlinedVector(size_type, const_reference)` constructor. + template = 0> + InlinedVector(ForwardIterator first, ForwardIterator last, + const allocator_type& allocator = allocator_type()) + : storage_(allocator) { + storage_.Initialize(IteratorValueAdapter(first), + static_cast(std::distance(first, last))); + } + + // Creates an inlined vector with elements constructed from the provided input + // iterator range [`first`, `last`). + template = 0> + InlinedVector(InputIterator first, InputIterator last, + const allocator_type& allocator = allocator_type()) + : storage_(allocator) { + std::copy(first, last, std::back_inserter(*this)); + } + + // Creates an inlined vector by copying the contents of `other` using + // `other`'s allocator. + InlinedVector(const InlinedVector& other) + : InlinedVector(other, other.storage_.GetAllocator()) {} + + // Creates an inlined vector by copying the contents of `other` using the + // provided `allocator`. + InlinedVector(const InlinedVector& other, const allocator_type& allocator) + : storage_(allocator) { + if (other.empty()) { + // Empty; nothing to do. + } else if (IsMemcpyOk::value && !other.storage_.GetIsAllocated()) { + // Memcpy-able and do not need allocation. + storage_.MemcpyFrom(other.storage_); + } else { + storage_.InitFrom(other.storage_); + } + } + + // Creates an inlined vector by moving in the contents of `other` without + // allocating. If `other` contains allocated memory, the newly-created inlined + // vector will take ownership of that memory. However, if `other` does not + // contain allocated memory, the newly-created inlined vector will perform + // element-wise move construction of the contents of `other`. + // + // NOTE: since no allocation is performed for the inlined vector in either + // case, the `noexcept(...)` specification depends on whether moving the + // underlying objects can throw. It is assumed assumed that... + // a) move constructors should only throw due to allocation failure. + // b) if `value_type`'s move constructor allocates, it uses the same + // allocation function as the inlined vector's allocator. + // Thus, the move constructor is non-throwing if the allocator is non-throwing + // or `value_type`'s move constructor is specified as `noexcept`. + InlinedVector(InlinedVector&& other) noexcept( + absl::allocator_is_nothrow::value || + std::is_nothrow_move_constructible::value) + : storage_(other.storage_.GetAllocator()) { + if (IsMemcpyOk::value) { + storage_.MemcpyFrom(other.storage_); + + other.storage_.SetInlinedSize(0); + } else if (other.storage_.GetIsAllocated()) { + storage_.SetAllocation({other.storage_.GetAllocatedData(), + other.storage_.GetAllocatedCapacity()}); + storage_.SetAllocatedSize(other.storage_.GetSize()); + + other.storage_.SetInlinedSize(0); + } else { + IteratorValueAdapter> other_values( + MoveIterator(other.storage_.GetInlinedData())); + + inlined_vector_internal::ConstructElements( + storage_.GetAllocator(), storage_.GetInlinedData(), other_values, + other.storage_.GetSize()); + + storage_.SetInlinedSize(other.storage_.GetSize()); + } + } + + // Creates an inlined vector by moving in the contents of `other` with a copy + // of `allocator`. + // + // NOTE: if `other`'s allocator is not equal to `allocator`, even if `other` + // contains allocated memory, this move constructor will still allocate. Since + // allocation is performed, this constructor can only be `noexcept` if the + // specified allocator is also `noexcept`. + InlinedVector( + InlinedVector&& other, + const allocator_type& + allocator) noexcept(absl::allocator_is_nothrow::value) + : storage_(allocator) { + if (IsMemcpyOk::value) { + storage_.MemcpyFrom(other.storage_); + + other.storage_.SetInlinedSize(0); + } else if ((storage_.GetAllocator() == other.storage_.GetAllocator()) && + other.storage_.GetIsAllocated()) { + storage_.SetAllocation({other.storage_.GetAllocatedData(), + other.storage_.GetAllocatedCapacity()}); + storage_.SetAllocatedSize(other.storage_.GetSize()); + + other.storage_.SetInlinedSize(0); + } else { + storage_.Initialize(IteratorValueAdapter>( + MoveIterator(other.data())), + other.size()); + } + } + + ~InlinedVector() {} + + // --------------------------------------------------------------------------- + // InlinedVector Member Accessors + // --------------------------------------------------------------------------- + + // `InlinedVector::empty()` + // + // Returns whether the inlined vector contains no elements. + bool empty() const noexcept { return !size(); } + + // `InlinedVector::size()` + // + // Returns the number of elements in the inlined vector. + size_type size() const noexcept { return storage_.GetSize(); } + + // `InlinedVector::max_size()` + // + // Returns the maximum number of elements the inlined vector can hold. + size_type max_size() const noexcept { + // One bit of the size storage is used to indicate whether the inlined + // vector contains allocated memory. As a result, the maximum size that the + // inlined vector can express is half of the max for `size_type`. + return (std::numeric_limits::max)() / 2; + } + + // `InlinedVector::capacity()` + // + // Returns the number of elements that could be stored in the inlined vector + // without requiring a reallocation. + // + // NOTE: for most inlined vectors, `capacity()` should be equal to the + // template parameter `N`. For inlined vectors which exceed this capacity, + // they will no longer be inlined and `capacity()` will equal the capactity of + // the allocated memory. + size_type capacity() const noexcept { + return storage_.GetIsAllocated() ? storage_.GetAllocatedCapacity() + : storage_.GetInlinedCapacity(); + } + + // `InlinedVector::data()` + // + // Returns a `pointer` to the elements of the inlined vector. This pointer + // can be used to access and modify the contained elements. + // + // NOTE: only elements within [`data()`, `data() + size()`) are valid. + pointer data() noexcept { + return storage_.GetIsAllocated() ? storage_.GetAllocatedData() + : storage_.GetInlinedData(); + } + + // Overload of `InlinedVector::data()` that returns a `const_pointer` to the + // elements of the inlined vector. This pointer can be used to access but not + // modify the contained elements. + // + // NOTE: only elements within [`data()`, `data() + size()`) are valid. + const_pointer data() const noexcept { + return storage_.GetIsAllocated() ? storage_.GetAllocatedData() + : storage_.GetInlinedData(); + } + + // `InlinedVector::operator[](...)` + // + // Returns a `reference` to the `i`th element of the inlined vector. + reference operator[](size_type i) { + ABSL_HARDENING_ASSERT(i < size()); + return data()[i]; + } + + // Overload of `InlinedVector::operator[](...)` that returns a + // `const_reference` to the `i`th element of the inlined vector. + const_reference operator[](size_type i) const { + ABSL_HARDENING_ASSERT(i < size()); + return data()[i]; + } + + // `InlinedVector::at(...)` + // + // Returns a `reference` to the `i`th element of the inlined vector. + // + // NOTE: if `i` is not within the required range of `InlinedVector::at(...)`, + // in both debug and non-debug builds, `std::out_of_range` will be thrown. + reference at(size_type i) { + if (ABSL_PREDICT_FALSE(i >= size())) { + base_internal::ThrowStdOutOfRange( + "`InlinedVector::at(size_type)` failed bounds check"); + } + return data()[i]; + } + + // Overload of `InlinedVector::at(...)` that returns a `const_reference` to + // the `i`th element of the inlined vector. + // + // NOTE: if `i` is not within the required range of `InlinedVector::at(...)`, + // in both debug and non-debug builds, `std::out_of_range` will be thrown. + const_reference at(size_type i) const { + if (ABSL_PREDICT_FALSE(i >= size())) { + base_internal::ThrowStdOutOfRange( + "`InlinedVector::at(size_type) const` failed bounds check"); + } + return data()[i]; + } + + // `InlinedVector::front()` + // + // Returns a `reference` to the first element of the inlined vector. + reference front() { + ABSL_HARDENING_ASSERT(!empty()); + return data()[0]; + } + + // Overload of `InlinedVector::front()` that returns a `const_reference` to + // the first element of the inlined vector. + const_reference front() const { + ABSL_HARDENING_ASSERT(!empty()); + return data()[0]; + } + + // `InlinedVector::back()` + // + // Returns a `reference` to the last element of the inlined vector. + reference back() { + ABSL_HARDENING_ASSERT(!empty()); + return data()[size() - 1]; + } + + // Overload of `InlinedVector::back()` that returns a `const_reference` to the + // last element of the inlined vector. + const_reference back() const { + ABSL_HARDENING_ASSERT(!empty()); + return data()[size() - 1]; + } + + // `InlinedVector::begin()` + // + // Returns an `iterator` to the beginning of the inlined vector. + iterator begin() noexcept { return data(); } + + // Overload of `InlinedVector::begin()` that returns a `const_iterator` to + // the beginning of the inlined vector. + const_iterator begin() const noexcept { return data(); } + + // `InlinedVector::end()` + // + // Returns an `iterator` to the end of the inlined vector. + iterator end() noexcept { return data() + size(); } + + // Overload of `InlinedVector::end()` that returns a `const_iterator` to the + // end of the inlined vector. + const_iterator end() const noexcept { return data() + size(); } + + // `InlinedVector::cbegin()` + // + // Returns a `const_iterator` to the beginning of the inlined vector. + const_iterator cbegin() const noexcept { return begin(); } + + // `InlinedVector::cend()` + // + // Returns a `const_iterator` to the end of the inlined vector. + const_iterator cend() const noexcept { return end(); } + + // `InlinedVector::rbegin()` + // + // Returns a `reverse_iterator` from the end of the inlined vector. + reverse_iterator rbegin() noexcept { return reverse_iterator(end()); } + + // Overload of `InlinedVector::rbegin()` that returns a + // `const_reverse_iterator` from the end of the inlined vector. + const_reverse_iterator rbegin() const noexcept { + return const_reverse_iterator(end()); + } + + // `InlinedVector::rend()` + // + // Returns a `reverse_iterator` from the beginning of the inlined vector. + reverse_iterator rend() noexcept { return reverse_iterator(begin()); } + + // Overload of `InlinedVector::rend()` that returns a `const_reverse_iterator` + // from the beginning of the inlined vector. + const_reverse_iterator rend() const noexcept { + return const_reverse_iterator(begin()); + } + + // `InlinedVector::crbegin()` + // + // Returns a `const_reverse_iterator` from the end of the inlined vector. + const_reverse_iterator crbegin() const noexcept { return rbegin(); } + + // `InlinedVector::crend()` + // + // Returns a `const_reverse_iterator` from the beginning of the inlined + // vector. + const_reverse_iterator crend() const noexcept { return rend(); } + + // `InlinedVector::get_allocator()` + // + // Returns a copy of the inlined vector's allocator. + allocator_type get_allocator() const { return storage_.GetAllocator(); } + + // --------------------------------------------------------------------------- + // InlinedVector Member Mutators + // --------------------------------------------------------------------------- + + // `InlinedVector::operator=(...)` + // + // Replaces the elements of the inlined vector with copies of the elements of + // `list`. + InlinedVector& operator=(std::initializer_list list) { + assign(list.begin(), list.end()); + + return *this; + } + + // Overload of `InlinedVector::operator=(...)` that replaces the elements of + // the inlined vector with copies of the elements of `other`. + InlinedVector& operator=(const InlinedVector& other) { + if (ABSL_PREDICT_TRUE(this != std::addressof(other))) { + const_pointer other_data = other.data(); + assign(other_data, other_data + other.size()); + } + + return *this; + } + + // Overload of `InlinedVector::operator=(...)` that moves the elements of + // `other` into the inlined vector. + // + // NOTE: as a result of calling this overload, `other` is left in a valid but + // unspecified state. + InlinedVector& operator=(InlinedVector&& other) { + if (ABSL_PREDICT_TRUE(this != std::addressof(other))) { + if (IsMemcpyOk::value || other.storage_.GetIsAllocated()) { + inlined_vector_internal::DestroyAdapter::DestroyElements( + storage_.GetAllocator(), data(), size()); + storage_.DeallocateIfAllocated(); + storage_.MemcpyFrom(other.storage_); + + other.storage_.SetInlinedSize(0); + } else { + storage_.Assign(IteratorValueAdapter>( + MoveIterator(other.storage_.GetInlinedData())), + other.size()); + } + } + + return *this; + } + + // `InlinedVector::assign(...)` + // + // Replaces the contents of the inlined vector with `n` copies of `v`. + void assign(size_type n, const_reference v) { + storage_.Assign(CopyValueAdapter(std::addressof(v)), n); + } + + // Overload of `InlinedVector::assign(...)` that replaces the contents of the + // inlined vector with copies of the elements of `list`. + void assign(std::initializer_list list) { + assign(list.begin(), list.end()); + } + + // Overload of `InlinedVector::assign(...)` to replace the contents of the + // inlined vector with the range [`first`, `last`). + // + // NOTE: this overload is for iterators that are "forward" category or better. + template = 0> + void assign(ForwardIterator first, ForwardIterator last) { + storage_.Assign(IteratorValueAdapter(first), + static_cast(std::distance(first, last))); + } + + // Overload of `InlinedVector::assign(...)` to replace the contents of the + // inlined vector with the range [`first`, `last`). + // + // NOTE: this overload is for iterators that are "input" category. + template = 0> + void assign(InputIterator first, InputIterator last) { + size_type i = 0; + for (; i < size() && first != last; ++i, static_cast(++first)) { + data()[i] = *first; + } + + erase(data() + i, data() + size()); + std::copy(first, last, std::back_inserter(*this)); + } + + // `InlinedVector::resize(...)` + // + // Resizes the inlined vector to contain `n` elements. + // + // NOTE: If `n` is smaller than `size()`, extra elements are destroyed. If `n` + // is larger than `size()`, new elements are value-initialized. + void resize(size_type n) { + ABSL_HARDENING_ASSERT(n <= max_size()); + storage_.Resize(DefaultValueAdapter(), n); + } + + // Overload of `InlinedVector::resize(...)` that resizes the inlined vector to + // contain `n` elements. + // + // NOTE: if `n` is smaller than `size()`, extra elements are destroyed. If `n` + // is larger than `size()`, new elements are copied-constructed from `v`. + void resize(size_type n, const_reference v) { + ABSL_HARDENING_ASSERT(n <= max_size()); + storage_.Resize(CopyValueAdapter(std::addressof(v)), n); + } + + // `InlinedVector::insert(...)` + // + // Inserts a copy of `v` at `pos`, returning an `iterator` to the newly + // inserted element. + iterator insert(const_iterator pos, const_reference v) { + return emplace(pos, v); + } + + // Overload of `InlinedVector::insert(...)` that inserts `v` at `pos` using + // move semantics, returning an `iterator` to the newly inserted element. + iterator insert(const_iterator pos, value_type&& v) { + return emplace(pos, std::move(v)); + } + + // Overload of `InlinedVector::insert(...)` that inserts `n` contiguous copies + // of `v` starting at `pos`, returning an `iterator` pointing to the first of + // the newly inserted elements. + iterator insert(const_iterator pos, size_type n, const_reference v) { + ABSL_HARDENING_ASSERT(pos >= begin()); + ABSL_HARDENING_ASSERT(pos <= end()); + + if (ABSL_PREDICT_TRUE(n != 0)) { + value_type dealias = v; + // https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102329#c2 + // It appears that GCC thinks that since `pos` is a const pointer and may + // point to uninitialized memory at this point, a warning should be + // issued. But `pos` is actually only used to compute an array index to + // write to. +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic push +// #pragma GCC diagnostic ignored "-Wmaybe-uninitialized" +#endif + return storage_.Insert(pos, CopyValueAdapter(std::addressof(dealias)), + n); +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic pop +#endif + } else { + return const_cast(pos); + } + } + + // Overload of `InlinedVector::insert(...)` that inserts copies of the + // elements of `list` starting at `pos`, returning an `iterator` pointing to + // the first of the newly inserted elements. + iterator insert(const_iterator pos, std::initializer_list list) { + return insert(pos, list.begin(), list.end()); + } + + // Overload of `InlinedVector::insert(...)` that inserts the range [`first`, + // `last`) starting at `pos`, returning an `iterator` pointing to the first + // of the newly inserted elements. + // + // NOTE: this overload is for iterators that are "forward" category or better. + template = 0> + iterator insert(const_iterator pos, ForwardIterator first, + ForwardIterator last) { + ABSL_HARDENING_ASSERT(pos >= begin()); + ABSL_HARDENING_ASSERT(pos <= end()); + + if (ABSL_PREDICT_TRUE(first != last)) { + return storage_.Insert(pos, + IteratorValueAdapter(first), + std::distance(first, last)); + } else { + return const_cast(pos); + } + } + + // Overload of `InlinedVector::insert(...)` that inserts the range [`first`, + // `last`) starting at `pos`, returning an `iterator` pointing to the first + // of the newly inserted elements. + // + // NOTE: this overload is for iterators that are "input" category. + template = 0> + iterator insert(const_iterator pos, InputIterator first, InputIterator last) { + ABSL_HARDENING_ASSERT(pos >= begin()); + ABSL_HARDENING_ASSERT(pos <= end()); + + size_type index = std::distance(cbegin(), pos); + for (size_type i = index; first != last; ++i, static_cast(++first)) { + insert(data() + i, *first); + } + + return iterator(data() + index); + } + + // `InlinedVector::emplace(...)` + // + // Constructs and inserts an element using `args...` in the inlined vector at + // `pos`, returning an `iterator` pointing to the newly emplaced element. + template + iterator emplace(const_iterator pos, Args&&... args) { + ABSL_HARDENING_ASSERT(pos >= begin()); + ABSL_HARDENING_ASSERT(pos <= end()); + + value_type dealias(std::forward(args)...); + return storage_.Insert(pos, + IteratorValueAdapter>( + MoveIterator(std::addressof(dealias))), + 1); + } + + // `InlinedVector::emplace_back(...)` + // + // Constructs and inserts an element using `args...` in the inlined vector at + // `end()`, returning a `reference` to the newly emplaced element. + template + reference emplace_back(Args&&... args) { + return storage_.EmplaceBack(std::forward(args)...); + } + + // `InlinedVector::push_back(...)` + // + // Inserts a copy of `v` in the inlined vector at `end()`. + void push_back(const_reference v) { static_cast(emplace_back(v)); } + + // Overload of `InlinedVector::push_back(...)` for inserting `v` at `end()` + // using move semantics. + void push_back(value_type&& v) { + static_cast(emplace_back(std::move(v))); + } + + // `InlinedVector::pop_back()` + // + // Destroys the element at `back()`, reducing the size by `1`. + void pop_back() noexcept { + ABSL_HARDENING_ASSERT(!empty()); + + AllocatorTraits::destroy(storage_.GetAllocator(), data() + (size() - 1)); + storage_.SubtractSize(1); + } + + // `InlinedVector::erase(...)` + // + // Erases the element at `pos`, returning an `iterator` pointing to where the + // erased element was located. + // + // NOTE: may return `end()`, which is not dereferencable. + iterator erase(const_iterator pos) { + ABSL_HARDENING_ASSERT(pos >= begin()); + ABSL_HARDENING_ASSERT(pos < end()); + + return storage_.Erase(pos, pos + 1); + } + + // Overload of `InlinedVector::erase(...)` that erases every element in the + // range [`from`, `to`), returning an `iterator` pointing to where the first + // erased element was located. + // + // NOTE: may return `end()`, which is not dereferencable. + iterator erase(const_iterator from, const_iterator to) { + ABSL_HARDENING_ASSERT(from >= begin()); + ABSL_HARDENING_ASSERT(from <= to); + ABSL_HARDENING_ASSERT(to <= end()); + + if (ABSL_PREDICT_TRUE(from != to)) { + return storage_.Erase(from, to); + } else { + return const_cast(from); + } + } + + // `InlinedVector::clear()` + // + // Destroys all elements in the inlined vector, setting the size to `0` and + // deallocating any held memory. + void clear() noexcept { + inlined_vector_internal::DestroyAdapter::DestroyElements( + storage_.GetAllocator(), data(), size()); + storage_.DeallocateIfAllocated(); + + storage_.SetInlinedSize(0); + } + + // `InlinedVector::reserve(...)` + // + // Ensures that there is enough room for at least `n` elements. + void reserve(size_type n) { storage_.Reserve(n); } + + // `InlinedVector::shrink_to_fit()` + // + // Attempts to reduce memory usage by moving elements to (or keeping elements + // in) the smallest available buffer sufficient for containing `size()` + // elements. + // + // If `size()` is sufficiently small, the elements will be moved into (or kept + // in) the inlined space. + void shrink_to_fit() { + if (storage_.GetIsAllocated()) { + storage_.ShrinkToFit(); + } + } + + // `InlinedVector::swap(...)` + // + // Swaps the contents of the inlined vector with `other`. + void swap(InlinedVector& other) { + if (ABSL_PREDICT_TRUE(this != std::addressof(other))) { + storage_.Swap(std::addressof(other.storage_)); + } + } + + private: + template + friend H AbslHashValue(H h, const absl::InlinedVector& a); + + Storage storage_; +}; + +// ----------------------------------------------------------------------------- +// InlinedVector Non-Member Functions +// ----------------------------------------------------------------------------- + +// `swap(...)` +// +// Swaps the contents of two inlined vectors. +template +void swap(absl::InlinedVector& a, + absl::InlinedVector& b) noexcept(noexcept(a.swap(b))) { + a.swap(b); +} + +// `operator==(...)` +// +// Tests for value-equality of two inlined vectors. +template +bool operator==(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + auto a_data = a.data(); + auto b_data = b.data(); + return absl::equal(a_data, a_data + a.size(), b_data, b_data + b.size()); +} + +// `operator!=(...)` +// +// Tests for value-inequality of two inlined vectors. +template +bool operator!=(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + return !(a == b); +} + +// `operator<(...)` +// +// Tests whether the value of an inlined vector is less than the value of +// another inlined vector using a lexicographical comparison algorithm. +template +bool operator<(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + auto a_data = a.data(); + auto b_data = b.data(); + return std::lexicographical_compare(a_data, a_data + a.size(), b_data, + b_data + b.size()); +} + +// `operator>(...)` +// +// Tests whether the value of an inlined vector is greater than the value of +// another inlined vector using a lexicographical comparison algorithm. +template +bool operator>(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + return b < a; +} + +// `operator<=(...)` +// +// Tests whether the value of an inlined vector is less than or equal to the +// value of another inlined vector using a lexicographical comparison algorithm. +template +bool operator<=(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + return !(b < a); +} + +// `operator>=(...)` +// +// Tests whether the value of an inlined vector is greater than or equal to the +// value of another inlined vector using a lexicographical comparison algorithm. +template +bool operator>=(const absl::InlinedVector& a, + const absl::InlinedVector& b) { + return !(a < b); +} + +// `AbslHashValue(...)` +// +// Provides `absl::Hash` support for `absl::InlinedVector`. It is uncommon to +// call this directly. +template +H AbslHashValue(H h, const absl::InlinedVector& a) { + auto size = a.size(); + return H::combine(H::combine_contiguous(std::move(h), a.data(), size), size); +} + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INLINED_VECTOR_H_ diff --git a/src/absl/container/internal/btree.h b/src/absl/container/internal/btree.h new file mode 100644 index 0000000..01f4e74 --- /dev/null +++ b/src/absl/container/internal/btree.h @@ -0,0 +1,2854 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// A btree implementation of the STL set and map interfaces. A btree is smaller +// and generally also faster than STL set/map (refer to the benchmarks below). +// The red-black tree implementation of STL set/map has an overhead of 3 +// pointers (left, right and parent) plus the node color information for each +// stored value. So a set consumes 40 bytes for each value stored in +// 64-bit mode. This btree implementation stores multiple values on fixed +// size nodes (usually 256 bytes) and doesn't store child pointers for leaf +// nodes. The result is that a btree_set may use much less memory per +// stored value. For the random insertion benchmark in btree_bench.cc, a +// btree_set with node-size of 256 uses 5.1 bytes per stored value. +// +// The packing of multiple values on to each node of a btree has another effect +// besides better space utilization: better cache locality due to fewer cache +// lines being accessed. Better cache locality translates into faster +// operations. +// +// CAVEATS +// +// Insertions and deletions on a btree can cause splitting, merging or +// rebalancing of btree nodes. And even without these operations, insertions +// and deletions on a btree will move values around within a node. In both +// cases, the result is that insertions and deletions can invalidate iterators +// pointing to values other than the one being inserted/deleted. Therefore, this +// container does not provide pointer stability. This is notably different from +// STL set/map which takes care to not invalidate iterators on insert/erase +// except, of course, for iterators pointing to the value being erased. A +// partial workaround when erasing is available: erase() returns an iterator +// pointing to the item just after the one that was erased (or end() if none +// exists). + +#ifndef ABSL_CONTAINER_INTERNAL_BTREE_H_ +#define ABSL_CONTAINER_INTERNAL_BTREE_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/internal/raw_logging.h" +#include "absl/base/macros.h" +#include "absl/container/internal/common.h" +#include "absl/container/internal/compressed_tuple.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/layout.h" +#include "absl/memory/memory.h" +#include "absl/meta/type_traits.h" +#include "absl/strings/cord.h" +#include "absl/strings/string_view.h" +#include "absl/types/compare.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +#ifdef ABSL_BTREE_ENABLE_GENERATIONS +#error ABSL_BTREE_ENABLE_GENERATIONS cannot be directly set +#elif defined(ABSL_HAVE_ADDRESS_SANITIZER) || \ + defined(ABSL_HAVE_MEMORY_SANITIZER) +// When compiled in sanitizer mode, we add generation integers to the nodes and +// iterators. When iterators are used, we validate that the container has not +// been mutated since the iterator was constructed. +#define ABSL_BTREE_ENABLE_GENERATIONS +#endif + +template +using compare_result_t = absl::result_of_t; + +// A helper class that indicates if the Compare parameter is a key-compare-to +// comparator. +template +using btree_is_key_compare_to = + std::is_convertible, absl::weak_ordering>; + +struct StringBtreeDefaultLess { + using is_transparent = void; + + StringBtreeDefaultLess() = default; + + // Compatibility constructor. + StringBtreeDefaultLess(std::less) {} // NOLINT + StringBtreeDefaultLess(std::less) {} // NOLINT + + // Allow converting to std::less for use in key_comp()/value_comp(). + explicit operator std::less() const { return {}; } + explicit operator std::less() const { return {}; } + explicit operator std::less() const { return {}; } + + absl::weak_ordering operator()(absl::string_view lhs, + absl::string_view rhs) const { + return compare_internal::compare_result_as_ordering(lhs.compare(rhs)); + } + StringBtreeDefaultLess(std::less) {} // NOLINT + absl::weak_ordering operator()(const absl::Cord &lhs, + const absl::Cord &rhs) const { + return compare_internal::compare_result_as_ordering(lhs.Compare(rhs)); + } + absl::weak_ordering operator()(const absl::Cord &lhs, + absl::string_view rhs) const { + return compare_internal::compare_result_as_ordering(lhs.Compare(rhs)); + } + absl::weak_ordering operator()(absl::string_view lhs, + const absl::Cord &rhs) const { + return compare_internal::compare_result_as_ordering(-rhs.Compare(lhs)); + } +}; + +struct StringBtreeDefaultGreater { + using is_transparent = void; + + StringBtreeDefaultGreater() = default; + + StringBtreeDefaultGreater(std::greater) {} // NOLINT + StringBtreeDefaultGreater(std::greater) {} // NOLINT + + // Allow converting to std::greater for use in key_comp()/value_comp(). + explicit operator std::greater() const { return {}; } + explicit operator std::greater() const { return {}; } + explicit operator std::greater() const { return {}; } + + absl::weak_ordering operator()(absl::string_view lhs, + absl::string_view rhs) const { + return compare_internal::compare_result_as_ordering(rhs.compare(lhs)); + } + StringBtreeDefaultGreater(std::greater) {} // NOLINT + absl::weak_ordering operator()(const absl::Cord &lhs, + const absl::Cord &rhs) const { + return compare_internal::compare_result_as_ordering(rhs.Compare(lhs)); + } + absl::weak_ordering operator()(const absl::Cord &lhs, + absl::string_view rhs) const { + return compare_internal::compare_result_as_ordering(-lhs.Compare(rhs)); + } + absl::weak_ordering operator()(absl::string_view lhs, + const absl::Cord &rhs) const { + return compare_internal::compare_result_as_ordering(rhs.Compare(lhs)); + } +}; + +// See below comments for checked_compare. +template ::value> +struct checked_compare_base : Compare { + using Compare::Compare; + explicit checked_compare_base(Compare c) : Compare(std::move(c)) {} + const Compare &comp() const { return *this; } +}; +template +struct checked_compare_base { + explicit checked_compare_base(Compare c) : compare(std::move(c)) {} + const Compare &comp() const { return compare; } + Compare compare; +}; + +// A mechanism for opting out of checked_compare for use only in btree_test.cc. +struct BtreeTestOnlyCheckedCompareOptOutBase {}; + +// A helper class to adapt the specified comparator for two use cases: +// (1) When using common Abseil string types with common comparison functors, +// convert a boolean comparison into a three-way comparison that returns an +// `absl::weak_ordering`. This helper class is specialized for +// less, greater, less, +// greater, less, and greater. +// (2) Adapt the comparator to diagnose cases of non-strict-weak-ordering (see +// https://en.cppreference.com/w/cpp/named_req/Compare) in debug mode. Whenever +// a comparison is made, we will make assertions to verify that the comparator +// is valid. +template +struct key_compare_adapter { + // Inherit from checked_compare_base to support function pointers and also + // keep empty-base-optimization (EBO) support for classes. + // Note: we can't use CompressedTuple here because that would interfere + // with the EBO for `btree::rightmost_`. `btree::rightmost_` is itself a + // CompressedTuple and nested `CompressedTuple`s don't support EBO. + // TODO(b/214288561): use CompressedTuple instead once it supports EBO for + // nested `CompressedTuple`s. + struct checked_compare : checked_compare_base { + private: + using Base = typename checked_compare::checked_compare_base; + using Base::comp; + + // If possible, returns whether `t` is equivalent to itself. We can only do + // this for `Key`s because we can't be sure that it's safe to call + // `comp()(k, k)` otherwise. Even if SFINAE allows it, there could be a + // compilation failure inside the implementation of the comparison operator. + bool is_self_equivalent(const Key &k) const { + // Note: this works for both boolean and three-way comparators. + return comp()(k, k) == 0; + } + // If we can't compare `t` with itself, returns true unconditionally. + template + bool is_self_equivalent(const T &) const { + return true; + } + + public: + using Base::Base; + checked_compare(Compare comp) : Base(std::move(comp)) {} // NOLINT + + // Allow converting to Compare for use in key_comp()/value_comp(). + explicit operator Compare() const { return comp(); } + + template >::value, + int> = 0> + bool operator()(const T &lhs, const U &rhs) const { + // NOTE: if any of these assertions fail, then the comparator does not + // establish a strict-weak-ordering (see + // https://en.cppreference.com/w/cpp/named_req/Compare). + assert(is_self_equivalent(lhs)); + assert(is_self_equivalent(rhs)); + const bool lhs_comp_rhs = comp()(lhs, rhs); + assert(!lhs_comp_rhs || !comp()(rhs, lhs)); + return lhs_comp_rhs; + } + + template < + typename T, typename U, + absl::enable_if_t, + absl::weak_ordering>::value, + int> = 0> + absl::weak_ordering operator()(const T &lhs, const U &rhs) const { + // NOTE: if any of these assertions fail, then the comparator does not + // establish a strict-weak-ordering (see + // https://en.cppreference.com/w/cpp/named_req/Compare). + assert(is_self_equivalent(lhs)); + assert(is_self_equivalent(rhs)); + const absl::weak_ordering lhs_comp_rhs = comp()(lhs, rhs); +#ifndef NDEBUG + const absl::weak_ordering rhs_comp_lhs = comp()(rhs, lhs); + if (lhs_comp_rhs > 0) { + assert(rhs_comp_lhs < 0 && "lhs_comp_rhs > 0 -> rhs_comp_lhs < 0"); + } else if (lhs_comp_rhs == 0) { + assert(rhs_comp_lhs == 0 && "lhs_comp_rhs == 0 -> rhs_comp_lhs == 0"); + } else { + assert(rhs_comp_lhs > 0 && "lhs_comp_rhs < 0 -> rhs_comp_lhs > 0"); + } +#endif + return lhs_comp_rhs; + } + }; + using type = absl::conditional_t< + std::is_base_of::value, + Compare, checked_compare>; +}; + +template <> +struct key_compare_adapter, std::string> { + using type = StringBtreeDefaultLess; +}; + +template <> +struct key_compare_adapter, std::string> { + using type = StringBtreeDefaultGreater; +}; + +template <> +struct key_compare_adapter, absl::string_view> { + using type = StringBtreeDefaultLess; +}; + +template <> +struct key_compare_adapter, absl::string_view> { + using type = StringBtreeDefaultGreater; +}; + +template <> +struct key_compare_adapter, absl::Cord> { + using type = StringBtreeDefaultLess; +}; + +template <> +struct key_compare_adapter, absl::Cord> { + using type = StringBtreeDefaultGreater; +}; + +// Detects an 'absl_btree_prefer_linear_node_search' member. This is +// a protocol used as an opt-in or opt-out of linear search. +// +// For example, this would be useful for key types that wrap an integer +// and define their own cheap operator<(). For example: +// +// class K { +// public: +// using absl_btree_prefer_linear_node_search = std::true_type; +// ... +// private: +// friend bool operator<(K a, K b) { return a.k_ < b.k_; } +// int k_; +// }; +// +// btree_map m; // Uses linear search +// +// If T has the preference tag, then it has a preference. +// Btree will use the tag's truth value. +template +struct has_linear_node_search_preference : std::false_type {}; +template +struct prefers_linear_node_search : std::false_type {}; +template +struct has_linear_node_search_preference< + T, absl::void_t> + : std::true_type {}; +template +struct prefers_linear_node_search< + T, absl::void_t> + : T::absl_btree_prefer_linear_node_search {}; + +template +constexpr bool compare_has_valid_result_type() { + using compare_result_type = compare_result_t; + return std::is_same::value || + std::is_convertible::value; +} + +template +class map_value_compare { + template + friend class btree; + + // Note: this `protected` is part of the API of std::map::value_compare. See + // https://en.cppreference.com/w/cpp/container/map/value_compare. + protected: + explicit map_value_compare(original_key_compare c) : comp(std::move(c)) {} + + original_key_compare comp; // NOLINT + + public: + auto operator()(const value_type &lhs, const value_type &rhs) const + -> decltype(comp(lhs.first, rhs.first)) { + return comp(lhs.first, rhs.first); + } +}; + +template +struct common_params { + using original_key_compare = Compare; + + // If Compare is a common comparator for a string-like type, then we adapt it + // to use heterogeneous lookup and to be a key-compare-to comparator. + // We also adapt the comparator to diagnose invalid comparators in debug mode. + // We disable this when `Compare` is invalid in a way that will cause + // adaptation to fail (having invalid return type) so that we can give a + // better compilation failure in static_assert_validation. If we don't do + // this, then there will be cascading compilation failures that are confusing + // for users. + using key_compare = + absl::conditional_t(), + Compare, + typename key_compare_adapter::type>; + + static constexpr bool kIsKeyCompareStringAdapted = + std::is_same::value || + std::is_same::value; + static constexpr bool kIsKeyCompareTransparent = + IsTransparent::value || + kIsKeyCompareStringAdapted; + static constexpr bool kEnableGenerations = +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + true; +#else + false; +#endif + + // A type which indicates if we have a key-compare-to functor or a plain old + // key-compare functor. + using is_key_compare_to = btree_is_key_compare_to; + + using allocator_type = Alloc; + using key_type = Key; + using size_type = size_t; + using difference_type = ptrdiff_t; + + using slot_policy = SlotPolicy; + using slot_type = typename slot_policy::slot_type; + using value_type = typename slot_policy::value_type; + using init_type = typename slot_policy::mutable_value_type; + using pointer = value_type *; + using const_pointer = const value_type *; + using reference = value_type &; + using const_reference = const value_type &; + + using value_compare = + absl::conditional_t, + original_key_compare>; + using is_map_container = std::integral_constant; + + // For the given lookup key type, returns whether we can have multiple + // equivalent keys in the btree. If this is a multi-container, then we can. + // Otherwise, we can have multiple equivalent keys only if all of the + // following conditions are met: + // - The comparator is transparent. + // - The lookup key type is not the same as key_type. + // - The comparator is not a StringBtreeDefault{Less,Greater} comparator + // that we know has the same equivalence classes for all lookup types. + template + constexpr static bool can_have_multiple_equivalent_keys() { + return IsMulti || (IsTransparent::value && + !std::is_same::value && + !kIsKeyCompareStringAdapted); + } + + enum { + kTargetNodeSize = TargetNodeSize, + + // Upper bound for the available space for slots. This is largest for leaf + // nodes, which have overhead of at least a pointer + 4 bytes (for storing + // 3 field_types and an enum). + kNodeSlotSpace = + TargetNodeSize - /*minimum overhead=*/(sizeof(void *) + 4), + }; + + // This is an integral type large enough to hold as many slots as will fit a + // node of TargetNodeSize bytes. + using node_count_type = + absl::conditional_t<(kNodeSlotSpace / sizeof(slot_type) > + (std::numeric_limits::max)()), + uint16_t, uint8_t>; // NOLINT + + // The following methods are necessary for passing this struct as PolicyTraits + // for node_handle and/or are used within btree. + static value_type &element(slot_type *slot) { + return slot_policy::element(slot); + } + static const value_type &element(const slot_type *slot) { + return slot_policy::element(slot); + } + template + static void construct(Alloc *alloc, slot_type *slot, Args &&... args) { + slot_policy::construct(alloc, slot, std::forward(args)...); + } + static void construct(Alloc *alloc, slot_type *slot, slot_type *other) { + slot_policy::construct(alloc, slot, other); + } + static void destroy(Alloc *alloc, slot_type *slot) { + slot_policy::destroy(alloc, slot); + } + static void transfer(Alloc *alloc, slot_type *new_slot, slot_type *old_slot) { + slot_policy::transfer(alloc, new_slot, old_slot); + } +}; + +// An adapter class that converts a lower-bound compare into an upper-bound +// compare. Note: there is no need to make a version of this adapter specialized +// for key-compare-to functors because the upper-bound (the first value greater +// than the input) is never an exact match. +template +struct upper_bound_adapter { + explicit upper_bound_adapter(const Compare &c) : comp(c) {} + template + bool operator()(const K1 &a, const K2 &b) const { + // Returns true when a is not greater than b. + return !compare_internal::compare_result_as_less_than(comp(b, a)); + } + + private: + Compare comp; +}; + +enum class MatchKind : uint8_t { kEq, kNe }; + +template +struct SearchResult { + V value; + MatchKind match; + + static constexpr bool HasMatch() { return true; } + bool IsEq() const { return match == MatchKind::kEq; } +}; + +// When we don't use CompareTo, `match` is not present. +// This ensures that callers can't use it accidentally when it provides no +// useful information. +template +struct SearchResult { + SearchResult() {} + explicit SearchResult(V v) : value(v) {} + SearchResult(V v, MatchKind /*match*/) : value(v) {} + + V value; + + static constexpr bool HasMatch() { return false; } + static constexpr bool IsEq() { return false; } +}; + +// A node in the btree holding. The same node type is used for both internal +// and leaf nodes in the btree, though the nodes are allocated in such a way +// that the children array is only valid in internal nodes. +template +class btree_node { + using is_key_compare_to = typename Params::is_key_compare_to; + using field_type = typename Params::node_count_type; + using allocator_type = typename Params::allocator_type; + using slot_type = typename Params::slot_type; + using original_key_compare = typename Params::original_key_compare; + + public: + using params_type = Params; + using key_type = typename Params::key_type; + using value_type = typename Params::value_type; + using pointer = typename Params::pointer; + using const_pointer = typename Params::const_pointer; + using reference = typename Params::reference; + using const_reference = typename Params::const_reference; + using key_compare = typename Params::key_compare; + using size_type = typename Params::size_type; + using difference_type = typename Params::difference_type; + + // Btree decides whether to use linear node search as follows: + // - If the comparator expresses a preference, use that. + // - If the key expresses a preference, use that. + // - If the key is arithmetic and the comparator is std::less or + // std::greater, choose linear. + // - Otherwise, choose binary. + // TODO(ezb): Might make sense to add condition(s) based on node-size. + using use_linear_search = std::integral_constant< + bool, has_linear_node_search_preference::value + ? prefers_linear_node_search::value + : has_linear_node_search_preference::value + ? prefers_linear_node_search::value + : std::is_arithmetic::value && + (std::is_same, + original_key_compare>::value || + std::is_same, + original_key_compare>::value)>; + + // This class is organized by absl::container_internal::Layout as if it had + // the following structure: + // // A pointer to the node's parent. + // btree_node *parent; + // + // // When ABSL_BTREE_ENABLE_GENERATIONS is defined, we also have a + // // generation integer in order to check that when iterators are + // // used, they haven't been invalidated already. Only the generation on + // // the root is used, but we have one on each node because whether a node + // // is root or not can change. + // uint32_t generation; + // + // // The position of the node in the node's parent. + // field_type position; + // // The index of the first populated value in `values`. + // // TODO(ezb): right now, `start` is always 0. Update insertion/merge + // // logic to allow for floating storage within nodes. + // field_type start; + // // The index after the last populated value in `values`. Currently, this + // // is the same as the count of values. + // field_type finish; + // // The maximum number of values the node can hold. This is an integer in + // // [1, kNodeSlots] for root leaf nodes, kNodeSlots for non-root leaf + // // nodes, and kInternalNodeMaxCount (as a sentinel value) for internal + // // nodes (even though there are still kNodeSlots values in the node). + // // TODO(ezb): make max_count use only 4 bits and record log2(capacity) + // // to free extra bits for is_root, etc. + // field_type max_count; + // + // // The array of values. The capacity is `max_count` for leaf nodes and + // // kNodeSlots for internal nodes. Only the values in + // // [start, finish) have been initialized and are valid. + // slot_type values[max_count]; + // + // // The array of child pointers. The keys in children[i] are all less + // // than key(i). The keys in children[i + 1] are all greater than key(i). + // // There are 0 children for leaf nodes and kNodeSlots + 1 children for + // // internal nodes. + // btree_node *children[kNodeSlots + 1]; + // + // This class is only constructed by EmptyNodeType. Normally, pointers to the + // layout above are allocated, cast to btree_node*, and de-allocated within + // the btree implementation. + ~btree_node() = default; + btree_node(btree_node const &) = delete; + btree_node &operator=(btree_node const &) = delete; + + // Public for EmptyNodeType. + constexpr static size_type Alignment() { + static_assert(LeafLayout(1).Alignment() == InternalLayout().Alignment(), + "Alignment of all nodes must be equal."); + return InternalLayout().Alignment(); + } + + protected: + btree_node() = default; + + private: + using layout_type = + absl::container_internal::Layout; + constexpr static size_type SizeWithNSlots(size_type n) { + return layout_type( + /*parent*/ 1, + /*generation*/ params_type::kEnableGenerations ? 1 : 0, + /*position, start, finish, max_count*/ 4, + /*slots*/ n, + /*children*/ 0) + .AllocSize(); + } + // A lower bound for the overhead of fields other than slots in a leaf node. + constexpr static size_type MinimumOverhead() { + return SizeWithNSlots(1) - sizeof(slot_type); + } + + // Compute how many values we can fit onto a leaf node taking into account + // padding. + constexpr static size_type NodeTargetSlots(const size_type begin, + const size_type end) { + return begin == end ? begin + : SizeWithNSlots((begin + end) / 2 + 1) > + params_type::kTargetNodeSize + ? NodeTargetSlots(begin, (begin + end) / 2) + : NodeTargetSlots((begin + end) / 2 + 1, end); + } + + enum { + kTargetNodeSize = params_type::kTargetNodeSize, + kNodeTargetSlots = NodeTargetSlots(0, params_type::kTargetNodeSize), + + // We need a minimum of 3 slots per internal node in order to perform + // splitting (1 value for the two nodes involved in the split and 1 value + // propagated to the parent as the delimiter for the split). For performance + // reasons, we don't allow 3 slots-per-node due to bad worst case occupancy + // of 1/3 (for a node, not a b-tree). + kMinNodeSlots = 4, + + kNodeSlots = + kNodeTargetSlots >= kMinNodeSlots ? kNodeTargetSlots : kMinNodeSlots, + + // The node is internal (i.e. is not a leaf node) if and only if `max_count` + // has this value. + kInternalNodeMaxCount = 0, + }; + + // Leaves can have less than kNodeSlots values. + constexpr static layout_type LeafLayout(const int slot_count = kNodeSlots) { + return layout_type( + /*parent*/ 1, + /*generation*/ params_type::kEnableGenerations ? 1 : 0, + /*position, start, finish, max_count*/ 4, + /*slots*/ slot_count, + /*children*/ 0); + } + constexpr static layout_type InternalLayout() { + return layout_type( + /*parent*/ 1, + /*generation*/ params_type::kEnableGenerations ? 1 : 0, + /*position, start, finish, max_count*/ 4, + /*slots*/ kNodeSlots, + /*children*/ kNodeSlots + 1); + } + constexpr static size_type LeafSize(const int slot_count = kNodeSlots) { + return LeafLayout(slot_count).AllocSize(); + } + constexpr static size_type InternalSize() { + return InternalLayout().AllocSize(); + } + + // N is the index of the type in the Layout definition. + // ElementType is the Nth type in the Layout definition. + template + inline typename layout_type::template ElementType *GetField() { + // We assert that we don't read from values that aren't there. + assert(N < 4 || is_internal()); + return InternalLayout().template Pointer(reinterpret_cast(this)); + } + template + inline const typename layout_type::template ElementType *GetField() const { + assert(N < 4 || is_internal()); + return InternalLayout().template Pointer( + reinterpret_cast(this)); + } + void set_parent(btree_node *p) { *GetField<0>() = p; } + field_type &mutable_finish() { return GetField<2>()[2]; } + slot_type *slot(int i) { return &GetField<3>()[i]; } + slot_type *start_slot() { return slot(start()); } + slot_type *finish_slot() { return slot(finish()); } + const slot_type *slot(int i) const { return &GetField<3>()[i]; } + void set_position(field_type v) { GetField<2>()[0] = v; } + void set_start(field_type v) { GetField<2>()[1] = v; } + void set_finish(field_type v) { GetField<2>()[2] = v; } + // This method is only called by the node init methods. + void set_max_count(field_type v) { GetField<2>()[3] = v; } + + public: + // Whether this is a leaf node or not. This value doesn't change after the + // node is created. + bool is_leaf() const { return GetField<2>()[3] != kInternalNodeMaxCount; } + // Whether this is an internal node or not. This value doesn't change after + // the node is created. + bool is_internal() const { return !is_leaf(); } + + // Getter for the position of this node in its parent. + field_type position() const { return GetField<2>()[0]; } + + // Getter for the offset of the first value in the `values` array. + field_type start() const { + // TODO(ezb): when floating storage is implemented, return GetField<2>()[1]; + assert(GetField<2>()[1] == 0); + return 0; + } + + // Getter for the offset after the last value in the `values` array. + field_type finish() const { return GetField<2>()[2]; } + + // Getters for the number of values stored in this node. + field_type count() const { + assert(finish() >= start()); + return finish() - start(); + } + field_type max_count() const { + // Internal nodes have max_count==kInternalNodeMaxCount. + // Leaf nodes have max_count in [1, kNodeSlots]. + const field_type max_count = GetField<2>()[3]; + return max_count == field_type{kInternalNodeMaxCount} + ? field_type{kNodeSlots} + : max_count; + } + + // Getter for the parent of this node. + btree_node *parent() const { return *GetField<0>(); } + // Getter for whether the node is the root of the tree. The parent of the + // root of the tree is the leftmost node in the tree which is guaranteed to + // be a leaf. + bool is_root() const { return parent()->is_leaf(); } + void make_root() { + assert(parent()->is_root()); + set_generation(parent()->generation()); + set_parent(parent()->parent()); + } + + // Gets the root node's generation integer, which is the one used by the tree. + uint32_t *get_root_generation() const { + assert(params_type::kEnableGenerations); + const btree_node *curr = this; + for (; !curr->is_root(); curr = curr->parent()) continue; + return const_cast(&curr->GetField<1>()[0]); + } + + // Returns the generation for iterator validation. + uint32_t generation() const { + return params_type::kEnableGenerations ? *get_root_generation() : 0; + } + // Updates generation. Should only be called on a root node or during node + // initialization. + void set_generation(uint32_t generation) { + if (params_type::kEnableGenerations) GetField<1>()[0] = generation; + } + // Updates the generation. We do this whenever the node is mutated. + void next_generation() { + if (params_type::kEnableGenerations) ++*get_root_generation(); + } + + // Getters for the key/value at position i in the node. + const key_type &key(int i) const { return params_type::key(slot(i)); } + reference value(int i) { return params_type::element(slot(i)); } + const_reference value(int i) const { return params_type::element(slot(i)); } + + // Getters/setter for the child at position i in the node. + btree_node *child(int i) const { return GetField<4>()[i]; } + btree_node *start_child() const { return child(start()); } + btree_node *&mutable_child(int i) { return GetField<4>()[i]; } + void clear_child(int i) { + absl::container_internal::SanitizerPoisonObject(&mutable_child(i)); + } + void set_child(int i, btree_node *c) { + absl::container_internal::SanitizerUnpoisonObject(&mutable_child(i)); + mutable_child(i) = c; + c->set_position(i); + } + void init_child(int i, btree_node *c) { + set_child(i, c); + c->set_parent(this); + } + + // Returns the position of the first value whose key is not less than k. + template + SearchResult lower_bound( + const K &k, const key_compare &comp) const { + return use_linear_search::value ? linear_search(k, comp) + : binary_search(k, comp); + } + // Returns the position of the first value whose key is greater than k. + template + int upper_bound(const K &k, const key_compare &comp) const { + auto upper_compare = upper_bound_adapter(comp); + return use_linear_search::value ? linear_search(k, upper_compare).value + : binary_search(k, upper_compare).value; + } + + template + SearchResult::value> + linear_search(const K &k, const Compare &comp) const { + return linear_search_impl(k, start(), finish(), comp, + btree_is_key_compare_to()); + } + + template + SearchResult::value> + binary_search(const K &k, const Compare &comp) const { + return binary_search_impl(k, start(), finish(), comp, + btree_is_key_compare_to()); + } + + // Returns the position of the first value whose key is not less than k using + // linear search performed using plain compare. + template + SearchResult linear_search_impl( + const K &k, int s, const int e, const Compare &comp, + std::false_type /* IsCompareTo */) const { + while (s < e) { + if (!comp(key(s), k)) { + break; + } + ++s; + } + return SearchResult{s}; + } + + // Returns the position of the first value whose key is not less than k using + // linear search performed using compare-to. + template + SearchResult linear_search_impl( + const K &k, int s, const int e, const Compare &comp, + std::true_type /* IsCompareTo */) const { + while (s < e) { + const absl::weak_ordering c = comp(key(s), k); + if (c == 0) { + return {s, MatchKind::kEq}; + } else if (c > 0) { + break; + } + ++s; + } + return {s, MatchKind::kNe}; + } + + // Returns the position of the first value whose key is not less than k using + // binary search performed using plain compare. + template + SearchResult binary_search_impl( + const K &k, int s, int e, const Compare &comp, + std::false_type /* IsCompareTo */) const { + while (s != e) { + const int mid = (s + e) >> 1; + if (comp(key(mid), k)) { + s = mid + 1; + } else { + e = mid; + } + } + return SearchResult{s}; + } + + // Returns the position of the first value whose key is not less than k using + // binary search performed using compare-to. + template + SearchResult binary_search_impl( + const K &k, int s, int e, const CompareTo &comp, + std::true_type /* IsCompareTo */) const { + if (params_type::template can_have_multiple_equivalent_keys()) { + MatchKind exact_match = MatchKind::kNe; + while (s != e) { + const int mid = (s + e) >> 1; + const absl::weak_ordering c = comp(key(mid), k); + if (c < 0) { + s = mid + 1; + } else { + e = mid; + if (c == 0) { + // Need to return the first value whose key is not less than k, + // which requires continuing the binary search if there could be + // multiple equivalent keys. + exact_match = MatchKind::kEq; + } + } + } + return {s, exact_match}; + } else { // Can't have multiple equivalent keys. + while (s != e) { + const int mid = (s + e) >> 1; + const absl::weak_ordering c = comp(key(mid), k); + if (c < 0) { + s = mid + 1; + } else if (c > 0) { + e = mid; + } else { + return {mid, MatchKind::kEq}; + } + } + return {s, MatchKind::kNe}; + } + } + + // Emplaces a value at position i, shifting all existing values and + // children at positions >= i to the right by 1. + template + void emplace_value(size_type i, allocator_type *alloc, Args &&... args); + + // Removes the values at positions [i, i + to_erase), shifting all existing + // values and children after that range to the left by to_erase. Clears all + // children between [i, i + to_erase). + void remove_values(field_type i, field_type to_erase, allocator_type *alloc); + + // Rebalances a node with its right sibling. + void rebalance_right_to_left(int to_move, btree_node *right, + allocator_type *alloc); + void rebalance_left_to_right(int to_move, btree_node *right, + allocator_type *alloc); + + // Splits a node, moving a portion of the node's values to its right sibling. + void split(int insert_position, btree_node *dest, allocator_type *alloc); + + // Merges a node with its right sibling, moving all of the values and the + // delimiting key in the parent node onto itself, and deleting the src node. + void merge(btree_node *src, allocator_type *alloc); + + // Node allocation/deletion routines. + void init_leaf(int max_count, btree_node *parent) { + set_generation(0); + set_parent(parent); + set_position(0); + set_start(0); + set_finish(0); + set_max_count(max_count); + absl::container_internal::SanitizerPoisonMemoryRegion( + start_slot(), max_count * sizeof(slot_type)); + } + void init_internal(btree_node *parent) { + init_leaf(kNodeSlots, parent); + // Set `max_count` to a sentinel value to indicate that this node is + // internal. + set_max_count(kInternalNodeMaxCount); + absl::container_internal::SanitizerPoisonMemoryRegion( + &mutable_child(start()), (kNodeSlots + 1) * sizeof(btree_node *)); + } + + static void deallocate(const size_type size, btree_node *node, + allocator_type *alloc) { + absl::container_internal::Deallocate(alloc, node, size); + } + + // Deletes a node and all of its children. + static void clear_and_delete(btree_node *node, allocator_type *alloc); + + private: + template + void value_init(const field_type i, allocator_type *alloc, Args &&... args) { + next_generation(); + absl::container_internal::SanitizerUnpoisonObject(slot(i)); + params_type::construct(alloc, slot(i), std::forward(args)...); + } + void value_destroy(const field_type i, allocator_type *alloc) { + next_generation(); + params_type::destroy(alloc, slot(i)); + absl::container_internal::SanitizerPoisonObject(slot(i)); + } + void value_destroy_n(const field_type i, const field_type n, + allocator_type *alloc) { + next_generation(); + for (slot_type *s = slot(i), *end = slot(i + n); s != end; ++s) { + params_type::destroy(alloc, s); + absl::container_internal::SanitizerPoisonObject(s); + } + } + + static void transfer(slot_type *dest, slot_type *src, allocator_type *alloc) { + absl::container_internal::SanitizerUnpoisonObject(dest); + params_type::transfer(alloc, dest, src); + absl::container_internal::SanitizerPoisonObject(src); + } + + // Transfers value from slot `src_i` in `src_node` to slot `dest_i` in `this`. + void transfer(const size_type dest_i, const size_type src_i, + btree_node *src_node, allocator_type *alloc) { + next_generation(); + transfer(slot(dest_i), src_node->slot(src_i), alloc); + } + + // Transfers `n` values starting at value `src_i` in `src_node` into the + // values starting at value `dest_i` in `this`. + void transfer_n(const size_type n, const size_type dest_i, + const size_type src_i, btree_node *src_node, + allocator_type *alloc) { + next_generation(); + for (slot_type *src = src_node->slot(src_i), *end = src + n, + *dest = slot(dest_i); + src != end; ++src, ++dest) { + transfer(dest, src, alloc); + } + } + + // Same as above, except that we start at the end and work our way to the + // beginning. + void transfer_n_backward(const size_type n, const size_type dest_i, + const size_type src_i, btree_node *src_node, + allocator_type *alloc) { + next_generation(); + for (slot_type *src = src_node->slot(src_i + n - 1), *end = src - n, + *dest = slot(dest_i + n - 1); + src != end; --src, --dest) { + transfer(dest, src, alloc); + } + } + + template + friend class btree; + template + friend class btree_iterator; + friend class BtreeNodePeer; + friend struct btree_access; +}; + +template +class btree_iterator { + using key_type = typename Node::key_type; + using size_type = typename Node::size_type; + using params_type = typename Node::params_type; + using is_map_container = typename params_type::is_map_container; + + using node_type = Node; + using normal_node = typename std::remove_const::type; + using const_node = const Node; + using normal_pointer = typename params_type::pointer; + using normal_reference = typename params_type::reference; + using const_pointer = typename params_type::const_pointer; + using const_reference = typename params_type::const_reference; + using slot_type = typename params_type::slot_type; + + using iterator = + btree_iterator; + using const_iterator = + btree_iterator; + + public: + // These aliases are public for std::iterator_traits. + using difference_type = typename Node::difference_type; + using value_type = typename params_type::value_type; + using pointer = Pointer; + using reference = Reference; + using iterator_category = std::bidirectional_iterator_tag; + + btree_iterator() : btree_iterator(nullptr, -1) {} + explicit btree_iterator(Node *n) : btree_iterator(n, n->start()) {} + btree_iterator(Node *n, int p) : node_(n), position_(p) { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + // Use `~uint32_t{}` as a sentinel value for iterator generations so it + // doesn't match the initial value for the actual generation. + generation_ = n != nullptr ? n->generation() : ~uint32_t{}; +#endif + } + + // NOTE: this SFINAE allows for implicit conversions from iterator to + // const_iterator, but it specifically avoids hiding the copy constructor so + // that the trivial one will be used when possible. + template , iterator>::value && + std::is_same::value, + int> = 0> + btree_iterator(const btree_iterator other) // NOLINT + : node_(other.node_), position_(other.position_) { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + generation_ = other.generation_; +#endif + } + + bool operator==(const iterator &other) const { + return node_ == other.node_ && position_ == other.position_; + } + bool operator==(const const_iterator &other) const { + return node_ == other.node_ && position_ == other.position_; + } + bool operator!=(const iterator &other) const { + return node_ != other.node_ || position_ != other.position_; + } + bool operator!=(const const_iterator &other) const { + return node_ != other.node_ || position_ != other.position_; + } + + // Accessors for the key/value the iterator is pointing at. + reference operator*() const { + ABSL_HARDENING_ASSERT(node_ != nullptr); + ABSL_HARDENING_ASSERT(node_->start() <= position_); + ABSL_HARDENING_ASSERT(node_->finish() > position_); + assert_valid_generation(); + return node_->value(position_); + } + pointer operator->() const { return &operator*(); } + + btree_iterator &operator++() { + increment(); + return *this; + } + btree_iterator &operator--() { + decrement(); + return *this; + } + btree_iterator operator++(int) { + btree_iterator tmp = *this; + ++*this; + return tmp; + } + btree_iterator operator--(int) { + btree_iterator tmp = *this; + --*this; + return tmp; + } + + private: + friend iterator; + friend const_iterator; + template + friend class btree; + template + friend class btree_container; + template + friend class btree_set_container; + template + friend class btree_map_container; + template + friend class btree_multiset_container; + template + friend class base_checker; + friend struct btree_access; + + // This SFINAE allows explicit conversions from const_iterator to + // iterator, but also avoids hiding the copy constructor. + // NOTE: the const_cast is safe because this constructor is only called by + // non-const methods and the container owns the nodes. + template , const_iterator>::value && + std::is_same::value, + int> = 0> + explicit btree_iterator(const btree_iterator other) + : node_(const_cast(other.node_)), + position_(other.position_) { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + generation_ = other.generation_; +#endif + } + + // Increment/decrement the iterator. + void increment() { + assert_valid_generation(); + if (node_->is_leaf() && ++position_ < node_->finish()) { + return; + } + increment_slow(); + } + void increment_slow(); + + void decrement() { + assert_valid_generation(); + if (node_->is_leaf() && --position_ >= node_->start()) { + return; + } + decrement_slow(); + } + void decrement_slow(); + + // Updates the generation. For use internally right before we return an + // iterator to the user. + void update_generation() { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + if (node_ != nullptr) generation_ = node_->generation(); +#endif + } + + const key_type &key() const { return node_->key(position_); } + decltype(std::declval()->slot(0)) slot() { + return node_->slot(position_); + } + + void assert_valid_generation() const { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + if (node_ != nullptr && node_->generation() != generation_) { + ABSL_INTERNAL_LOG( + FATAL, + "Attempting to use an invalidated iterator. The corresponding b-tree " + "container has been mutated since this iterator was constructed."); + } +#endif + } + + // The node in the tree the iterator is pointing at. + Node *node_; + // The position within the node of the tree the iterator is pointing at. + // NOTE: this is an int rather than a field_type because iterators can point + // to invalid positions (such as -1) in certain circumstances. + int position_; +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + // Used to check that the iterator hasn't been invalidated. + uint32_t generation_; +#endif +}; + +template +class btree { + using node_type = btree_node; + using is_key_compare_to = typename Params::is_key_compare_to; + using field_type = typename node_type::field_type; + + // We use a static empty node for the root/leftmost/rightmost of empty btrees + // in order to avoid branching in begin()/end(). + struct alignas(node_type::Alignment()) EmptyNodeType : node_type { + using field_type = typename node_type::field_type; + node_type *parent; +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + uint32_t generation = 0; +#endif + field_type position = 0; + field_type start = 0; + field_type finish = 0; + // max_count must be != kInternalNodeMaxCount (so that this node is regarded + // as a leaf node). max_count() is never called when the tree is empty. + field_type max_count = node_type::kInternalNodeMaxCount + 1; + +#ifdef _MSC_VER + // MSVC has constexpr code generations bugs here. + EmptyNodeType() : parent(this) {} +#else + constexpr EmptyNodeType(node_type *p) : parent(p) {} +#endif + }; + + static node_type *EmptyNode() { +#ifdef _MSC_VER + static EmptyNodeType *empty_node = new EmptyNodeType; + // This assert fails on some other construction methods. + assert(empty_node->parent == empty_node); + return empty_node; +#else + static constexpr EmptyNodeType empty_node( + const_cast(&empty_node)); + return const_cast(&empty_node); +#endif + } + + enum : uint32_t { + kNodeSlots = node_type::kNodeSlots, + kMinNodeValues = kNodeSlots / 2, + }; + + struct node_stats { + using size_type = typename Params::size_type; + + node_stats(size_type l, size_type i) : leaf_nodes(l), internal_nodes(i) {} + + node_stats &operator+=(const node_stats &other) { + leaf_nodes += other.leaf_nodes; + internal_nodes += other.internal_nodes; + return *this; + } + + size_type leaf_nodes; + size_type internal_nodes; + }; + + public: + using key_type = typename Params::key_type; + using value_type = typename Params::value_type; + using size_type = typename Params::size_type; + using difference_type = typename Params::difference_type; + using key_compare = typename Params::key_compare; + using original_key_compare = typename Params::original_key_compare; + using value_compare = typename Params::value_compare; + using allocator_type = typename Params::allocator_type; + using reference = typename Params::reference; + using const_reference = typename Params::const_reference; + using pointer = typename Params::pointer; + using const_pointer = typename Params::const_pointer; + using iterator = + typename btree_iterator::iterator; + using const_iterator = typename iterator::const_iterator; + using reverse_iterator = std::reverse_iterator; + using const_reverse_iterator = std::reverse_iterator; + using node_handle_type = node_handle; + + // Internal types made public for use by btree_container types. + using params_type = Params; + using slot_type = typename Params::slot_type; + + private: + // Copies or moves (depending on the template parameter) the values in + // other into this btree in their order in other. This btree must be empty + // before this method is called. This method is used in copy construction, + // copy assignment, and move assignment. + template + void copy_or_move_values_in_order(Btree &other); + + // Validates that various assumptions/requirements are true at compile time. + constexpr static bool static_assert_validation(); + + public: + btree(const key_compare &comp, const allocator_type &alloc) + : root_(EmptyNode()), rightmost_(comp, alloc, EmptyNode()), size_(0) {} + + btree(const btree &other) : btree(other, other.allocator()) {} + btree(const btree &other, const allocator_type &alloc) + : btree(other.key_comp(), alloc) { + copy_or_move_values_in_order(other); + } + btree(btree &&other) noexcept + : root_(absl::exchange(other.root_, EmptyNode())), + rightmost_(std::move(other.rightmost_)), + size_(absl::exchange(other.size_, 0)) { + other.mutable_rightmost() = EmptyNode(); + } + btree(btree &&other, const allocator_type &alloc) + : btree(other.key_comp(), alloc) { + if (alloc == other.allocator()) { + swap(other); + } else { + // Move values from `other` one at a time when allocators are different. + copy_or_move_values_in_order(other); + } + } + + ~btree() { + // Put static_asserts in destructor to avoid triggering them before the type + // is complete. + static_assert(static_assert_validation(), "This call must be elided."); + clear(); + } + + // Assign the contents of other to *this. + btree &operator=(const btree &other); + btree &operator=(btree &&other) noexcept; + + iterator begin() { return iterator(leftmost()); } + const_iterator begin() const { return const_iterator(leftmost()); } + iterator end() { return iterator(rightmost(), rightmost()->finish()); } + const_iterator end() const { + return const_iterator(rightmost(), rightmost()->finish()); + } + reverse_iterator rbegin() { return reverse_iterator(end()); } + const_reverse_iterator rbegin() const { + return const_reverse_iterator(end()); + } + reverse_iterator rend() { return reverse_iterator(begin()); } + const_reverse_iterator rend() const { + return const_reverse_iterator(begin()); + } + + // Finds the first element whose key is not less than `key`. + template + iterator lower_bound(const K &key) { + return internal_end(internal_lower_bound(key).value); + } + template + const_iterator lower_bound(const K &key) const { + return internal_end(internal_lower_bound(key).value); + } + + // Finds the first element whose key is not less than `key` and also returns + // whether that element is equal to `key`. + template + std::pair lower_bound_equal(const K &key) const; + + // Finds the first element whose key is greater than `key`. + template + iterator upper_bound(const K &key) { + return internal_end(internal_upper_bound(key)); + } + template + const_iterator upper_bound(const K &key) const { + return internal_end(internal_upper_bound(key)); + } + + // Finds the range of values which compare equal to key. The first member of + // the returned pair is equal to lower_bound(key). The second member of the + // pair is equal to upper_bound(key). + template + std::pair equal_range(const K &key); + template + std::pair equal_range(const K &key) const { + return const_cast(this)->equal_range(key); + } + + // Inserts a value into the btree only if it does not already exist. The + // boolean return value indicates whether insertion succeeded or failed. + // Requirement: if `key` already exists in the btree, does not consume `args`. + // Requirement: `key` is never referenced after consuming `args`. + template + std::pair insert_unique(const K &key, Args &&... args); + + // Inserts with hint. Checks to see if the value should be placed immediately + // before `position` in the tree. If so, then the insertion will take + // amortized constant time. If not, the insertion will take amortized + // logarithmic time as if a call to insert_unique() were made. + // Requirement: if `key` already exists in the btree, does not consume `args`. + // Requirement: `key` is never referenced after consuming `args`. + template + std::pair insert_hint_unique(iterator position, + const K &key, + Args &&... args); + + // Insert a range of values into the btree. + // Note: the first overload avoids constructing a value_type if the key + // already exists in the btree. + template ()( + params_type::key(*std::declval()), + std::declval()))> + void insert_iterator_unique(InputIterator b, InputIterator e, int); + // We need the second overload for cases in which we need to construct a + // value_type in order to compare it with the keys already in the btree. + template + void insert_iterator_unique(InputIterator b, InputIterator e, char); + + // Inserts a value into the btree. + template + iterator insert_multi(const key_type &key, ValueType &&v); + + // Inserts a value into the btree. + template + iterator insert_multi(ValueType &&v) { + return insert_multi(params_type::key(v), std::forward(v)); + } + + // Insert with hint. Check to see if the value should be placed immediately + // before position in the tree. If it does, then the insertion will take + // amortized constant time. If not, the insertion will take amortized + // logarithmic time as if a call to insert_multi(v) were made. + template + iterator insert_hint_multi(iterator position, ValueType &&v); + + // Insert a range of values into the btree. + template + void insert_iterator_multi(InputIterator b, InputIterator e); + + // Erase the specified iterator from the btree. The iterator must be valid + // (i.e. not equal to end()). Return an iterator pointing to the node after + // the one that was erased (or end() if none exists). + // Requirement: does not read the value at `*iter`. + iterator erase(iterator iter); + + // Erases range. Returns the number of keys erased and an iterator pointing + // to the element after the last erased element. + std::pair erase_range(iterator begin, iterator end); + + // Finds an element with key equivalent to `key` or returns `end()` if `key` + // is not present. + template + iterator find(const K &key) { + return internal_end(internal_find(key)); + } + template + const_iterator find(const K &key) const { + return internal_end(internal_find(key)); + } + + // Clear the btree, deleting all of the values it contains. + void clear(); + + // Swaps the contents of `this` and `other`. + void swap(btree &other); + + const key_compare &key_comp() const noexcept { + return rightmost_.template get<0>(); + } + template + bool compare_keys(const K1 &a, const K2 &b) const { + return compare_internal::compare_result_as_less_than(key_comp()(a, b)); + } + + value_compare value_comp() const { + return value_compare(original_key_compare(key_comp())); + } + + // Verifies the structure of the btree. + void verify() const; + + // Size routines. + size_type size() const { return size_; } + size_type max_size() const { return (std::numeric_limits::max)(); } + bool empty() const { return size_ == 0; } + + // The height of the btree. An empty tree will have height 0. + size_type height() const { + size_type h = 0; + if (!empty()) { + // Count the length of the chain from the leftmost node up to the + // root. We actually count from the root back around to the level below + // the root, but the calculation is the same because of the circularity + // of that traversal. + const node_type *n = root(); + do { + ++h; + n = n->parent(); + } while (n != root()); + } + return h; + } + + // The number of internal, leaf and total nodes used by the btree. + size_type leaf_nodes() const { return internal_stats(root()).leaf_nodes; } + size_type internal_nodes() const { + return internal_stats(root()).internal_nodes; + } + size_type nodes() const { + node_stats stats = internal_stats(root()); + return stats.leaf_nodes + stats.internal_nodes; + } + + // The total number of bytes used by the btree. + // TODO(b/169338300): update to support node_btree_*. + size_type bytes_used() const { + node_stats stats = internal_stats(root()); + if (stats.leaf_nodes == 1 && stats.internal_nodes == 0) { + return sizeof(*this) + node_type::LeafSize(root()->max_count()); + } else { + return sizeof(*this) + stats.leaf_nodes * node_type::LeafSize() + + stats.internal_nodes * node_type::InternalSize(); + } + } + + // The average number of bytes used per value stored in the btree assuming + // random insertion order. + static double average_bytes_per_value() { + // The expected number of values per node with random insertion order is the + // average of the maximum and minimum numbers of values per node. + const double expected_values_per_node = + (kNodeSlots + kMinNodeValues) / 2.0; + return node_type::LeafSize() / expected_values_per_node; + } + + // The fullness of the btree. Computed as the number of elements in the btree + // divided by the maximum number of elements a tree with the current number + // of nodes could hold. A value of 1 indicates perfect space + // utilization. Smaller values indicate space wastage. + // Returns 0 for empty trees. + double fullness() const { + if (empty()) return 0.0; + return static_cast(size()) / (nodes() * kNodeSlots); + } + // The overhead of the btree structure in bytes per node. Computed as the + // total number of bytes used by the btree minus the number of bytes used for + // storing elements divided by the number of elements. + // Returns 0 for empty trees. + double overhead() const { + if (empty()) return 0.0; + return (bytes_used() - size() * sizeof(value_type)) / + static_cast(size()); + } + + // The allocator used by the btree. + allocator_type get_allocator() const { return allocator(); } + + private: + friend struct btree_access; + + // Internal accessor routines. + node_type *root() { return root_; } + const node_type *root() const { return root_; } + node_type *&mutable_root() noexcept { return root_; } + node_type *rightmost() { return rightmost_.template get<2>(); } + const node_type *rightmost() const { return rightmost_.template get<2>(); } + node_type *&mutable_rightmost() noexcept { + return rightmost_.template get<2>(); + } + key_compare *mutable_key_comp() noexcept { + return &rightmost_.template get<0>(); + } + + // The leftmost node is stored as the parent of the root node. + node_type *leftmost() { return root()->parent(); } + const node_type *leftmost() const { return root()->parent(); } + + // Allocator routines. + allocator_type *mutable_allocator() noexcept { + return &rightmost_.template get<1>(); + } + const allocator_type &allocator() const noexcept { + return rightmost_.template get<1>(); + } + + // Allocates a correctly aligned node of at least size bytes using the + // allocator. + node_type *allocate(const size_type size) { + return reinterpret_cast( + absl::container_internal::Allocate( + mutable_allocator(), size)); + } + + // Node creation/deletion routines. + node_type *new_internal_node(node_type *parent) { + node_type *n = allocate(node_type::InternalSize()); + n->init_internal(parent); + return n; + } + node_type *new_leaf_node(node_type *parent) { + node_type *n = allocate(node_type::LeafSize()); + n->init_leaf(kNodeSlots, parent); + return n; + } + node_type *new_leaf_root_node(const int max_count) { + node_type *n = allocate(node_type::LeafSize(max_count)); + n->init_leaf(max_count, /*parent=*/n); + return n; + } + + // Deletion helper routines. + iterator rebalance_after_delete(iterator iter); + + // Rebalances or splits the node iter points to. + void rebalance_or_split(iterator *iter); + + // Merges the values of left, right and the delimiting key on their parent + // onto left, removing the delimiting key and deleting right. + void merge_nodes(node_type *left, node_type *right); + + // Tries to merge node with its left or right sibling, and failing that, + // rebalance with its left or right sibling. Returns true if a merge + // occurred, at which point it is no longer valid to access node. Returns + // false if no merging took place. + bool try_merge_or_rebalance(iterator *iter); + + // Tries to shrink the height of the tree by 1. + void try_shrink(); + + iterator internal_end(iterator iter) { + return iter.node_ != nullptr ? iter : end(); + } + const_iterator internal_end(const_iterator iter) const { + return iter.node_ != nullptr ? iter : end(); + } + + // Emplaces a value into the btree immediately before iter. Requires that + // key(v) <= iter.key() and (--iter).key() <= key(v). + template + iterator internal_emplace(iterator iter, Args &&... args); + + // Returns an iterator pointing to the first value >= the value "iter" is + // pointing at. Note that "iter" might be pointing to an invalid location such + // as iter.position_ == iter.node_->finish(). This routine simply moves iter + // up in the tree to a valid location. Requires: iter.node_ is non-null. + template + static IterType internal_last(IterType iter); + + // Returns an iterator pointing to the leaf position at which key would + // reside in the tree, unless there is an exact match - in which case, the + // result may not be on a leaf. When there's a three-way comparator, we can + // return whether there was an exact match. This allows the caller to avoid a + // subsequent comparison to determine if an exact match was made, which is + // important for keys with expensive comparison, such as strings. + template + SearchResult internal_locate( + const K &key) const; + + // Internal routine which implements lower_bound(). + template + SearchResult internal_lower_bound( + const K &key) const; + + // Internal routine which implements upper_bound(). + template + iterator internal_upper_bound(const K &key) const; + + // Internal routine which implements find(). + template + iterator internal_find(const K &key) const; + + // Verifies the tree structure of node. + int internal_verify(const node_type *node, const key_type *lo, + const key_type *hi) const; + + node_stats internal_stats(const node_type *node) const { + // The root can be a static empty node. + if (node == nullptr || (node == root() && empty())) { + return node_stats(0, 0); + } + if (node->is_leaf()) { + return node_stats(1, 0); + } + node_stats res(0, 1); + for (int i = node->start(); i <= node->finish(); ++i) { + res += internal_stats(node->child(i)); + } + return res; + } + + node_type *root_; + + // A pointer to the rightmost node. Note that the leftmost node is stored as + // the root's parent. We use compressed tuple in order to save space because + // key_compare and allocator_type are usually empty. + absl::container_internal::CompressedTuple + rightmost_; + + // Number of values. + size_type size_; +}; + +//// +// btree_node methods +template +template +inline void btree_node

::emplace_value(const size_type i, + allocator_type *alloc, + Args &&... args) { + assert(i >= start()); + assert(i <= finish()); + // Shift old values to create space for new value and then construct it in + // place. + if (i < finish()) { + transfer_n_backward(finish() - i, /*dest_i=*/i + 1, /*src_i=*/i, this, + alloc); + } + value_init(i, alloc, std::forward(args)...); + set_finish(finish() + 1); + + if (is_internal() && finish() > i + 1) { + for (field_type j = finish(); j > i + 1; --j) { + set_child(j, child(j - 1)); + } + clear_child(i + 1); + } +} + +template +inline void btree_node

::remove_values(const field_type i, + const field_type to_erase, + allocator_type *alloc) { + // Transfer values after the removed range into their new places. + value_destroy_n(i, to_erase, alloc); + const field_type orig_finish = finish(); + const field_type src_i = i + to_erase; + transfer_n(orig_finish - src_i, i, src_i, this, alloc); + + if (is_internal()) { + // Delete all children between begin and end. + for (int j = 0; j < to_erase; ++j) { + clear_and_delete(child(i + j + 1), alloc); + } + // Rotate children after end into new positions. + for (int j = i + to_erase + 1; j <= orig_finish; ++j) { + set_child(j - to_erase, child(j)); + clear_child(j); + } + } + set_finish(orig_finish - to_erase); +} + +template +void btree_node

::rebalance_right_to_left(const int to_move, + btree_node *right, + allocator_type *alloc) { + assert(parent() == right->parent()); + assert(position() + 1 == right->position()); + assert(right->count() >= count()); + assert(to_move >= 1); + assert(to_move <= right->count()); + + // 1) Move the delimiting value in the parent to the left node. + transfer(finish(), position(), parent(), alloc); + + // 2) Move the (to_move - 1) values from the right node to the left node. + transfer_n(to_move - 1, finish() + 1, right->start(), right, alloc); + + // 3) Move the new delimiting value to the parent from the right node. + parent()->transfer(position(), right->start() + to_move - 1, right, alloc); + + // 4) Shift the values in the right node to their correct positions. + right->transfer_n(right->count() - to_move, right->start(), + right->start() + to_move, right, alloc); + + if (is_internal()) { + // Move the child pointers from the right to the left node. + for (int i = 0; i < to_move; ++i) { + init_child(finish() + i + 1, right->child(i)); + } + for (int i = right->start(); i <= right->finish() - to_move; ++i) { + assert(i + to_move <= right->max_count()); + right->init_child(i, right->child(i + to_move)); + right->clear_child(i + to_move); + } + } + + // Fixup `finish` on the left and right nodes. + set_finish(finish() + to_move); + right->set_finish(right->finish() - to_move); +} + +template +void btree_node

::rebalance_left_to_right(const int to_move, + btree_node *right, + allocator_type *alloc) { + assert(parent() == right->parent()); + assert(position() + 1 == right->position()); + assert(count() >= right->count()); + assert(to_move >= 1); + assert(to_move <= count()); + + // Values in the right node are shifted to the right to make room for the + // new to_move values. Then, the delimiting value in the parent and the + // other (to_move - 1) values in the left node are moved into the right node. + // Lastly, a new delimiting value is moved from the left node into the + // parent, and the remaining empty left node entries are destroyed. + + // 1) Shift existing values in the right node to their correct positions. + right->transfer_n_backward(right->count(), right->start() + to_move, + right->start(), right, alloc); + + // 2) Move the delimiting value in the parent to the right node. + right->transfer(right->start() + to_move - 1, position(), parent(), alloc); + + // 3) Move the (to_move - 1) values from the left node to the right node. + right->transfer_n(to_move - 1, right->start(), finish() - (to_move - 1), this, + alloc); + + // 4) Move the new delimiting value to the parent from the left node. + parent()->transfer(position(), finish() - to_move, this, alloc); + + if (is_internal()) { + // Move the child pointers from the left to the right node. + for (int i = right->finish(); i >= right->start(); --i) { + right->init_child(i + to_move, right->child(i)); + right->clear_child(i); + } + for (int i = 1; i <= to_move; ++i) { + right->init_child(i - 1, child(finish() - to_move + i)); + clear_child(finish() - to_move + i); + } + } + + // Fixup the counts on the left and right nodes. + set_finish(finish() - to_move); + right->set_finish(right->finish() + to_move); +} + +template +void btree_node

::split(const int insert_position, btree_node *dest, + allocator_type *alloc) { + assert(dest->count() == 0); + assert(max_count() == kNodeSlots); + + // We bias the split based on the position being inserted. If we're + // inserting at the beginning of the left node then bias the split to put + // more values on the right node. If we're inserting at the end of the + // right node then bias the split to put more values on the left node. + if (insert_position == start()) { + dest->set_finish(dest->start() + finish() - 1); + } else if (insert_position == kNodeSlots) { + dest->set_finish(dest->start()); + } else { + dest->set_finish(dest->start() + count() / 2); + } + set_finish(finish() - dest->count()); + assert(count() >= 1); + + // Move values from the left sibling to the right sibling. + dest->transfer_n(dest->count(), dest->start(), finish(), this, alloc); + + // The split key is the largest value in the left sibling. + --mutable_finish(); + parent()->emplace_value(position(), alloc, finish_slot()); + value_destroy(finish(), alloc); + parent()->init_child(position() + 1, dest); + + if (is_internal()) { + for (int i = dest->start(), j = finish() + 1; i <= dest->finish(); + ++i, ++j) { + assert(child(j) != nullptr); + dest->init_child(i, child(j)); + clear_child(j); + } + } +} + +template +void btree_node

::merge(btree_node *src, allocator_type *alloc) { + assert(parent() == src->parent()); + assert(position() + 1 == src->position()); + + // Move the delimiting value to the left node. + value_init(finish(), alloc, parent()->slot(position())); + + // Move the values from the right to the left node. + transfer_n(src->count(), finish() + 1, src->start(), src, alloc); + + if (is_internal()) { + // Move the child pointers from the right to the left node. + for (int i = src->start(), j = finish() + 1; i <= src->finish(); ++i, ++j) { + init_child(j, src->child(i)); + src->clear_child(i); + } + } + + // Fixup `finish` on the src and dest nodes. + set_finish(start() + 1 + count() + src->count()); + src->set_finish(src->start()); + + // Remove the value on the parent node and delete the src node. + parent()->remove_values(position(), /*to_erase=*/1, alloc); +} + +template +void btree_node

::clear_and_delete(btree_node *node, allocator_type *alloc) { + if (node->is_leaf()) { + node->value_destroy_n(node->start(), node->count(), alloc); + deallocate(LeafSize(node->max_count()), node, alloc); + return; + } + if (node->count() == 0) { + deallocate(InternalSize(), node, alloc); + return; + } + + // The parent of the root of the subtree we are deleting. + btree_node *delete_root_parent = node->parent(); + + // Navigate to the leftmost leaf under node, and then delete upwards. + while (node->is_internal()) node = node->start_child(); +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + // When generations are enabled, we delete the leftmost leaf last in case it's + // the parent of the root and we need to check whether it's a leaf before we + // can update the root's generation. + // TODO(ezb): if we change btree_node::is_root to check a bool inside the node + // instead of checking whether the parent is a leaf, we can remove this logic. + btree_node *leftmost_leaf = node; +#endif + // Use `int` because `pos` needs to be able to hold `kNodeSlots+1`, which + // isn't guaranteed to be a valid `field_type`. + int pos = node->position(); + btree_node *parent = node->parent(); + for (;;) { + // In each iteration of the next loop, we delete one leaf node and go right. + assert(pos <= parent->finish()); + do { + node = parent->child(pos); + if (node->is_internal()) { + // Navigate to the leftmost leaf under node. + while (node->is_internal()) node = node->start_child(); + pos = node->position(); + parent = node->parent(); + } + node->value_destroy_n(node->start(), node->count(), alloc); +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + if (leftmost_leaf != node) +#endif + deallocate(LeafSize(node->max_count()), node, alloc); + ++pos; + } while (pos <= parent->finish()); + + // Once we've deleted all children of parent, delete parent and go up/right. + assert(pos > parent->finish()); + do { + node = parent; + pos = node->position(); + parent = node->parent(); + node->value_destroy_n(node->start(), node->count(), alloc); + deallocate(InternalSize(), node, alloc); + if (parent == delete_root_parent) { +#ifdef ABSL_BTREE_ENABLE_GENERATIONS + deallocate(LeafSize(leftmost_leaf->max_count()), leftmost_leaf, alloc); +#endif + return; + } + ++pos; + } while (pos > parent->finish()); + } +} + +//// +// btree_iterator methods +template +void btree_iterator::increment_slow() { + if (node_->is_leaf()) { + assert(position_ >= node_->finish()); + btree_iterator save(*this); + while (position_ == node_->finish() && !node_->is_root()) { + assert(node_->parent()->child(node_->position()) == node_); + position_ = node_->position(); + node_ = node_->parent(); + } + // TODO(ezb): assert we aren't incrementing end() instead of handling. + if (position_ == node_->finish()) { + *this = save; + } + } else { + assert(position_ < node_->finish()); + node_ = node_->child(position_ + 1); + while (node_->is_internal()) { + node_ = node_->start_child(); + } + position_ = node_->start(); + } +} + +template +void btree_iterator::decrement_slow() { + if (node_->is_leaf()) { + assert(position_ <= -1); + btree_iterator save(*this); + while (position_ < node_->start() && !node_->is_root()) { + assert(node_->parent()->child(node_->position()) == node_); + position_ = node_->position() - 1; + node_ = node_->parent(); + } + // TODO(ezb): assert we aren't decrementing begin() instead of handling. + if (position_ < node_->start()) { + *this = save; + } + } else { + assert(position_ >= node_->start()); + node_ = node_->child(position_); + while (node_->is_internal()) { + node_ = node_->child(node_->finish()); + } + position_ = node_->finish() - 1; + } +} + +//// +// btree methods +template +template +void btree

::copy_or_move_values_in_order(Btree &other) { + static_assert(std::is_same::value || + std::is_same::value, + "Btree type must be same or const."); + assert(empty()); + + // We can avoid key comparisons because we know the order of the + // values is the same order we'll store them in. + auto iter = other.begin(); + if (iter == other.end()) return; + insert_multi(iter.slot()); + ++iter; + for (; iter != other.end(); ++iter) { + // If the btree is not empty, we can just insert the new value at the end + // of the tree. + internal_emplace(end(), iter.slot()); + } +} + +template +constexpr bool btree

::static_assert_validation() { + static_assert(std::is_nothrow_copy_constructible::value, + "Key comparison must be nothrow copy constructible"); + static_assert(std::is_nothrow_copy_constructible::value, + "Allocator must be nothrow copy constructible"); + static_assert(type_traits_internal::is_trivially_copyable::value, + "iterator not trivially copyable."); + + // Note: We assert that kTargetValues, which is computed from + // Params::kTargetNodeSize, must fit the node_type::field_type. + static_assert( + kNodeSlots < (1 << (8 * sizeof(typename node_type::field_type))), + "target node size too large"); + + // Verify that key_compare returns an absl::{weak,strong}_ordering or bool. + static_assert( + compare_has_valid_result_type(), + "key comparison function must return absl::{weak,strong}_ordering or " + "bool."); + + // Test the assumption made in setting kNodeSlotSpace. + static_assert(node_type::MinimumOverhead() >= sizeof(void *) + 4, + "node space assumption incorrect"); + + return true; +} + +template +template +auto btree

::lower_bound_equal(const K &key) const + -> std::pair { + const SearchResult res = + internal_lower_bound(key); + const iterator lower = iterator(internal_end(res.value)); + const bool equal = res.HasMatch() + ? res.IsEq() + : lower != end() && !compare_keys(key, lower.key()); + return {lower, equal}; +} + +template +template +auto btree

::equal_range(const K &key) -> std::pair { + const std::pair lower_and_equal = lower_bound_equal(key); + const iterator lower = lower_and_equal.first; + if (!lower_and_equal.second) { + return {lower, lower}; + } + + const iterator next = std::next(lower); + if (!params_type::template can_have_multiple_equivalent_keys()) { + // The next iterator after lower must point to a key greater than `key`. + // Note: if this assert fails, then it may indicate that the comparator does + // not meet the equivalence requirements for Compare + // (see https://en.cppreference.com/w/cpp/named_req/Compare). + assert(next == end() || compare_keys(key, next.key())); + return {lower, next}; + } + // Try once more to avoid the call to upper_bound() if there's only one + // equivalent key. This should prevent all calls to upper_bound() in cases of + // unique-containers with heterogeneous comparators in which all comparison + // operators have the same equivalence classes. + if (next == end() || compare_keys(key, next.key())) return {lower, next}; + + // In this case, we need to call upper_bound() to avoid worst case O(N) + // behavior if we were to iterate over equal keys. + return {lower, upper_bound(key)}; +} + +template +template +auto btree

::insert_unique(const K &key, Args &&... args) + -> std::pair { + if (empty()) { + mutable_root() = mutable_rightmost() = new_leaf_root_node(1); + } + + SearchResult res = internal_locate(key); + iterator iter = res.value; + + if (res.HasMatch()) { + if (res.IsEq()) { + // The key already exists in the tree, do nothing. + return {iter, false}; + } + } else { + iterator last = internal_last(iter); + if (last.node_ && !compare_keys(key, last.key())) { + // The key already exists in the tree, do nothing. + return {last, false}; + } + } + return {internal_emplace(iter, std::forward(args)...), true}; +} + +template +template +inline auto btree

::insert_hint_unique(iterator position, const K &key, + Args &&... args) + -> std::pair { + if (!empty()) { + if (position == end() || compare_keys(key, position.key())) { + if (position == begin() || compare_keys(std::prev(position).key(), key)) { + // prev.key() < key < position.key() + return {internal_emplace(position, std::forward(args)...), true}; + } + } else if (compare_keys(position.key(), key)) { + ++position; + if (position == end() || compare_keys(key, position.key())) { + // {original `position`}.key() < key < {current `position`}.key() + return {internal_emplace(position, std::forward(args)...), true}; + } + } else { + // position.key() == key + return {position, false}; + } + } + return insert_unique(key, std::forward(args)...); +} + +template +template +void btree

::insert_iterator_unique(InputIterator b, InputIterator e, int) { + for (; b != e; ++b) { + insert_hint_unique(end(), params_type::key(*b), *b); + } +} + +template +template +void btree

::insert_iterator_unique(InputIterator b, InputIterator e, char) { + for (; b != e; ++b) { + // Use a node handle to manage a temp slot. + auto node_handle = + CommonAccess::Construct(get_allocator(), *b); + slot_type *slot = CommonAccess::GetSlot(node_handle); + insert_hint_unique(end(), params_type::key(slot), slot); + } +} + +template +template +auto btree

::insert_multi(const key_type &key, ValueType &&v) -> iterator { + if (empty()) { + mutable_root() = mutable_rightmost() = new_leaf_root_node(1); + } + + iterator iter = internal_upper_bound(key); + if (iter.node_ == nullptr) { + iter = end(); + } + return internal_emplace(iter, std::forward(v)); +} + +template +template +auto btree

::insert_hint_multi(iterator position, ValueType &&v) -> iterator { + if (!empty()) { + const key_type &key = params_type::key(v); + if (position == end() || !compare_keys(position.key(), key)) { + if (position == begin() || + !compare_keys(key, std::prev(position).key())) { + // prev.key() <= key <= position.key() + return internal_emplace(position, std::forward(v)); + } + } else { + ++position; + if (position == end() || !compare_keys(position.key(), key)) { + // {original `position`}.key() < key < {current `position`}.key() + return internal_emplace(position, std::forward(v)); + } + } + } + return insert_multi(std::forward(v)); +} + +template +template +void btree

::insert_iterator_multi(InputIterator b, InputIterator e) { + for (; b != e; ++b) { + insert_hint_multi(end(), *b); + } +} + +template +auto btree

::operator=(const btree &other) -> btree & { + if (this != &other) { + clear(); + + *mutable_key_comp() = other.key_comp(); + if (absl::allocator_traits< + allocator_type>::propagate_on_container_copy_assignment::value) { + *mutable_allocator() = other.allocator(); + } + + copy_or_move_values_in_order(other); + } + return *this; +} + +template +auto btree

::operator=(btree &&other) noexcept -> btree & { + if (this != &other) { + clear(); + + using std::swap; + if (absl::allocator_traits< + allocator_type>::propagate_on_container_copy_assignment::value) { + swap(root_, other.root_); + // Note: `rightmost_` also contains the allocator and the key comparator. + swap(rightmost_, other.rightmost_); + swap(size_, other.size_); + } else { + if (allocator() == other.allocator()) { + swap(mutable_root(), other.mutable_root()); + swap(*mutable_key_comp(), *other.mutable_key_comp()); + swap(mutable_rightmost(), other.mutable_rightmost()); + swap(size_, other.size_); + } else { + // We aren't allowed to propagate the allocator and the allocator is + // different so we can't take over its memory. We must move each element + // individually. We need both `other` and `this` to have `other`s key + // comparator while moving the values so we can't swap the key + // comparators. + *mutable_key_comp() = other.key_comp(); + copy_or_move_values_in_order(other); + } + } + } + return *this; +} + +template +auto btree

::erase(iterator iter) -> iterator { + iter.node_->value_destroy(iter.position_, mutable_allocator()); + iter.update_generation(); + + const bool internal_delete = iter.node_->is_internal(); + if (internal_delete) { + // Deletion of a value on an internal node. First, transfer the largest + // value from our left child here, then erase/rebalance from that position. + // We can get to the largest value from our left child by decrementing iter. + iterator internal_iter(iter); + --iter; + assert(iter.node_->is_leaf()); + internal_iter.node_->transfer(internal_iter.position_, iter.position_, + iter.node_, mutable_allocator()); + } else { + // Shift values after erased position in leaf. In the internal case, we + // don't need to do this because the leaf position is the end of the node. + const field_type transfer_from = iter.position_ + 1; + const field_type num_to_transfer = iter.node_->finish() - transfer_from; + iter.node_->transfer_n(num_to_transfer, iter.position_, transfer_from, + iter.node_, mutable_allocator()); + } + // Update node finish and container size. + iter.node_->set_finish(iter.node_->finish() - 1); + --size_; + + // We want to return the next value after the one we just erased. If we + // erased from an internal node (internal_delete == true), then the next + // value is ++(++iter). If we erased from a leaf node (internal_delete == + // false) then the next value is ++iter. Note that ++iter may point to an + // internal node and the value in the internal node may move to a leaf node + // (iter.node_) when rebalancing is performed at the leaf level. + + iterator res = rebalance_after_delete(iter); + + // If we erased from an internal node, advance the iterator. + if (internal_delete) { + ++res; + } + return res; +} + +template +auto btree

::rebalance_after_delete(iterator iter) -> iterator { + // Merge/rebalance as we walk back up the tree. + iterator res(iter); + bool first_iteration = true; + for (;;) { + if (iter.node_ == root()) { + try_shrink(); + if (empty()) { + return end(); + } + break; + } + if (iter.node_->count() >= kMinNodeValues) { + break; + } + bool merged = try_merge_or_rebalance(&iter); + // On the first iteration, we should update `res` with `iter` because `res` + // may have been invalidated. + if (first_iteration) { + res = iter; + first_iteration = false; + } + if (!merged) { + break; + } + iter.position_ = iter.node_->position(); + iter.node_ = iter.node_->parent(); + } + res.update_generation(); + + // Adjust our return value. If we're pointing at the end of a node, advance + // the iterator. + if (res.position_ == res.node_->finish()) { + res.position_ = res.node_->finish() - 1; + ++res; + } + + return res; +} + +template +auto btree

::erase_range(iterator begin, iterator end) + -> std::pair { + difference_type count = std::distance(begin, end); + assert(count >= 0); + + if (count == 0) { + return {0, begin}; + } + + if (static_cast(count) == size_) { + clear(); + return {count, this->end()}; + } + + if (begin.node_ == end.node_) { + assert(end.position_ > begin.position_); + begin.node_->remove_values(begin.position_, end.position_ - begin.position_, + mutable_allocator()); + size_ -= count; + return {count, rebalance_after_delete(begin)}; + } + + const size_type target_size = size_ - count; + while (size_ > target_size) { + if (begin.node_->is_leaf()) { + const size_type remaining_to_erase = size_ - target_size; + const size_type remaining_in_node = + begin.node_->finish() - begin.position_; + const size_type to_erase = + (std::min)(remaining_to_erase, remaining_in_node); + begin.node_->remove_values(begin.position_, to_erase, + mutable_allocator()); + size_ -= to_erase; + begin = rebalance_after_delete(begin); + } else { + begin = erase(begin); + } + } + begin.update_generation(); + return {count, begin}; +} + +template +void btree

::clear() { + if (!empty()) { + node_type::clear_and_delete(root(), mutable_allocator()); + } + mutable_root() = mutable_rightmost() = EmptyNode(); + size_ = 0; +} + +template +void btree

::swap(btree &other) { + using std::swap; + if (absl::allocator_traits< + allocator_type>::propagate_on_container_swap::value) { + // Note: `rightmost_` also contains the allocator and the key comparator. + swap(rightmost_, other.rightmost_); + } else { + // It's undefined behavior if the allocators are unequal here. + assert(allocator() == other.allocator()); + swap(mutable_rightmost(), other.mutable_rightmost()); + swap(*mutable_key_comp(), *other.mutable_key_comp()); + } + swap(mutable_root(), other.mutable_root()); + swap(size_, other.size_); +} + +template +void btree

::verify() const { + assert(root() != nullptr); + assert(leftmost() != nullptr); + assert(rightmost() != nullptr); + assert(empty() || size() == internal_verify(root(), nullptr, nullptr)); + assert(leftmost() == (++const_iterator(root(), -1)).node_); + assert(rightmost() == (--const_iterator(root(), root()->finish())).node_); + assert(leftmost()->is_leaf()); + assert(rightmost()->is_leaf()); +} + +template +void btree

::rebalance_or_split(iterator *iter) { + node_type *&node = iter->node_; + int &insert_position = iter->position_; + assert(node->count() == node->max_count()); + assert(kNodeSlots == node->max_count()); + + // First try to make room on the node by rebalancing. + node_type *parent = node->parent(); + if (node != root()) { + if (node->position() > parent->start()) { + // Try rebalancing with our left sibling. + node_type *left = parent->child(node->position() - 1); + assert(left->max_count() == kNodeSlots); + if (left->count() < kNodeSlots) { + // We bias rebalancing based on the position being inserted. If we're + // inserting at the end of the right node then we bias rebalancing to + // fill up the left node. + int to_move = (kNodeSlots - left->count()) / + (1 + (insert_position < static_cast(kNodeSlots))); + to_move = (std::max)(1, to_move); + + if (insert_position - to_move >= node->start() || + left->count() + to_move < static_cast(kNodeSlots)) { + left->rebalance_right_to_left(to_move, node, mutable_allocator()); + + assert(node->max_count() - node->count() == to_move); + insert_position = insert_position - to_move; + if (insert_position < node->start()) { + insert_position = insert_position + left->count() + 1; + node = left; + } + + assert(node->count() < node->max_count()); + return; + } + } + } + + if (node->position() < parent->finish()) { + // Try rebalancing with our right sibling. + node_type *right = parent->child(node->position() + 1); + assert(right->max_count() == kNodeSlots); + if (right->count() < kNodeSlots) { + // We bias rebalancing based on the position being inserted. If we're + // inserting at the beginning of the left node then we bias rebalancing + // to fill up the right node. + int to_move = (static_cast(kNodeSlots) - right->count()) / + (1 + (insert_position > node->start())); + to_move = (std::max)(1, to_move); + + if (insert_position <= node->finish() - to_move || + right->count() + to_move < static_cast(kNodeSlots)) { + node->rebalance_left_to_right(to_move, right, mutable_allocator()); + + if (insert_position > node->finish()) { + insert_position = insert_position - node->count() - 1; + node = right; + } + + assert(node->count() < node->max_count()); + return; + } + } + } + + // Rebalancing failed, make sure there is room on the parent node for a new + // value. + assert(parent->max_count() == kNodeSlots); + if (parent->count() == kNodeSlots) { + iterator parent_iter(node->parent(), node->position()); + rebalance_or_split(&parent_iter); + } + } else { + // Rebalancing not possible because this is the root node. + // Create a new root node and set the current root node as the child of the + // new root. + parent = new_internal_node(parent); + parent->set_generation(root()->generation()); + parent->init_child(parent->start(), root()); + mutable_root() = parent; + // If the former root was a leaf node, then it's now the rightmost node. + assert(parent->start_child()->is_internal() || + parent->start_child() == rightmost()); + } + + // Split the node. + node_type *split_node; + if (node->is_leaf()) { + split_node = new_leaf_node(parent); + node->split(insert_position, split_node, mutable_allocator()); + if (rightmost() == node) mutable_rightmost() = split_node; + } else { + split_node = new_internal_node(parent); + node->split(insert_position, split_node, mutable_allocator()); + } + + if (insert_position > node->finish()) { + insert_position = insert_position - node->count() - 1; + node = split_node; + } +} + +template +void btree

::merge_nodes(node_type *left, node_type *right) { + left->merge(right, mutable_allocator()); + if (rightmost() == right) mutable_rightmost() = left; +} + +template +bool btree

::try_merge_or_rebalance(iterator *iter) { + node_type *parent = iter->node_->parent(); + if (iter->node_->position() > parent->start()) { + // Try merging with our left sibling. + node_type *left = parent->child(iter->node_->position() - 1); + assert(left->max_count() == kNodeSlots); + if (1U + left->count() + iter->node_->count() <= kNodeSlots) { + iter->position_ += 1 + left->count(); + merge_nodes(left, iter->node_); + iter->node_ = left; + return true; + } + } + if (iter->node_->position() < parent->finish()) { + // Try merging with our right sibling. + node_type *right = parent->child(iter->node_->position() + 1); + assert(right->max_count() == kNodeSlots); + if (1U + iter->node_->count() + right->count() <= kNodeSlots) { + merge_nodes(iter->node_, right); + return true; + } + // Try rebalancing with our right sibling. We don't perform rebalancing if + // we deleted the first element from iter->node_ and the node is not + // empty. This is a small optimization for the common pattern of deleting + // from the front of the tree. + if (right->count() > kMinNodeValues && + (iter->node_->count() == 0 || iter->position_ > iter->node_->start())) { + int to_move = (right->count() - iter->node_->count()) / 2; + to_move = (std::min)(to_move, right->count() - 1); + iter->node_->rebalance_right_to_left(to_move, right, mutable_allocator()); + return false; + } + } + if (iter->node_->position() > parent->start()) { + // Try rebalancing with our left sibling. We don't perform rebalancing if + // we deleted the last element from iter->node_ and the node is not + // empty. This is a small optimization for the common pattern of deleting + // from the back of the tree. + node_type *left = parent->child(iter->node_->position() - 1); + if (left->count() > kMinNodeValues && + (iter->node_->count() == 0 || + iter->position_ < iter->node_->finish())) { + int to_move = (left->count() - iter->node_->count()) / 2; + to_move = (std::min)(to_move, left->count() - 1); + left->rebalance_left_to_right(to_move, iter->node_, mutable_allocator()); + iter->position_ += to_move; + return false; + } + } + return false; +} + +template +void btree

::try_shrink() { + node_type *orig_root = root(); + if (orig_root->count() > 0) { + return; + } + // Deleted the last item on the root node, shrink the height of the tree. + if (orig_root->is_leaf()) { + assert(size() == 0); + mutable_root() = mutable_rightmost() = EmptyNode(); + } else { + node_type *child = orig_root->start_child(); + child->make_root(); + mutable_root() = child; + } + node_type::clear_and_delete(orig_root, mutable_allocator()); +} + +template +template +inline IterType btree

::internal_last(IterType iter) { + assert(iter.node_ != nullptr); + while (iter.position_ == iter.node_->finish()) { + iter.position_ = iter.node_->position(); + iter.node_ = iter.node_->parent(); + if (iter.node_->is_leaf()) { + iter.node_ = nullptr; + break; + } + } + iter.update_generation(); + return iter; +} + +template +template +inline auto btree

::internal_emplace(iterator iter, Args &&... args) + -> iterator { + if (iter.node_->is_internal()) { + // We can't insert on an internal node. Instead, we'll insert after the + // previous value which is guaranteed to be on a leaf node. + --iter; + ++iter.position_; + } + const field_type max_count = iter.node_->max_count(); + allocator_type *alloc = mutable_allocator(); + if (iter.node_->count() == max_count) { + // Make room in the leaf for the new item. + if (max_count < kNodeSlots) { + // Insertion into the root where the root is smaller than the full node + // size. Simply grow the size of the root node. + assert(iter.node_ == root()); + iter.node_ = + new_leaf_root_node((std::min)(kNodeSlots, 2 * max_count)); + // Transfer the values from the old root to the new root. + node_type *old_root = root(); + node_type *new_root = iter.node_; + new_root->transfer_n(old_root->count(), new_root->start(), + old_root->start(), old_root, alloc); + new_root->set_finish(old_root->finish()); + old_root->set_finish(old_root->start()); + new_root->set_generation(old_root->generation()); + node_type::clear_and_delete(old_root, alloc); + mutable_root() = mutable_rightmost() = new_root; + } else { + rebalance_or_split(&iter); + } + } + iter.node_->emplace_value(iter.position_, alloc, std::forward(args)...); + ++size_; + iter.update_generation(); + return iter; +} + +template +template +inline auto btree

::internal_locate(const K &key) const + -> SearchResult { + iterator iter(const_cast(root())); + for (;;) { + SearchResult res = + iter.node_->lower_bound(key, key_comp()); + iter.position_ = res.value; + if (res.IsEq()) { + return {iter, MatchKind::kEq}; + } + // Note: in the non-key-compare-to case, we don't need to walk all the way + // down the tree if the keys are equal, but determining equality would + // require doing an extra comparison on each node on the way down, and we + // will need to go all the way to the leaf node in the expected case. + if (iter.node_->is_leaf()) { + break; + } + iter.node_ = iter.node_->child(iter.position_); + } + // Note: in the non-key-compare-to case, the key may actually be equivalent + // here (and the MatchKind::kNe is ignored). + return {iter, MatchKind::kNe}; +} + +template +template +auto btree

::internal_lower_bound(const K &key) const + -> SearchResult { + if (!params_type::template can_have_multiple_equivalent_keys()) { + SearchResult ret = internal_locate(key); + ret.value = internal_last(ret.value); + return ret; + } + iterator iter(const_cast(root())); + SearchResult res; + bool seen_eq = false; + for (;;) { + res = iter.node_->lower_bound(key, key_comp()); + iter.position_ = res.value; + if (iter.node_->is_leaf()) { + break; + } + seen_eq = seen_eq || res.IsEq(); + iter.node_ = iter.node_->child(iter.position_); + } + if (res.IsEq()) return {iter, MatchKind::kEq}; + return {internal_last(iter), seen_eq ? MatchKind::kEq : MatchKind::kNe}; +} + +template +template +auto btree

::internal_upper_bound(const K &key) const -> iterator { + iterator iter(const_cast(root())); + for (;;) { + iter.position_ = iter.node_->upper_bound(key, key_comp()); + if (iter.node_->is_leaf()) { + break; + } + iter.node_ = iter.node_->child(iter.position_); + } + return internal_last(iter); +} + +template +template +auto btree

::internal_find(const K &key) const -> iterator { + SearchResult res = internal_locate(key); + if (res.HasMatch()) { + if (res.IsEq()) { + return res.value; + } + } else { + const iterator iter = internal_last(res.value); + if (iter.node_ != nullptr && !compare_keys(key, iter.key())) { + return iter; + } + } + return {nullptr, 0}; +} + +template +int btree

::internal_verify(const node_type *node, const key_type *lo, + const key_type *hi) const { + assert(node->count() > 0); + assert(node->count() <= node->max_count()); + if (lo) { + assert(!compare_keys(node->key(node->start()), *lo)); + } + if (hi) { + assert(!compare_keys(*hi, node->key(node->finish() - 1))); + } + for (int i = node->start() + 1; i < node->finish(); ++i) { + assert(!compare_keys(node->key(i), node->key(i - 1))); + } + int count = node->count(); + if (node->is_internal()) { + for (int i = node->start(); i <= node->finish(); ++i) { + assert(node->child(i) != nullptr); + assert(node->child(i)->parent() == node); + assert(node->child(i)->position() == i); + count += internal_verify(node->child(i), + i == node->start() ? lo : &node->key(i - 1), + i == node->finish() ? hi : &node->key(i)); + } + } + return count; +} + +struct btree_access { + template + static auto erase_if(BtreeContainer &container, Pred pred) + -> typename BtreeContainer::size_type { + const auto initial_size = container.size(); + auto &tree = container.tree_; + auto *alloc = tree.mutable_allocator(); + for (auto it = container.begin(); it != container.end();) { + if (!pred(*it)) { + ++it; + continue; + } + auto *node = it.node_; + if (node->is_internal()) { + // Handle internal nodes normally. + it = container.erase(it); + continue; + } + // If this is a leaf node, then we do all the erases from this node + // at once before doing rebalancing. + + // The current position to transfer slots to. + int to_pos = it.position_; + node->value_destroy(it.position_, alloc); + while (++it.position_ < node->finish()) { + it.update_generation(); + if (pred(*it)) { + node->value_destroy(it.position_, alloc); + } else { + node->transfer(node->slot(to_pos++), node->slot(it.position_), alloc); + } + } + const int num_deleted = node->finish() - to_pos; + tree.size_ -= num_deleted; + node->set_finish(to_pos); + it.position_ = to_pos; + it = tree.rebalance_after_delete(it); + } + return initial_size - container.size(); + } +}; + +#undef ABSL_BTREE_ENABLE_GENERATIONS + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_BTREE_H_ diff --git a/src/absl/container/internal/btree_container.h b/src/absl/container/internal/btree_container.h new file mode 100644 index 0000000..fc2f740 --- /dev/null +++ b/src/absl/container/internal/btree_container.h @@ -0,0 +1,699 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_BTREE_CONTAINER_H_ +#define ABSL_CONTAINER_INTERNAL_BTREE_CONTAINER_H_ + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/internal/throw_delegate.h" +#include "absl/container/internal/btree.h" // IWYU pragma: export +#include "absl/container/internal/common.h" +#include "absl/memory/memory.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// A common base class for btree_set, btree_map, btree_multiset, and +// btree_multimap. +template +class btree_container { + using params_type = typename Tree::params_type; + + protected: + // Alias used for heterogeneous lookup functions. + // `key_arg` evaluates to `K` when the functors are transparent and to + // `key_type` otherwise. It permits template argument deduction on `K` for the + // transparent case. + template + using key_arg = + typename KeyArg::template type< + K, typename Tree::key_type>; + + public: + using key_type = typename Tree::key_type; + using value_type = typename Tree::value_type; + using size_type = typename Tree::size_type; + using difference_type = typename Tree::difference_type; + using key_compare = typename Tree::original_key_compare; + using value_compare = typename Tree::value_compare; + using allocator_type = typename Tree::allocator_type; + using reference = typename Tree::reference; + using const_reference = typename Tree::const_reference; + using pointer = typename Tree::pointer; + using const_pointer = typename Tree::const_pointer; + using iterator = typename Tree::iterator; + using const_iterator = typename Tree::const_iterator; + using reverse_iterator = typename Tree::reverse_iterator; + using const_reverse_iterator = typename Tree::const_reverse_iterator; + using node_type = typename Tree::node_handle_type; + + // Constructors/assignments. + btree_container() : tree_(key_compare(), allocator_type()) {} + explicit btree_container(const key_compare &comp, + const allocator_type &alloc = allocator_type()) + : tree_(comp, alloc) {} + explicit btree_container(const allocator_type &alloc) + : tree_(key_compare(), alloc) {} + + btree_container(const btree_container &other) + : btree_container(other, absl::allocator_traits:: + select_on_container_copy_construction( + other.get_allocator())) {} + btree_container(const btree_container &other, const allocator_type &alloc) + : tree_(other.tree_, alloc) {} + + btree_container(btree_container &&other) noexcept( + std::is_nothrow_move_constructible::value) = default; + btree_container(btree_container &&other, const allocator_type &alloc) + : tree_(std::move(other.tree_), alloc) {} + + btree_container &operator=(const btree_container &other) = default; + btree_container &operator=(btree_container &&other) noexcept( + std::is_nothrow_move_assignable::value) = default; + + // Iterator routines. + iterator begin() { return tree_.begin(); } + const_iterator begin() const { return tree_.begin(); } + const_iterator cbegin() const { return tree_.begin(); } + iterator end() { return tree_.end(); } + const_iterator end() const { return tree_.end(); } + const_iterator cend() const { return tree_.end(); } + reverse_iterator rbegin() { return tree_.rbegin(); } + const_reverse_iterator rbegin() const { return tree_.rbegin(); } + const_reverse_iterator crbegin() const { return tree_.rbegin(); } + reverse_iterator rend() { return tree_.rend(); } + const_reverse_iterator rend() const { return tree_.rend(); } + const_reverse_iterator crend() const { return tree_.rend(); } + + // Lookup routines. + template + size_type count(const key_arg &key) const { + auto equal_range = this->equal_range(key); + return std::distance(equal_range.first, equal_range.second); + } + template + iterator find(const key_arg &key) { + return tree_.find(key); + } + template + const_iterator find(const key_arg &key) const { + return tree_.find(key); + } + template + bool contains(const key_arg &key) const { + return find(key) != end(); + } + template + iterator lower_bound(const key_arg &key) { + return tree_.lower_bound(key); + } + template + const_iterator lower_bound(const key_arg &key) const { + return tree_.lower_bound(key); + } + template + iterator upper_bound(const key_arg &key) { + return tree_.upper_bound(key); + } + template + const_iterator upper_bound(const key_arg &key) const { + return tree_.upper_bound(key); + } + template + std::pair equal_range(const key_arg &key) { + return tree_.equal_range(key); + } + template + std::pair equal_range( + const key_arg &key) const { + return tree_.equal_range(key); + } + + // Deletion routines. Note that there is also a deletion routine that is + // specific to btree_set_container/btree_multiset_container. + + // Erase the specified iterator from the btree. The iterator must be valid + // (i.e. not equal to end()). Return an iterator pointing to the node after + // the one that was erased (or end() if none exists). + iterator erase(const_iterator iter) { return tree_.erase(iterator(iter)); } + iterator erase(iterator iter) { return tree_.erase(iter); } + iterator erase(const_iterator first, const_iterator last) { + return tree_.erase_range(iterator(first), iterator(last)).second; + } + template + size_type erase(const key_arg &key) { + auto equal_range = this->equal_range(key); + return tree_.erase_range(equal_range.first, equal_range.second).first; + } + + // Extract routines. + node_type extract(iterator position) { + // Use Construct instead of Transfer because the rebalancing code will + // destroy the slot later. + auto node = + CommonAccess::Construct(get_allocator(), position.slot()); + erase(position); + return node; + } + node_type extract(const_iterator position) { + return extract(iterator(position)); + } + + // Utility routines. + ABSL_ATTRIBUTE_REINITIALIZES void clear() { tree_.clear(); } + void swap(btree_container &other) { tree_.swap(other.tree_); } + void verify() const { tree_.verify(); } + + // Size routines. + size_type size() const { return tree_.size(); } + size_type max_size() const { return tree_.max_size(); } + bool empty() const { return tree_.empty(); } + + friend bool operator==(const btree_container &x, const btree_container &y) { + if (x.size() != y.size()) return false; + return std::equal(x.begin(), x.end(), y.begin()); + } + + friend bool operator!=(const btree_container &x, const btree_container &y) { + return !(x == y); + } + + friend bool operator<(const btree_container &x, const btree_container &y) { + return std::lexicographical_compare(x.begin(), x.end(), y.begin(), y.end()); + } + + friend bool operator>(const btree_container &x, const btree_container &y) { + return y < x; + } + + friend bool operator<=(const btree_container &x, const btree_container &y) { + return !(y < x); + } + + friend bool operator>=(const btree_container &x, const btree_container &y) { + return !(x < y); + } + + // The allocator used by the btree. + allocator_type get_allocator() const { return tree_.get_allocator(); } + + // The key comparator used by the btree. + key_compare key_comp() const { return key_compare(tree_.key_comp()); } + value_compare value_comp() const { return tree_.value_comp(); } + + // Support absl::Hash. + template + friend State AbslHashValue(State h, const btree_container &b) { + for (const auto &v : b) { + h = State::combine(std::move(h), v); + } + return State::combine(std::move(h), b.size()); + } + + protected: + friend struct btree_access; + Tree tree_; +}; + +// A common base class for btree_set and btree_map. +template +class btree_set_container : public btree_container { + using super_type = btree_container; + using params_type = typename Tree::params_type; + using init_type = typename params_type::init_type; + using is_key_compare_to = typename params_type::is_key_compare_to; + friend class BtreeNodePeer; + + protected: + template + using key_arg = typename super_type::template key_arg; + + public: + using key_type = typename Tree::key_type; + using value_type = typename Tree::value_type; + using size_type = typename Tree::size_type; + using key_compare = typename Tree::original_key_compare; + using allocator_type = typename Tree::allocator_type; + using iterator = typename Tree::iterator; + using const_iterator = typename Tree::const_iterator; + using node_type = typename super_type::node_type; + using insert_return_type = InsertReturnType; + + // Inherit constructors. + using super_type::super_type; + btree_set_container() {} + + // Range constructors. + template + btree_set_container(InputIterator b, InputIterator e, + const key_compare &comp = key_compare(), + const allocator_type &alloc = allocator_type()) + : super_type(comp, alloc) { + insert(b, e); + } + template + btree_set_container(InputIterator b, InputIterator e, + const allocator_type &alloc) + : btree_set_container(b, e, key_compare(), alloc) {} + + // Initializer list constructors. + btree_set_container(std::initializer_list init, + const key_compare &comp = key_compare(), + const allocator_type &alloc = allocator_type()) + : btree_set_container(init.begin(), init.end(), comp, alloc) {} + btree_set_container(std::initializer_list init, + const allocator_type &alloc) + : btree_set_container(init.begin(), init.end(), alloc) {} + + // Insertion routines. + std::pair insert(const value_type &v) { + return this->tree_.insert_unique(params_type::key(v), v); + } + std::pair insert(value_type &&v) { + return this->tree_.insert_unique(params_type::key(v), std::move(v)); + } + template + std::pair emplace(Args &&... args) { + // Use a node handle to manage a temp slot. + auto node = CommonAccess::Construct(this->get_allocator(), + std::forward(args)...); + auto *slot = CommonAccess::GetSlot(node); + return this->tree_.insert_unique(params_type::key(slot), slot); + } + iterator insert(const_iterator hint, const value_type &v) { + return this->tree_ + .insert_hint_unique(iterator(hint), params_type::key(v), v) + .first; + } + iterator insert(const_iterator hint, value_type &&v) { + return this->tree_ + .insert_hint_unique(iterator(hint), params_type::key(v), std::move(v)) + .first; + } + template + iterator emplace_hint(const_iterator hint, Args &&... args) { + // Use a node handle to manage a temp slot. + auto node = CommonAccess::Construct(this->get_allocator(), + std::forward(args)...); + auto *slot = CommonAccess::GetSlot(node); + return this->tree_ + .insert_hint_unique(iterator(hint), params_type::key(slot), slot) + .first; + } + template + void insert(InputIterator b, InputIterator e) { + this->tree_.insert_iterator_unique(b, e, 0); + } + void insert(std::initializer_list init) { + this->tree_.insert_iterator_unique(init.begin(), init.end(), 0); + } + insert_return_type insert(node_type &&node) { + if (!node) return {this->end(), false, node_type()}; + std::pair res = + this->tree_.insert_unique(params_type::key(CommonAccess::GetSlot(node)), + CommonAccess::GetSlot(node)); + if (res.second) { + CommonAccess::Destroy(&node); + return {res.first, true, node_type()}; + } else { + return {res.first, false, std::move(node)}; + } + } + iterator insert(const_iterator hint, node_type &&node) { + if (!node) return this->end(); + std::pair res = this->tree_.insert_hint_unique( + iterator(hint), params_type::key(CommonAccess::GetSlot(node)), + CommonAccess::GetSlot(node)); + if (res.second) CommonAccess::Destroy(&node); + return res.first; + } + + // Node extraction routines. + template + node_type extract(const key_arg &key) { + const std::pair lower_and_equal = + this->tree_.lower_bound_equal(key); + return lower_and_equal.second ? extract(lower_and_equal.first) + : node_type(); + } + using super_type::extract; + + // Merge routines. + // Moves elements from `src` into `this`. If the element already exists in + // `this`, it is left unmodified in `src`. + template < + typename T, + typename absl::enable_if_t< + absl::conjunction< + std::is_same, + std::is_same, + std::is_same>::value, + int> = 0> + void merge(btree_container &src) { // NOLINT + for (auto src_it = src.begin(); src_it != src.end();) { + if (insert(std::move(params_type::element(src_it.slot()))).second) { + src_it = src.erase(src_it); + } else { + ++src_it; + } + } + } + + template < + typename T, + typename absl::enable_if_t< + absl::conjunction< + std::is_same, + std::is_same, + std::is_same>::value, + int> = 0> + void merge(btree_container &&src) { + merge(src); + } +}; + +// Base class for btree_map. +template +class btree_map_container : public btree_set_container { + using super_type = btree_set_container; + using params_type = typename Tree::params_type; + friend class BtreeNodePeer; + + private: + template + using key_arg = typename super_type::template key_arg; + + public: + using key_type = typename Tree::key_type; + using mapped_type = typename params_type::mapped_type; + using value_type = typename Tree::value_type; + using key_compare = typename Tree::original_key_compare; + using allocator_type = typename Tree::allocator_type; + using iterator = typename Tree::iterator; + using const_iterator = typename Tree::const_iterator; + + // Inherit constructors. + using super_type::super_type; + btree_map_container() {} + + // Insertion routines. + // Note: the nullptr template arguments and extra `const M&` overloads allow + // for supporting bitfield arguments. + template + std::pair insert_or_assign(const key_arg &k, + const M &obj) { + return insert_or_assign_impl(k, obj); + } + template + std::pair insert_or_assign(key_arg &&k, const M &obj) { + return insert_or_assign_impl(std::forward(k), obj); + } + template + std::pair insert_or_assign(const key_arg &k, M &&obj) { + return insert_or_assign_impl(k, std::forward(obj)); + } + template + std::pair insert_or_assign(key_arg &&k, M &&obj) { + return insert_or_assign_impl(std::forward(k), std::forward(obj)); + } + template + iterator insert_or_assign(const_iterator hint, const key_arg &k, + const M &obj) { + return insert_or_assign_hint_impl(hint, k, obj); + } + template + iterator insert_or_assign(const_iterator hint, key_arg &&k, const M &obj) { + return insert_or_assign_hint_impl(hint, std::forward(k), obj); + } + template + iterator insert_or_assign(const_iterator hint, const key_arg &k, M &&obj) { + return insert_or_assign_hint_impl(hint, k, std::forward(obj)); + } + template + iterator insert_or_assign(const_iterator hint, key_arg &&k, M &&obj) { + return insert_or_assign_hint_impl(hint, std::forward(k), + std::forward(obj)); + } + + template ::value, int> = 0> + std::pair try_emplace(const key_arg &k, Args &&... args) { + return try_emplace_impl(k, std::forward(args)...); + } + template ::value, int> = 0> + std::pair try_emplace(key_arg &&k, Args &&... args) { + return try_emplace_impl(std::forward(k), std::forward(args)...); + } + template + iterator try_emplace(const_iterator hint, const key_arg &k, + Args &&... args) { + return try_emplace_hint_impl(hint, k, std::forward(args)...); + } + template + iterator try_emplace(const_iterator hint, key_arg &&k, Args &&... args) { + return try_emplace_hint_impl(hint, std::forward(k), + std::forward(args)...); + } + + template + mapped_type &operator[](const key_arg &k) { + return try_emplace(k).first->second; + } + template + mapped_type &operator[](key_arg &&k) { + return try_emplace(std::forward(k)).first->second; + } + + template + mapped_type &at(const key_arg &key) { + auto it = this->find(key); + if (it == this->end()) + base_internal::ThrowStdOutOfRange("absl::btree_map::at"); + return it->second; + } + template + const mapped_type &at(const key_arg &key) const { + auto it = this->find(key); + if (it == this->end()) + base_internal::ThrowStdOutOfRange("absl::btree_map::at"); + return it->second; + } + + private: + // Note: when we call `std::forward(obj)` twice, it's safe because + // insert_unique/insert_hint_unique are guaranteed to not consume `obj` when + // `ret.second` is false. + template + std::pair insert_or_assign_impl(K &&k, M &&obj) { + const std::pair ret = + this->tree_.insert_unique(k, std::forward(k), std::forward(obj)); + if (!ret.second) ret.first->second = std::forward(obj); + return ret; + } + template + iterator insert_or_assign_hint_impl(const_iterator hint, K &&k, M &&obj) { + const std::pair ret = this->tree_.insert_hint_unique( + iterator(hint), k, std::forward(k), std::forward(obj)); + if (!ret.second) ret.first->second = std::forward(obj); + return ret.first; + } + + template + std::pair try_emplace_impl(K &&k, Args &&... args) { + return this->tree_.insert_unique( + k, std::piecewise_construct, std::forward_as_tuple(std::forward(k)), + std::forward_as_tuple(std::forward(args)...)); + } + template + iterator try_emplace_hint_impl(const_iterator hint, K &&k, Args &&... args) { + return this->tree_ + .insert_hint_unique(iterator(hint), k, std::piecewise_construct, + std::forward_as_tuple(std::forward(k)), + std::forward_as_tuple(std::forward(args)...)) + .first; + } +}; + +// A common base class for btree_multiset and btree_multimap. +template +class btree_multiset_container : public btree_container { + using super_type = btree_container; + using params_type = typename Tree::params_type; + using init_type = typename params_type::init_type; + using is_key_compare_to = typename params_type::is_key_compare_to; + friend class BtreeNodePeer; + + template + using key_arg = typename super_type::template key_arg; + + public: + using key_type = typename Tree::key_type; + using value_type = typename Tree::value_type; + using size_type = typename Tree::size_type; + using key_compare = typename Tree::original_key_compare; + using allocator_type = typename Tree::allocator_type; + using iterator = typename Tree::iterator; + using const_iterator = typename Tree::const_iterator; + using node_type = typename super_type::node_type; + + // Inherit constructors. + using super_type::super_type; + btree_multiset_container() {} + + // Range constructors. + template + btree_multiset_container(InputIterator b, InputIterator e, + const key_compare &comp = key_compare(), + const allocator_type &alloc = allocator_type()) + : super_type(comp, alloc) { + insert(b, e); + } + template + btree_multiset_container(InputIterator b, InputIterator e, + const allocator_type &alloc) + : btree_multiset_container(b, e, key_compare(), alloc) {} + + // Initializer list constructors. + btree_multiset_container(std::initializer_list init, + const key_compare &comp = key_compare(), + const allocator_type &alloc = allocator_type()) + : btree_multiset_container(init.begin(), init.end(), comp, alloc) {} + btree_multiset_container(std::initializer_list init, + const allocator_type &alloc) + : btree_multiset_container(init.begin(), init.end(), alloc) {} + + // Insertion routines. + iterator insert(const value_type &v) { return this->tree_.insert_multi(v); } + iterator insert(value_type &&v) { + return this->tree_.insert_multi(std::move(v)); + } + iterator insert(const_iterator hint, const value_type &v) { + return this->tree_.insert_hint_multi(iterator(hint), v); + } + iterator insert(const_iterator hint, value_type &&v) { + return this->tree_.insert_hint_multi(iterator(hint), std::move(v)); + } + template + void insert(InputIterator b, InputIterator e) { + this->tree_.insert_iterator_multi(b, e); + } + void insert(std::initializer_list init) { + this->tree_.insert_iterator_multi(init.begin(), init.end()); + } + template + iterator emplace(Args &&... args) { + // Use a node handle to manage a temp slot. + auto node = CommonAccess::Construct(this->get_allocator(), + std::forward(args)...); + return this->tree_.insert_multi(CommonAccess::GetSlot(node)); + } + template + iterator emplace_hint(const_iterator hint, Args &&... args) { + // Use a node handle to manage a temp slot. + auto node = CommonAccess::Construct(this->get_allocator(), + std::forward(args)...); + return this->tree_.insert_hint_multi(iterator(hint), + CommonAccess::GetSlot(node)); + } + iterator insert(node_type &&node) { + if (!node) return this->end(); + iterator res = + this->tree_.insert_multi(params_type::key(CommonAccess::GetSlot(node)), + CommonAccess::GetSlot(node)); + CommonAccess::Destroy(&node); + return res; + } + iterator insert(const_iterator hint, node_type &&node) { + if (!node) return this->end(); + iterator res = this->tree_.insert_hint_multi( + iterator(hint), + std::move(params_type::element(CommonAccess::GetSlot(node)))); + CommonAccess::Destroy(&node); + return res; + } + + // Node extraction routines. + template + node_type extract(const key_arg &key) { + const std::pair lower_and_equal = + this->tree_.lower_bound_equal(key); + return lower_and_equal.second ? extract(lower_and_equal.first) + : node_type(); + } + using super_type::extract; + + // Merge routines. + // Moves all elements from `src` into `this`. + template < + typename T, + typename absl::enable_if_t< + absl::conjunction< + std::is_same, + std::is_same, + std::is_same>::value, + int> = 0> + void merge(btree_container &src) { // NOLINT + for (auto src_it = src.begin(), end = src.end(); src_it != end; ++src_it) { + insert(std::move(params_type::element(src_it.slot()))); + } + src.clear(); + } + + template < + typename T, + typename absl::enable_if_t< + absl::conjunction< + std::is_same, + std::is_same, + std::is_same>::value, + int> = 0> + void merge(btree_container &&src) { + merge(src); + } +}; + +// A base class for btree_multimap. +template +class btree_multimap_container : public btree_multiset_container { + using super_type = btree_multiset_container; + using params_type = typename Tree::params_type; + friend class BtreeNodePeer; + + public: + using mapped_type = typename params_type::mapped_type; + + // Inherit constructors. + using super_type::super_type; + btree_multimap_container() {} +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_BTREE_CONTAINER_H_ diff --git a/src/absl/container/internal/common.h b/src/absl/container/internal/common.h new file mode 100644 index 0000000..416d9aa --- /dev/null +++ b/src/absl/container/internal/common.h @@ -0,0 +1,207 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_CONTAINER_H_ +#define ABSL_CONTAINER_INTERNAL_CONTAINER_H_ + +#include +#include + +#include "absl/meta/type_traits.h" +#include "absl/types/optional.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +struct IsTransparent : std::false_type {}; +template +struct IsTransparent> + : std::true_type {}; + +template +struct KeyArg { + // Transparent. Forward `K`. + template + using type = K; +}; + +template <> +struct KeyArg { + // Not transparent. Always use `key_type`. + template + using type = key_type; +}; + +// The node_handle concept from C++17. +// We specialize node_handle for sets and maps. node_handle_base holds the +// common API of both. +template +class node_handle_base { + protected: + using slot_type = typename PolicyTraits::slot_type; + + public: + using allocator_type = Alloc; + + constexpr node_handle_base() = default; + node_handle_base(node_handle_base&& other) noexcept { + *this = std::move(other); + } + ~node_handle_base() { destroy(); } + node_handle_base& operator=(node_handle_base&& other) noexcept { + destroy(); + if (!other.empty()) { + alloc_ = other.alloc_; + PolicyTraits::transfer(alloc(), slot(), other.slot()); + other.reset(); + } + return *this; + } + + bool empty() const noexcept { return !alloc_; } + explicit operator bool() const noexcept { return !empty(); } + allocator_type get_allocator() const { return *alloc_; } + + protected: + friend struct CommonAccess; + + struct transfer_tag_t {}; + node_handle_base(transfer_tag_t, const allocator_type& a, slot_type* s) + : alloc_(a) { + PolicyTraits::transfer(alloc(), slot(), s); + } + + struct construct_tag_t {}; + template + node_handle_base(construct_tag_t, const allocator_type& a, Args&&... args) + : alloc_(a) { + PolicyTraits::construct(alloc(), slot(), std::forward(args)...); + } + + void destroy() { + if (!empty()) { + PolicyTraits::destroy(alloc(), slot()); + reset(); + } + } + + void reset() { + assert(alloc_.has_value()); + alloc_ = absl::nullopt; + } + + slot_type* slot() const { + assert(!empty()); + return reinterpret_cast(std::addressof(slot_space_)); + } + allocator_type* alloc() { return std::addressof(*alloc_); } + + private: + absl::optional alloc_ = {}; + alignas(slot_type) mutable unsigned char slot_space_[sizeof(slot_type)] = {}; +}; + +// For sets. +template +class node_handle : public node_handle_base { + using Base = node_handle_base; + + public: + using value_type = typename PolicyTraits::value_type; + + constexpr node_handle() {} + + value_type& value() const { return PolicyTraits::element(this->slot()); } + + private: + friend struct CommonAccess; + + using Base::Base; +}; + +// For maps. +template +class node_handle> + : public node_handle_base { + using Base = node_handle_base; + using slot_type = typename PolicyTraits::slot_type; + + public: + using key_type = typename Policy::key_type; + using mapped_type = typename Policy::mapped_type; + + constexpr node_handle() {} + + // When C++17 is available, we can use std::launder to provide mutable + // access to the key. Otherwise, we provide const access. + auto key() const + -> decltype(PolicyTraits::mutable_key(std::declval())) { + return PolicyTraits::mutable_key(this->slot()); + } + + mapped_type& mapped() const { + return PolicyTraits::value(&PolicyTraits::element(this->slot())); + } + + private: + friend struct CommonAccess; + + using Base::Base; +}; + +// Provide access to non-public node-handle functions. +struct CommonAccess { + template + static auto GetSlot(const Node& node) -> decltype(node.slot()) { + return node.slot(); + } + + template + static void Destroy(Node* node) { + node->destroy(); + } + + template + static void Reset(Node* node) { + node->reset(); + } + + template + static T Transfer(Args&&... args) { + return T(typename T::transfer_tag_t{}, std::forward(args)...); + } + + template + static T Construct(Args&&... args) { + return T(typename T::construct_tag_t{}, std::forward(args)...); + } +}; + +// Implement the insert_return_type<> concept of C++17. +template +struct InsertReturnType { + Iterator position; + bool inserted; + NodeType node; +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_CONTAINER_H_ diff --git a/src/absl/container/internal/compressed_tuple.h b/src/absl/container/internal/compressed_tuple.h new file mode 100644 index 0000000..5ebe164 --- /dev/null +++ b/src/absl/container/internal/compressed_tuple.h @@ -0,0 +1,290 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Helper class to perform the Empty Base Optimization. +// Ts can contain classes and non-classes, empty or not. For the ones that +// are empty classes, we perform the optimization. If all types in Ts are empty +// classes, then CompressedTuple is itself an empty class. +// +// To access the members, use member get() function. +// +// Eg: +// absl::container_internal::CompressedTuple value(7, t1, t2, +// t3); +// assert(value.get<0>() == 7); +// T1& t1 = value.get<1>(); +// const T2& t2 = value.get<2>(); +// ... +// +// https://en.cppreference.com/w/cpp/language/ebo + +#ifndef ABSL_CONTAINER_INTERNAL_COMPRESSED_TUPLE_H_ +#define ABSL_CONTAINER_INTERNAL_COMPRESSED_TUPLE_H_ + +#include +#include +#include +#include + +#include "absl/utility/utility.h" + +#if defined(_MSC_VER) && !defined(__NVCC__) +// We need to mark these classes with this declspec to ensure that +// CompressedTuple happens. +#define ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC __declspec(empty_bases) +#else +#define ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +class CompressedTuple; + +namespace internal_compressed_tuple { + +template +struct Elem; +template +struct Elem, I> + : std::tuple_element> {}; +template +using ElemT = typename Elem::type; + +// Use the __is_final intrinsic if available. Where it's not available, classes +// declared with the 'final' specifier cannot be used as CompressedTuple +// elements. +// TODO(sbenza): Replace this with std::is_final in C++14. +template +constexpr bool IsFinal() { +#if defined(__clang__) || defined(__GNUC__) + return __is_final(T); +#else + return false; +#endif +} + +// We can't use EBCO on other CompressedTuples because that would mean that we +// derive from multiple Storage<> instantiations with the same I parameter, +// and potentially from multiple identical Storage<> instantiations. So anytime +// we use type inheritance rather than encapsulation, we mark +// CompressedTupleImpl, to make this easy to detect. +struct uses_inheritance {}; + +template +constexpr bool ShouldUseBase() { + return std::is_class::value && std::is_empty::value && !IsFinal() && + !std::is_base_of::value; +} + +// The storage class provides two specializations: +// - For empty classes, it stores T as a base class. +// - For everything else, it stores T as a member. +template ::type>()> +#else + bool UseBase = ShouldUseBase()> +#endif +struct Storage { + T value; + constexpr Storage() = default; + template + explicit constexpr Storage(absl::in_place_t, V&& v) + : value(absl::forward(v)) {} + constexpr const T& get() const& { return value; } + T& get() & { return value; } + constexpr const T&& get() const&& { return absl::move(*this).value; } + T&& get() && { return std::move(*this).value; } +}; + +template +struct ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC Storage : T { + constexpr Storage() = default; + + template + explicit constexpr Storage(absl::in_place_t, V&& v) + : T(absl::forward(v)) {} + + constexpr const T& get() const& { return *this; } + T& get() & { return *this; } + constexpr const T&& get() const&& { return absl::move(*this); } + T&& get() && { return std::move(*this); } +}; + +template +struct ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC CompressedTupleImpl; + +template +struct ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC CompressedTupleImpl< + CompressedTuple, absl::index_sequence, ShouldAnyUseBase> + // We use the dummy identity function through std::integral_constant to + // convince MSVC of accepting and expanding I in that context. Without it + // you would get: + // error C3548: 'I': parameter pack cannot be used in this context + : uses_inheritance, + Storage::value>... { + constexpr CompressedTupleImpl() = default; + template + explicit constexpr CompressedTupleImpl(absl::in_place_t, Vs&&... args) + : Storage(absl::in_place, absl::forward(args))... {} + friend CompressedTuple; +}; + +template +struct ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC CompressedTupleImpl< + CompressedTuple, absl::index_sequence, false> + // We use the dummy identity function as above... + : Storage::value, false>... { + constexpr CompressedTupleImpl() = default; + template + explicit constexpr CompressedTupleImpl(absl::in_place_t, Vs&&... args) + : Storage(absl::in_place, absl::forward(args))... {} + friend CompressedTuple; +}; + +std::false_type Or(std::initializer_list); +std::true_type Or(std::initializer_list); + +// MSVC requires this to be done separately rather than within the declaration +// of CompressedTuple below. +template +constexpr bool ShouldAnyUseBase() { + return decltype( + Or({std::integral_constant()>()...})){}; +} + +template +using TupleElementMoveConstructible = + typename std::conditional::value, + std::is_convertible, + std::is_constructible>::type; + +template +struct TupleMoveConstructible : std::false_type {}; + +template +struct TupleMoveConstructible, Vs...> + : std::integral_constant< + bool, absl::conjunction< + TupleElementMoveConstructible...>::value> {}; + +template +struct compressed_tuple_size; + +template +struct compressed_tuple_size> + : public std::integral_constant {}; + +template +struct TupleItemsMoveConstructible + : std::integral_constant< + bool, TupleMoveConstructible::value == + sizeof...(Vs), + T, Vs...>::value> {}; + +} // namespace internal_compressed_tuple + +// Helper class to perform the Empty Base Class Optimization. +// Ts can contain classes and non-classes, empty or not. For the ones that +// are empty classes, we perform the CompressedTuple. If all types in Ts are +// empty classes, then CompressedTuple is itself an empty class. (This +// does not apply when one or more of those empty classes is itself an empty +// CompressedTuple.) +// +// To access the members, use member .get() function. +// +// Eg: +// absl::container_internal::CompressedTuple value(7, t1, t2, +// t3); +// assert(value.get<0>() == 7); +// T1& t1 = value.get<1>(); +// const T2& t2 = value.get<2>(); +// ... +// +// https://en.cppreference.com/w/cpp/language/ebo +template +class ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC CompressedTuple + : private internal_compressed_tuple::CompressedTupleImpl< + CompressedTuple, absl::index_sequence_for, + internal_compressed_tuple::ShouldAnyUseBase()> { + private: + template + using ElemT = internal_compressed_tuple::ElemT; + + template + using StorageT = internal_compressed_tuple::Storage, I>; + + public: + // There seems to be a bug in MSVC dealing in which using '=default' here will + // cause the compiler to ignore the body of other constructors. The work- + // around is to explicitly implement the default constructor. +#if defined(_MSC_VER) + constexpr CompressedTuple() : CompressedTuple::CompressedTupleImpl() {} +#else + constexpr CompressedTuple() = default; +#endif + explicit constexpr CompressedTuple(const Ts&... base) + : CompressedTuple::CompressedTupleImpl(absl::in_place, base...) {} + + template )>>, + internal_compressed_tuple::TupleItemsMoveConstructible< + CompressedTuple, First, Vs...>>::value, + bool> = true> + explicit constexpr CompressedTuple(First&& first, Vs&&... base) + : CompressedTuple::CompressedTupleImpl(absl::in_place, + absl::forward(first), + absl::forward(base)...) {} + + template + ElemT& get() & { + return StorageT::get(); + } + + template + constexpr const ElemT& get() const& { + return StorageT::get(); + } + + template + ElemT&& get() && { + return std::move(*this).StorageT::get(); + } + + template + constexpr const ElemT&& get() const&& { + return absl::move(*this).StorageT::get(); + } +}; + +// Explicit specialization for a zero-element tuple +// (needed to avoid ambiguous overloads for the default constructor). +template <> +class ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC CompressedTuple<> {}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#undef ABSL_INTERNAL_COMPRESSED_TUPLE_DECLSPEC + +#endif // ABSL_CONTAINER_INTERNAL_COMPRESSED_TUPLE_H_ diff --git a/src/absl/container/internal/container_memory.h b/src/absl/container/internal/container_memory.h new file mode 100644 index 0000000..00e9f6d --- /dev/null +++ b/src/absl/container/internal/container_memory.h @@ -0,0 +1,442 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_CONTAINER_MEMORY_H_ +#define ABSL_CONTAINER_INTERNAL_CONTAINER_MEMORY_H_ + +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/memory/memory.h" +#include "absl/meta/type_traits.h" +#include "absl/utility/utility.h" + +#ifdef ABSL_HAVE_ADDRESS_SANITIZER +#include +#endif + +#ifdef ABSL_HAVE_MEMORY_SANITIZER +#include +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +struct alignas(Alignment) AlignedType {}; + +// Allocates at least n bytes aligned to the specified alignment. +// Alignment must be a power of 2. It must be positive. +// +// Note that many allocators don't honor alignment requirements above certain +// threshold (usually either alignof(std::max_align_t) or alignof(void*)). +// Allocate() doesn't apply alignment corrections. If the underlying allocator +// returns insufficiently alignment pointer, that's what you are going to get. +template +void* Allocate(Alloc* alloc, size_t n) { + static_assert(Alignment > 0, ""); + assert(n && "n must be positive"); + using M = AlignedType; + using A = typename absl::allocator_traits::template rebind_alloc; + using AT = typename absl::allocator_traits::template rebind_traits; + // On macOS, "mem_alloc" is a #define with one argument defined in + // rpc/types.h, so we can't name the variable "mem_alloc" and initialize it + // with the "foo(bar)" syntax. + A my_mem_alloc(*alloc); + void* p = AT::allocate(my_mem_alloc, (n + sizeof(M) - 1) / sizeof(M)); + assert(reinterpret_cast(p) % Alignment == 0 && + "allocator does not respect alignment"); + return p; +} + +// The pointer must have been previously obtained by calling +// Allocate(alloc, n). +template +void Deallocate(Alloc* alloc, void* p, size_t n) { + static_assert(Alignment > 0, ""); + assert(n && "n must be positive"); + using M = AlignedType; + using A = typename absl::allocator_traits::template rebind_alloc; + using AT = typename absl::allocator_traits::template rebind_traits; + // On macOS, "mem_alloc" is a #define with one argument defined in + // rpc/types.h, so we can't name the variable "mem_alloc" and initialize it + // with the "foo(bar)" syntax. + A my_mem_alloc(*alloc); + AT::deallocate(my_mem_alloc, static_cast(p), + (n + sizeof(M) - 1) / sizeof(M)); +} + +namespace memory_internal { + +// Constructs T into uninitialized storage pointed by `ptr` using the args +// specified in the tuple. +template +void ConstructFromTupleImpl(Alloc* alloc, T* ptr, Tuple&& t, + absl::index_sequence) { + absl::allocator_traits::construct( + *alloc, ptr, std::get(std::forward(t))...); +} + +template +struct WithConstructedImplF { + template + decltype(std::declval()(std::declval())) operator()( + Args&&... args) const { + return std::forward(f)(T(std::forward(args)...)); + } + F&& f; +}; + +template +decltype(std::declval()(std::declval())) WithConstructedImpl( + Tuple&& t, absl::index_sequence, F&& f) { + return WithConstructedImplF{std::forward(f)}( + std::get(std::forward(t))...); +} + +template +auto TupleRefImpl(T&& t, absl::index_sequence) + -> decltype(std::forward_as_tuple(std::get(std::forward(t))...)) { + return std::forward_as_tuple(std::get(std::forward(t))...); +} + +// Returns a tuple of references to the elements of the input tuple. T must be a +// tuple. +template +auto TupleRef(T&& t) -> decltype( + TupleRefImpl(std::forward(t), + absl::make_index_sequence< + std::tuple_size::type>::value>())) { + return TupleRefImpl( + std::forward(t), + absl::make_index_sequence< + std::tuple_size::type>::value>()); +} + +template +decltype(std::declval()(std::declval(), std::piecewise_construct, + std::declval>(), std::declval())) +DecomposePairImpl(F&& f, std::pair, V> p) { + const auto& key = std::get<0>(p.first); + return std::forward(f)(key, std::piecewise_construct, std::move(p.first), + std::move(p.second)); +} + +} // namespace memory_internal + +// Constructs T into uninitialized storage pointed by `ptr` using the args +// specified in the tuple. +template +void ConstructFromTuple(Alloc* alloc, T* ptr, Tuple&& t) { + memory_internal::ConstructFromTupleImpl( + alloc, ptr, std::forward(t), + absl::make_index_sequence< + std::tuple_size::type>::value>()); +} + +// Constructs T using the args specified in the tuple and calls F with the +// constructed value. +template +decltype(std::declval()(std::declval())) WithConstructed( + Tuple&& t, F&& f) { + return memory_internal::WithConstructedImpl( + std::forward(t), + absl::make_index_sequence< + std::tuple_size::type>::value>(), + std::forward(f)); +} + +// Given arguments of an std::pair's consructor, PairArgs() returns a pair of +// tuples with references to the passed arguments. The tuples contain +// constructor arguments for the first and the second elements of the pair. +// +// The following two snippets are equivalent. +// +// 1. std::pair p(args...); +// +// 2. auto a = PairArgs(args...); +// std::pair p(std::piecewise_construct, +// std::move(a.first), std::move(a.second)); +inline std::pair, std::tuple<>> PairArgs() { return {}; } +template +std::pair, std::tuple> PairArgs(F&& f, S&& s) { + return {std::piecewise_construct, std::forward_as_tuple(std::forward(f)), + std::forward_as_tuple(std::forward(s))}; +} +template +std::pair, std::tuple> PairArgs( + const std::pair& p) { + return PairArgs(p.first, p.second); +} +template +std::pair, std::tuple> PairArgs(std::pair&& p) { + return PairArgs(std::forward(p.first), std::forward(p.second)); +} +template +auto PairArgs(std::piecewise_construct_t, F&& f, S&& s) + -> decltype(std::make_pair(memory_internal::TupleRef(std::forward(f)), + memory_internal::TupleRef(std::forward(s)))) { + return std::make_pair(memory_internal::TupleRef(std::forward(f)), + memory_internal::TupleRef(std::forward(s))); +} + +// A helper function for implementing apply() in map policies. +template +auto DecomposePair(F&& f, Args&&... args) + -> decltype(memory_internal::DecomposePairImpl( + std::forward(f), PairArgs(std::forward(args)...))) { + return memory_internal::DecomposePairImpl( + std::forward(f), PairArgs(std::forward(args)...)); +} + +// A helper function for implementing apply() in set policies. +template +decltype(std::declval()(std::declval(), std::declval())) +DecomposeValue(F&& f, Arg&& arg) { + const auto& key = arg; + return std::forward(f)(key, std::forward(arg)); +} + +// Helper functions for asan and msan. +inline void SanitizerPoisonMemoryRegion(const void* m, size_t s) { +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + ASAN_POISON_MEMORY_REGION(m, s); +#endif +#ifdef ABSL_HAVE_MEMORY_SANITIZER + __msan_poison(m, s); +#endif + (void)m; + (void)s; +} + +inline void SanitizerUnpoisonMemoryRegion(const void* m, size_t s) { +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + ASAN_UNPOISON_MEMORY_REGION(m, s); +#endif +#ifdef ABSL_HAVE_MEMORY_SANITIZER + __msan_unpoison(m, s); +#endif + (void)m; + (void)s; +} + +template +inline void SanitizerPoisonObject(const T* object) { + SanitizerPoisonMemoryRegion(object, sizeof(T)); +} + +template +inline void SanitizerUnpoisonObject(const T* object) { + SanitizerUnpoisonMemoryRegion(object, sizeof(T)); +} + +namespace memory_internal { + +// If Pair is a standard-layout type, OffsetOf::kFirst and +// OffsetOf::kSecond are equivalent to offsetof(Pair, first) and +// offsetof(Pair, second) respectively. Otherwise they are -1. +// +// The purpose of OffsetOf is to avoid calling offsetof() on non-standard-layout +// type, which is non-portable. +template +struct OffsetOf { + static constexpr size_t kFirst = static_cast(-1); + static constexpr size_t kSecond = static_cast(-1); +}; + +template +struct OffsetOf::type> { + static constexpr size_t kFirst = offsetof(Pair, first); + static constexpr size_t kSecond = offsetof(Pair, second); +}; + +template +struct IsLayoutCompatible { + private: + struct Pair { + K first; + V second; + }; + + // Is P layout-compatible with Pair? + template + static constexpr bool LayoutCompatible() { + return std::is_standard_layout

() && sizeof(P) == sizeof(Pair) && + alignof(P) == alignof(Pair) && + memory_internal::OffsetOf

::kFirst == + memory_internal::OffsetOf::kFirst && + memory_internal::OffsetOf

::kSecond == + memory_internal::OffsetOf::kSecond; + } + + public: + // Whether pair and pair are layout-compatible. If they are, + // then it is safe to store them in a union and read from either. + static constexpr bool value = std::is_standard_layout() && + std::is_standard_layout() && + memory_internal::OffsetOf::kFirst == 0 && + LayoutCompatible>() && + LayoutCompatible>(); +}; + +} // namespace memory_internal + +// The internal storage type for key-value containers like flat_hash_map. +// +// It is convenient for the value_type of a flat_hash_map to be +// pair; the "const K" prevents accidental modification of the key +// when dealing with the reference returned from find() and similar methods. +// However, this creates other problems; we want to be able to emplace(K, V) +// efficiently with move operations, and similarly be able to move a +// pair in insert(). +// +// The solution is this union, which aliases the const and non-const versions +// of the pair. This also allows flat_hash_map to work, even though +// that has the same efficiency issues with move in emplace() and insert() - +// but people do it anyway. +// +// If kMutableKeys is false, only the value member can be accessed. +// +// If kMutableKeys is true, key can be accessed through all slots while value +// and mutable_value must be accessed only via INITIALIZED slots. Slots are +// created and destroyed via mutable_value so that the key can be moved later. +// +// Accessing one of the union fields while the other is active is safe as +// long as they are layout-compatible, which is guaranteed by the definition of +// kMutableKeys. For C++11, the relevant section of the standard is +// https://timsong-cpp.github.io/cppwp/n3337/class.mem#19 (9.2.19) +template +union map_slot_type { + map_slot_type() {} + ~map_slot_type() = delete; + using value_type = std::pair; + using mutable_value_type = + std::pair, absl::remove_const_t>; + + value_type value; + mutable_value_type mutable_value; + absl::remove_const_t key; +}; + +template +struct map_slot_policy { + using slot_type = map_slot_type; + using value_type = std::pair; + using mutable_value_type = std::pair; + + private: + static void emplace(slot_type* slot) { + // The construction of union doesn't do anything at runtime but it allows us + // to access its members without violating aliasing rules. + new (slot) slot_type; + } + // If pair and pair are layout-compatible, we can accept one + // or the other via slot_type. We are also free to access the key via + // slot_type::key in this case. + using kMutableKeys = memory_internal::IsLayoutCompatible; + + public: + static value_type& element(slot_type* slot) { return slot->value; } + static const value_type& element(const slot_type* slot) { + return slot->value; + } + + // When C++17 is available, we can use std::launder to provide mutable + // access to the key for use in node handle. +#if defined(__cpp_lib_launder) && __cpp_lib_launder >= 201606 + static K& mutable_key(slot_type* slot) { + // Still check for kMutableKeys so that we can avoid calling std::launder + // unless necessary because it can interfere with optimizations. + return kMutableKeys::value ? slot->key + : *std::launder(const_cast( + std::addressof(slot->value.first))); + } +#else // !(defined(__cpp_lib_launder) && __cpp_lib_launder >= 201606) + static const K& mutable_key(slot_type* slot) { return key(slot); } +#endif + + static const K& key(const slot_type* slot) { + return kMutableKeys::value ? slot->key : slot->value.first; + } + + template + static void construct(Allocator* alloc, slot_type* slot, Args&&... args) { + emplace(slot); + if (kMutableKeys::value) { + absl::allocator_traits::construct(*alloc, &slot->mutable_value, + std::forward(args)...); + } else { + absl::allocator_traits::construct(*alloc, &slot->value, + std::forward(args)...); + } + } + + // Construct this slot by moving from another slot. + template + static void construct(Allocator* alloc, slot_type* slot, slot_type* other) { + emplace(slot); + if (kMutableKeys::value) { + absl::allocator_traits::construct( + *alloc, &slot->mutable_value, std::move(other->mutable_value)); + } else { + absl::allocator_traits::construct(*alloc, &slot->value, + std::move(other->value)); + } + } + + // Construct this slot by copying from another slot. + template + static void construct(Allocator* alloc, slot_type* slot, + const slot_type* other) { + emplace(slot); + absl::allocator_traits::construct(*alloc, &slot->value, + other->value); + } + + template + static void destroy(Allocator* alloc, slot_type* slot) { + if (kMutableKeys::value) { + absl::allocator_traits::destroy(*alloc, &slot->mutable_value); + } else { + absl::allocator_traits::destroy(*alloc, &slot->value); + } + } + + template + static void transfer(Allocator* alloc, slot_type* new_slot, + slot_type* old_slot) { + emplace(new_slot); + if (kMutableKeys::value) { + absl::allocator_traits::construct( + *alloc, &new_slot->mutable_value, std::move(old_slot->mutable_value)); + } else { + absl::allocator_traits::construct(*alloc, &new_slot->value, + std::move(old_slot->value)); + } + destroy(alloc, old_slot); + } +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_CONTAINER_MEMORY_H_ diff --git a/src/absl/container/internal/counting_allocator.h b/src/absl/container/internal/counting_allocator.h new file mode 100644 index 0000000..b1a8519 --- /dev/null +++ b/src/absl/container/internal/counting_allocator.h @@ -0,0 +1,122 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_COUNTING_ALLOCATOR_H_ +#define ABSL_CONTAINER_INTERNAL_COUNTING_ALLOCATOR_H_ + +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// This is a stateful allocator, but the state lives outside of the +// allocator (in whatever test is using the allocator). This is odd +// but helps in tests where the allocator is propagated into nested +// containers - that chain of allocators uses the same state and is +// thus easier to query for aggregate allocation information. +template +class CountingAllocator { + public: + using Allocator = std::allocator; + using AllocatorTraits = std::allocator_traits; + using value_type = typename AllocatorTraits::value_type; + using pointer = typename AllocatorTraits::pointer; + using const_pointer = typename AllocatorTraits::const_pointer; + using size_type = typename AllocatorTraits::size_type; + using difference_type = typename AllocatorTraits::difference_type; + + CountingAllocator() = default; + explicit CountingAllocator(int64_t* bytes_used) : bytes_used_(bytes_used) {} + CountingAllocator(int64_t* bytes_used, int64_t* instance_count) + : bytes_used_(bytes_used), instance_count_(instance_count) {} + + template + CountingAllocator(const CountingAllocator& x) + : bytes_used_(x.bytes_used_), instance_count_(x.instance_count_) {} + + pointer allocate( + size_type n, + typename AllocatorTraits::const_void_pointer hint = nullptr) { + Allocator allocator; + pointer ptr = AllocatorTraits::allocate(allocator, n, hint); + if (bytes_used_ != nullptr) { + *bytes_used_ += n * sizeof(T); + } + return ptr; + } + + void deallocate(pointer p, size_type n) { + Allocator allocator; + AllocatorTraits::deallocate(allocator, p, n); + if (bytes_used_ != nullptr) { + *bytes_used_ -= n * sizeof(T); + } + } + + template + void construct(U* p, Args&&... args) { + Allocator allocator; + AllocatorTraits::construct(allocator, p, std::forward(args)...); + if (instance_count_ != nullptr) { + *instance_count_ += 1; + } + } + + template + void destroy(U* p) { + Allocator allocator; + // Ignore GCC warning bug. +#if ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(12, 0) +// #pragma GCC diagnostic push +// #pragma GCC diagnostic ignored "-Wuse-after-free" +#endif + AllocatorTraits::destroy(allocator, p); +#if ABSL_INTERNAL_HAVE_MIN_GNUC_VERSION(12, 0) +// #pragma GCC diagnostic pop +#endif + if (instance_count_ != nullptr) { + *instance_count_ -= 1; + } + } + + template + class rebind { + public: + using other = CountingAllocator; + }; + + friend bool operator==(const CountingAllocator& a, + const CountingAllocator& b) { + return a.bytes_used_ == b.bytes_used_ && + a.instance_count_ == b.instance_count_; + } + + friend bool operator!=(const CountingAllocator& a, + const CountingAllocator& b) { + return !(a == b); + } + + int64_t* bytes_used_ = nullptr; + int64_t* instance_count_ = nullptr; +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_COUNTING_ALLOCATOR_H_ diff --git a/src/absl/container/internal/hash_function_defaults.h b/src/absl/container/internal/hash_function_defaults.h new file mode 100644 index 0000000..250e662 --- /dev/null +++ b/src/absl/container/internal/hash_function_defaults.h @@ -0,0 +1,163 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Define the default Hash and Eq functions for SwissTable containers. +// +// std::hash and std::equal_to are not appropriate hash and equal +// functions for SwissTable containers. There are two reasons for this. +// +// SwissTable containers are power of 2 sized containers: +// +// This means they use the lower bits of the hash value to find the slot for +// each entry. The typical hash function for integral types is the identity. +// This is a very weak hash function for SwissTable and any power of 2 sized +// hashtable implementation which will lead to excessive collisions. For +// SwissTable we use murmur3 style mixing to reduce collisions to a minimum. +// +// SwissTable containers support heterogeneous lookup: +// +// In order to make heterogeneous lookup work, hash and equal functions must be +// polymorphic. At the same time they have to satisfy the same requirements the +// C++ standard imposes on hash functions and equality operators. That is: +// +// if hash_default_eq(a, b) returns true for any a and b of type T, then +// hash_default_hash(a) must equal hash_default_hash(b) +// +// For SwissTable containers this requirement is relaxed to allow a and b of +// any and possibly different types. Note that like the standard the hash and +// equal functions are still bound to T. This is important because some type U +// can be hashed by/tested for equality differently depending on T. A notable +// example is `const char*`. `const char*` is treated as a c-style string when +// the hash function is hash but as a pointer when the hash +// function is hash. +// +#ifndef ABSL_CONTAINER_INTERNAL_HASH_FUNCTION_DEFAULTS_H_ +#define ABSL_CONTAINER_INTERNAL_HASH_FUNCTION_DEFAULTS_H_ + +#include +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/hash/hash.h" +#include "absl/strings/cord.h" +#include "absl/strings/string_view.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// The hash of an object of type T is computed by using absl::Hash. +template +struct HashEq { + using Hash = absl::Hash; + using Eq = std::equal_to; +}; + +struct StringHash { + using is_transparent = void; + + size_t operator()(absl::string_view v) const { + return absl::Hash{}(v); + } + size_t operator()(const absl::Cord& v) const { + return absl::Hash{}(v); + } +}; + +struct StringEq { + using is_transparent = void; + bool operator()(absl::string_view lhs, absl::string_view rhs) const { + return lhs == rhs; + } + bool operator()(const absl::Cord& lhs, const absl::Cord& rhs) const { + return lhs == rhs; + } + bool operator()(const absl::Cord& lhs, absl::string_view rhs) const { + return lhs == rhs; + } + bool operator()(absl::string_view lhs, const absl::Cord& rhs) const { + return lhs == rhs; + } +}; + +// Supports heterogeneous lookup for string-like elements. +struct StringHashEq { + using Hash = StringHash; + using Eq = StringEq; +}; + +template <> +struct HashEq : StringHashEq {}; +template <> +struct HashEq : StringHashEq {}; +template <> +struct HashEq : StringHashEq {}; + +// Supports heterogeneous lookup for pointers and smart pointers. +template +struct HashEq { + struct Hash { + using is_transparent = void; + template + size_t operator()(const U& ptr) const { + return absl::Hash{}(HashEq::ToPtr(ptr)); + } + }; + struct Eq { + using is_transparent = void; + template + bool operator()(const A& a, const B& b) const { + return HashEq::ToPtr(a) == HashEq::ToPtr(b); + } + }; + + private: + static const T* ToPtr(const T* ptr) { return ptr; } + template + static const T* ToPtr(const std::unique_ptr& ptr) { + return ptr.get(); + } + template + static const T* ToPtr(const std::shared_ptr& ptr) { + return ptr.get(); + } +}; + +template +struct HashEq> : HashEq {}; +template +struct HashEq> : HashEq {}; + +// This header's visibility is restricted. If you need to access the default +// hasher please use the container's ::hasher alias instead. +// +// Example: typename Hash = typename absl::flat_hash_map::hasher +template +using hash_default_hash = typename container_internal::HashEq::Hash; + +// This header's visibility is restricted. If you need to access the default +// key equal please use the container's ::key_equal alias instead. +// +// Example: typename Eq = typename absl::flat_hash_map::key_equal +template +using hash_default_eq = typename container_internal::HashEq::Eq; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_HASH_FUNCTION_DEFAULTS_H_ diff --git a/src/absl/container/internal/hash_policy_traits.h b/src/absl/container/internal/hash_policy_traits.h new file mode 100644 index 0000000..46c97b1 --- /dev/null +++ b/src/absl/container/internal/hash_policy_traits.h @@ -0,0 +1,208 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_HASH_POLICY_TRAITS_H_ +#define ABSL_CONTAINER_INTERNAL_HASH_POLICY_TRAITS_H_ + +#include +#include +#include +#include +#include + +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// Defines how slots are initialized/destroyed/moved. +template +struct hash_policy_traits { + // The type of the keys stored in the hashtable. + using key_type = typename Policy::key_type; + + private: + struct ReturnKey { + // When C++17 is available, we can use std::launder to provide mutable + // access to the key for use in node handle. +#if defined(__cpp_lib_launder) && __cpp_lib_launder >= 201606 + template ::value, int> = 0> + static key_type& Impl(Key&& k, int) { + return *std::launder( + const_cast(std::addressof(std::forward(k)))); + } +#endif + + template + static Key Impl(Key&& k, char) { + return std::forward(k); + } + + // When Key=T&, we forward the lvalue reference. + // When Key=T, we return by value to avoid a dangling reference. + // eg, for string_hash_map. + template + auto operator()(Key&& k, const Args&...) const + -> decltype(Impl(std::forward(k), 0)) { + return Impl(std::forward(k), 0); + } + }; + + template + struct ConstantIteratorsImpl : std::false_type {}; + + template + struct ConstantIteratorsImpl> + : P::constant_iterators {}; + + public: + // The actual object stored in the hash table. + using slot_type = typename Policy::slot_type; + + // The argument type for insertions into the hashtable. This is different + // from value_type for increased performance. See initializer_list constructor + // and insert() member functions for more details. + using init_type = typename Policy::init_type; + + using reference = decltype(Policy::element(std::declval())); + using pointer = typename std::remove_reference::type*; + using value_type = typename std::remove_reference::type; + + // Policies can set this variable to tell raw_hash_set that all iterators + // should be constant, even `iterator`. This is useful for set-like + // containers. + // Defaults to false if not provided by the policy. + using constant_iterators = ConstantIteratorsImpl<>; + + // PRECONDITION: `slot` is UNINITIALIZED + // POSTCONDITION: `slot` is INITIALIZED + template + static void construct(Alloc* alloc, slot_type* slot, Args&&... args) { + Policy::construct(alloc, slot, std::forward(args)...); + } + + // PRECONDITION: `slot` is INITIALIZED + // POSTCONDITION: `slot` is UNINITIALIZED + template + static void destroy(Alloc* alloc, slot_type* slot) { + Policy::destroy(alloc, slot); + } + + // Transfers the `old_slot` to `new_slot`. Any memory allocated by the + // allocator inside `old_slot` to `new_slot` can be transferred. + // + // OPTIONAL: defaults to: + // + // clone(new_slot, std::move(*old_slot)); + // destroy(old_slot); + // + // PRECONDITION: `new_slot` is UNINITIALIZED and `old_slot` is INITIALIZED + // POSTCONDITION: `new_slot` is INITIALIZED and `old_slot` is + // UNINITIALIZED + template + static void transfer(Alloc* alloc, slot_type* new_slot, slot_type* old_slot) { + transfer_impl(alloc, new_slot, old_slot, 0); + } + + // PRECONDITION: `slot` is INITIALIZED + // POSTCONDITION: `slot` is INITIALIZED + template + static auto element(slot_type* slot) -> decltype(P::element(slot)) { + return P::element(slot); + } + + // Returns the amount of memory owned by `slot`, exclusive of `sizeof(*slot)`. + // + // If `slot` is nullptr, returns the constant amount of memory owned by any + // full slot or -1 if slots own variable amounts of memory. + // + // PRECONDITION: `slot` is INITIALIZED or nullptr + template + static size_t space_used(const slot_type* slot) { + return P::space_used(slot); + } + + // Provides generalized access to the key for elements, both for elements in + // the table and for elements that have not yet been inserted (or even + // constructed). We would like an API that allows us to say: `key(args...)` + // but we cannot do that for all cases, so we use this more general API that + // can be used for many things, including the following: + // + // - Given an element in a table, get its key. + // - Given an element initializer, get its key. + // - Given `emplace()` arguments, get the element key. + // + // Implementations of this must adhere to a very strict technical + // specification around aliasing and consuming arguments: + // + // Let `value_type` be the result type of `element()` without ref- and + // cv-qualifiers. The first argument is a functor, the rest are constructor + // arguments for `value_type`. Returns `std::forward(f)(k, xs...)`, where + // `k` is the element key, and `xs...` are the new constructor arguments for + // `value_type`. It's allowed for `k` to alias `xs...`, and for both to alias + // `ts...`. The key won't be touched once `xs...` are used to construct an + // element; `ts...` won't be touched at all, which allows `apply()` to consume + // any rvalues among them. + // + // If `value_type` is constructible from `Ts&&...`, `Policy::apply()` must not + // trigger a hard compile error unless it originates from `f`. In other words, + // `Policy::apply()` must be SFINAE-friendly. If `value_type` is not + // constructible from `Ts&&...`, either SFINAE or a hard compile error is OK. + // + // If `Ts...` is `[cv] value_type[&]` or `[cv] init_type[&]`, + // `Policy::apply()` must work. A compile error is not allowed, SFINAE or not. + template + static auto apply(F&& f, Ts&&... ts) + -> decltype(P::apply(std::forward(f), std::forward(ts)...)) { + return P::apply(std::forward(f), std::forward(ts)...); + } + + // Returns the "key" portion of the slot. + // Used for node handle manipulation. + template + static auto mutable_key(slot_type* slot) + -> decltype(P::apply(ReturnKey(), element(slot))) { + return P::apply(ReturnKey(), element(slot)); + } + + // Returns the "value" (as opposed to the "key") portion of the element. Used + // by maps to implement `operator[]`, `at()` and `insert_or_assign()`. + template + static auto value(T* elem) -> decltype(P::value(elem)) { + return P::value(elem); + } + + private: + // Use auto -> decltype as an enabler. + template + static auto transfer_impl(Alloc* alloc, slot_type* new_slot, + slot_type* old_slot, int) + -> decltype((void)P::transfer(alloc, new_slot, old_slot)) { + P::transfer(alloc, new_slot, old_slot); + } + template + static void transfer_impl(Alloc* alloc, slot_type* new_slot, + slot_type* old_slot, char) { + construct(alloc, new_slot, std::move(element(old_slot))); + destroy(alloc, old_slot); + } +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_HASH_POLICY_TRAITS_H_ diff --git a/src/absl/container/internal/hashtable_debug.h b/src/absl/container/internal/hashtable_debug.h new file mode 100644 index 0000000..19d5212 --- /dev/null +++ b/src/absl/container/internal/hashtable_debug.h @@ -0,0 +1,110 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This library provides APIs to debug the probing behavior of hash tables. +// +// In general, the probing behavior is a black box for users and only the +// side effects can be measured in the form of performance differences. +// These APIs give a glimpse on the actual behavior of the probing algorithms in +// these hashtables given a specified hash function and a set of elements. +// +// The probe count distribution can be used to assess the quality of the hash +// function for that particular hash table. Note that a hash function that +// performs well in one hash table implementation does not necessarily performs +// well in a different one. +// +// This library supports std::unordered_{set,map}, dense_hash_{set,map} and +// absl::{flat,node,string}_hash_{set,map}. + +#ifndef ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_H_ +#define ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_H_ + +#include +#include +#include +#include + +#include "absl/container/internal/hashtable_debug_hooks.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// Returns the number of probes required to lookup `key`. Returns 0 for a +// search with no collisions. Higher values mean more hash collisions occurred; +// however, the exact meaning of this number varies according to the container +// type. +template +size_t GetHashtableDebugNumProbes( + const C& c, const typename C::key_type& key) { + return absl::container_internal::hashtable_debug_internal:: + HashtableDebugAccess::GetNumProbes(c, key); +} + +// Gets a histogram of the number of probes for each elements in the container. +// The sum of all the values in the vector is equal to container.size(). +template +std::vector GetHashtableDebugNumProbesHistogram(const C& container) { + std::vector v; + for (auto it = container.begin(); it != container.end(); ++it) { + size_t num_probes = GetHashtableDebugNumProbes( + container, + absl::container_internal::hashtable_debug_internal::GetKey(*it, 0)); + v.resize((std::max)(v.size(), num_probes + 1)); + v[num_probes]++; + } + return v; +} + +struct HashtableDebugProbeSummary { + size_t total_elements; + size_t total_num_probes; + double mean; +}; + +// Gets a summary of the probe count distribution for the elements in the +// container. +template +HashtableDebugProbeSummary GetHashtableDebugProbeSummary(const C& container) { + auto probes = GetHashtableDebugNumProbesHistogram(container); + HashtableDebugProbeSummary summary = {}; + for (size_t i = 0; i < probes.size(); ++i) { + summary.total_elements += probes[i]; + summary.total_num_probes += probes[i] * i; + } + summary.mean = 1.0 * summary.total_num_probes / summary.total_elements; + return summary; +} + +// Returns the number of bytes requested from the allocator by the container +// and not freed. +template +size_t AllocatedByteSize(const C& c) { + return absl::container_internal::hashtable_debug_internal:: + HashtableDebugAccess::AllocatedByteSize(c); +} + +// Returns a tight lower bound for AllocatedByteSize(c) where `c` is of type `C` +// and `c.size()` is equal to `num_elements`. +template +size_t LowerBoundAllocatedByteSize(size_t num_elements) { + return absl::container_internal::hashtable_debug_internal:: + HashtableDebugAccess::LowerBoundAllocatedByteSize(num_elements); +} + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_H_ diff --git a/src/absl/container/internal/hashtable_debug_hooks.h b/src/absl/container/internal/hashtable_debug_hooks.h new file mode 100644 index 0000000..3e9ea59 --- /dev/null +++ b/src/absl/container/internal/hashtable_debug_hooks.h @@ -0,0 +1,85 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Provides the internal API for hashtable_debug.h. + +#ifndef ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_HOOKS_H_ +#define ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_HOOKS_H_ + +#include + +#include +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { +namespace hashtable_debug_internal { + +// If it is a map, call get<0>(). +using std::get; +template +auto GetKey(const typename T::value_type& pair, int) -> decltype(get<0>(pair)) { + return get<0>(pair); +} + +// If it is not a map, return the value directly. +template +const typename T::key_type& GetKey(const typename T::key_type& key, char) { + return key; +} + +// Containers should specialize this to provide debug information for that +// container. +template +struct HashtableDebugAccess { + // Returns the number of probes required to find `key` in `c`. The "number of + // probes" is a concept that can vary by container. Implementations should + // return 0 when `key` was found in the minimum number of operations and + // should increment the result for each non-trivial operation required to find + // `key`. + // + // The default implementation uses the bucket api from the standard and thus + // works for `std::unordered_*` containers. + static size_t GetNumProbes(const Container& c, + const typename Container::key_type& key) { + if (!c.bucket_count()) return {}; + size_t num_probes = 0; + size_t bucket = c.bucket(key); + for (auto it = c.begin(bucket), e = c.end(bucket);; ++it, ++num_probes) { + if (it == e) return num_probes; + if (c.key_eq()(key, GetKey(*it, 0))) return num_probes; + } + } + + // Returns the number of bytes requested from the allocator by the container + // and not freed. + // + // static size_t AllocatedByteSize(const Container& c); + + // Returns a tight lower bound for AllocatedByteSize(c) where `c` is of type + // `Container` and `c.size()` is equal to `num_elements`. + // + // static size_t LowerBoundAllocatedByteSize(size_t num_elements); +}; + +} // namespace hashtable_debug_internal +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_HASHTABLE_DEBUG_HOOKS_H_ diff --git a/src/absl/container/internal/hashtablez_sampler.cc b/src/absl/container/internal/hashtablez_sampler.cc new file mode 100644 index 0000000..efc1be5 --- /dev/null +++ b/src/absl/container/internal/hashtablez_sampler.cc @@ -0,0 +1,238 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/container/internal/hashtablez_sampler.h" + +#include +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/config.h" +#include "absl/debugging/stacktrace.h" +#include "absl/memory/memory.h" +#include "absl/profiling/internal/exponential_biased.h" +#include "absl/profiling/internal/sample_recorder.h" +#include "absl/synchronization/mutex.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +constexpr int HashtablezInfo::kMaxStackDepth; +#endif + +namespace { +ABSL_CONST_INIT std::atomic g_hashtablez_enabled{ + false +}; +ABSL_CONST_INIT std::atomic g_hashtablez_sample_parameter{1 << 10}; +std::atomic g_hashtablez_config_listener{nullptr}; + +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) +ABSL_PER_THREAD_TLS_KEYWORD absl::profiling_internal::ExponentialBiased + g_exponential_biased_generator; +#endif + +void TriggerHashtablezConfigListener() { + auto* listener = g_hashtablez_config_listener.load(std::memory_order_acquire); + if (listener != nullptr) listener(); +} + +} // namespace + +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) +ABSL_PER_THREAD_TLS_KEYWORD SamplingState global_next_sample = {0, 0}; +#endif // defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + +HashtablezSampler& GlobalHashtablezSampler() { + static auto* sampler = new HashtablezSampler(); + return *sampler; +} + +HashtablezInfo::HashtablezInfo() = default; +HashtablezInfo::~HashtablezInfo() = default; + +void HashtablezInfo::PrepareForSampling(int64_t stride, + size_t inline_element_size_value) { + capacity.store(0, std::memory_order_relaxed); + size.store(0, std::memory_order_relaxed); + num_erases.store(0, std::memory_order_relaxed); + num_rehashes.store(0, std::memory_order_relaxed); + max_probe_length.store(0, std::memory_order_relaxed); + total_probe_length.store(0, std::memory_order_relaxed); + hashes_bitwise_or.store(0, std::memory_order_relaxed); + hashes_bitwise_and.store(~size_t{}, std::memory_order_relaxed); + hashes_bitwise_xor.store(0, std::memory_order_relaxed); + max_reserve.store(0, std::memory_order_relaxed); + + create_time = absl::Now(); + weight = stride; + // The inliner makes hardcoded skip_count difficult (especially when combined + // with LTO). We use the ability to exclude stacks by regex when encoding + // instead. + depth = absl::GetStackTrace(stack, HashtablezInfo::kMaxStackDepth, + /* skip_count= */ 0); + inline_element_size = inline_element_size_value; +} + +static bool ShouldForceSampling() { + enum ForceState { + kDontForce, + kForce, + kUninitialized + }; + ABSL_CONST_INIT static std::atomic global_state{ + kUninitialized}; + ForceState state = global_state.load(std::memory_order_relaxed); + if (ABSL_PREDICT_TRUE(state == kDontForce)) return false; + + if (state == kUninitialized) { + state = ABSL_INTERNAL_C_SYMBOL(AbslContainerInternalSampleEverything)() + ? kForce + : kDontForce; + global_state.store(state, std::memory_order_relaxed); + } + return state == kForce; +} + +HashtablezInfo* SampleSlow(SamplingState& next_sample, + size_t inline_element_size) { + if (ABSL_PREDICT_FALSE(ShouldForceSampling())) { + next_sample.next_sample = 1; + const int64_t old_stride = exchange(next_sample.sample_stride, 1); + HashtablezInfo* result = + GlobalHashtablezSampler().Register(old_stride, inline_element_size); + return result; + } + +#if !defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + next_sample = { + std::numeric_limits::max(), + std::numeric_limits::max(), + }; + return nullptr; +#else + bool first = next_sample.next_sample < 0; + + const int64_t next_stride = g_exponential_biased_generator.GetStride( + g_hashtablez_sample_parameter.load(std::memory_order_relaxed)); + + next_sample.next_sample = next_stride; + const int64_t old_stride = exchange(next_sample.sample_stride, next_stride); + // Small values of interval are equivalent to just sampling next time. + ABSL_ASSERT(next_stride >= 1); + + // g_hashtablez_enabled can be dynamically flipped, we need to set a threshold + // low enough that we will start sampling in a reasonable time, so we just use + // the default sampling rate. + if (!g_hashtablez_enabled.load(std::memory_order_relaxed)) return nullptr; + + // We will only be negative on our first count, so we should just retry in + // that case. + if (first) { + if (ABSL_PREDICT_TRUE(--next_sample.next_sample > 0)) return nullptr; + return SampleSlow(next_sample, inline_element_size); + } + + return GlobalHashtablezSampler().Register(old_stride, inline_element_size); +#endif +} + +void UnsampleSlow(HashtablezInfo* info) { + GlobalHashtablezSampler().Unregister(info); +} + +void RecordInsertSlow(HashtablezInfo* info, size_t hash, + size_t distance_from_desired) { + // SwissTables probe in groups of 16, so scale this to count items probes and + // not offset from desired. + size_t probe_length = distance_from_desired; +#ifdef ABSL_INTERNAL_HAVE_SSE2 + probe_length /= 16; +#else + probe_length /= 8; +#endif + + info->hashes_bitwise_and.fetch_and(hash, std::memory_order_relaxed); + info->hashes_bitwise_or.fetch_or(hash, std::memory_order_relaxed); + info->hashes_bitwise_xor.fetch_xor(hash, std::memory_order_relaxed); + info->max_probe_length.store( + std::max(info->max_probe_length.load(std::memory_order_relaxed), + probe_length), + std::memory_order_relaxed); + info->total_probe_length.fetch_add(probe_length, std::memory_order_relaxed); + info->size.fetch_add(1, std::memory_order_relaxed); +} + +void SetHashtablezConfigListener(HashtablezConfigListener l) { + g_hashtablez_config_listener.store(l, std::memory_order_release); +} + +bool IsHashtablezEnabled() { + return g_hashtablez_enabled.load(std::memory_order_acquire); +} + +void SetHashtablezEnabled(bool enabled) { + SetHashtablezEnabledInternal(enabled); + TriggerHashtablezConfigListener(); +} + +void SetHashtablezEnabledInternal(bool enabled) { + g_hashtablez_enabled.store(enabled, std::memory_order_release); +} + +int32_t GetHashtablezSampleParameter() { + return g_hashtablez_sample_parameter.load(std::memory_order_acquire); +} + +void SetHashtablezSampleParameter(int32_t rate) { + SetHashtablezSampleParameterInternal(rate); + TriggerHashtablezConfigListener(); +} + +void SetHashtablezSampleParameterInternal(int32_t rate) { + if (rate > 0) { + g_hashtablez_sample_parameter.store(rate, std::memory_order_release); + } else { + ABSL_RAW_LOG(ERROR, "Invalid hashtablez sample rate: %lld", + static_cast(rate)); // NOLINT(runtime/int) + } +} + +int32_t GetHashtablezMaxSamples() { + return GlobalHashtablezSampler().GetMaxSamples(); +} + +void SetHashtablezMaxSamples(int32_t max) { + SetHashtablezMaxSamplesInternal(max); + TriggerHashtablezConfigListener(); +} + +void SetHashtablezMaxSamplesInternal(int32_t max) { + if (max > 0) { + GlobalHashtablezSampler().SetMaxSamples(max); + } else { + ABSL_RAW_LOG(ERROR, "Invalid hashtablez max samples: %lld", + static_cast(max)); // NOLINT(runtime/int) + } +} + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/container/internal/hashtablez_sampler.h b/src/absl/container/internal/hashtablez_sampler.h new file mode 100644 index 0000000..d4016d8 --- /dev/null +++ b/src/absl/container/internal/hashtablez_sampler.h @@ -0,0 +1,299 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: hashtablez_sampler.h +// ----------------------------------------------------------------------------- +// +// This header file defines the API for a low level library to sample hashtables +// and collect runtime statistics about them. +// +// `HashtablezSampler` controls the lifecycle of `HashtablezInfo` objects which +// store information about a single sample. +// +// `Record*` methods store information into samples. +// `Sample()` and `Unsample()` make use of a single global sampler with +// properties controlled by the flags hashtablez_enabled, +// hashtablez_sample_rate, and hashtablez_max_samples. +// +// WARNING +// +// Using this sampling API may cause sampled Swiss tables to use the global +// allocator (operator `new`) in addition to any custom allocator. If you +// are using a table in an unusual circumstance where allocation or calling a +// linux syscall is unacceptable, this could interfere. +// +// This utility is internal-only. Use at your own risk. + +#ifndef ABSL_CONTAINER_INTERNAL_HASHTABLEZ_SAMPLER_H_ +#define ABSL_CONTAINER_INTERNAL_HASHTABLEZ_SAMPLER_H_ + +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/base/internal/per_thread_tls.h" +#include "absl/base/optimization.h" +#include "absl/profiling/internal/sample_recorder.h" +#include "absl/synchronization/mutex.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// Stores information about a sampled hashtable. All mutations to this *must* +// be made through `Record*` functions below. All reads from this *must* only +// occur in the callback to `HashtablezSampler::Iterate`. +struct HashtablezInfo : public profiling_internal::Sample { + // Constructs the object but does not fill in any fields. + HashtablezInfo(); + ~HashtablezInfo(); + HashtablezInfo(const HashtablezInfo&) = delete; + HashtablezInfo& operator=(const HashtablezInfo&) = delete; + + // Puts the object into a clean state, fills in the logically `const` members, + // blocking for any readers that are currently sampling the object. + void PrepareForSampling(int64_t stride, size_t inline_element_size_value) + ABSL_EXCLUSIVE_LOCKS_REQUIRED(init_mu); + + // These fields are mutated by the various Record* APIs and need to be + // thread-safe. + std::atomic capacity; + std::atomic size; + std::atomic num_erases; + std::atomic num_rehashes; + std::atomic max_probe_length; + std::atomic total_probe_length; + std::atomic hashes_bitwise_or; + std::atomic hashes_bitwise_and; + std::atomic hashes_bitwise_xor; + std::atomic max_reserve; + + // All of the fields below are set by `PrepareForSampling`, they must not be + // mutated in `Record*` functions. They are logically `const` in that sense. + // These are guarded by init_mu, but that is not externalized to clients, + // which can read them only during `SampleRecorder::Iterate` which will hold + // the lock. + static constexpr int kMaxStackDepth = 64; + absl::Time create_time; + int32_t depth; + void* stack[kMaxStackDepth]; + size_t inline_element_size; // How big is the slot? +}; + +inline void RecordRehashSlow(HashtablezInfo* info, size_t total_probe_length) { +#ifdef ABSL_INTERNAL_HAVE_SSE2 + total_probe_length /= 16; +#else + total_probe_length /= 8; +#endif + info->total_probe_length.store(total_probe_length, std::memory_order_relaxed); + info->num_erases.store(0, std::memory_order_relaxed); + // There is only one concurrent writer, so `load` then `store` is sufficient + // instead of using `fetch_add`. + info->num_rehashes.store( + 1 + info->num_rehashes.load(std::memory_order_relaxed), + std::memory_order_relaxed); +} + +inline void RecordReservationSlow(HashtablezInfo* info, + size_t target_capacity) { + info->max_reserve.store( + (std::max)(info->max_reserve.load(std::memory_order_relaxed), + target_capacity), + std::memory_order_relaxed); +} + +inline void RecordClearedReservationSlow(HashtablezInfo* info) { + info->max_reserve.store(0, std::memory_order_relaxed); +} + +inline void RecordStorageChangedSlow(HashtablezInfo* info, size_t size, + size_t capacity) { + info->size.store(size, std::memory_order_relaxed); + info->capacity.store(capacity, std::memory_order_relaxed); + if (size == 0) { + // This is a clear, reset the total/num_erases too. + info->total_probe_length.store(0, std::memory_order_relaxed); + info->num_erases.store(0, std::memory_order_relaxed); + } +} + +void RecordInsertSlow(HashtablezInfo* info, size_t hash, + size_t distance_from_desired); + +inline void RecordEraseSlow(HashtablezInfo* info) { + info->size.fetch_sub(1, std::memory_order_relaxed); + // There is only one concurrent writer, so `load` then `store` is sufficient + // instead of using `fetch_add`. + info->num_erases.store( + 1 + info->num_erases.load(std::memory_order_relaxed), + std::memory_order_relaxed); +} + +struct SamplingState { + int64_t next_sample; + // When we make a sampling decision, we record that distance so we can weight + // each sample. + int64_t sample_stride; +}; + +HashtablezInfo* SampleSlow(SamplingState& next_sample, + size_t inline_element_size); +void UnsampleSlow(HashtablezInfo* info); + +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) +#error ABSL_INTERNAL_HASHTABLEZ_SAMPLE cannot be directly set +#endif // defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) +class HashtablezInfoHandle { + public: + explicit HashtablezInfoHandle() : info_(nullptr) {} + explicit HashtablezInfoHandle(HashtablezInfo* info) : info_(info) {} + ~HashtablezInfoHandle() { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + UnsampleSlow(info_); + } + + HashtablezInfoHandle(const HashtablezInfoHandle&) = delete; + HashtablezInfoHandle& operator=(const HashtablezInfoHandle&) = delete; + + HashtablezInfoHandle(HashtablezInfoHandle&& o) noexcept + : info_(absl::exchange(o.info_, nullptr)) {} + HashtablezInfoHandle& operator=(HashtablezInfoHandle&& o) noexcept { + if (ABSL_PREDICT_FALSE(info_ != nullptr)) { + UnsampleSlow(info_); + } + info_ = absl::exchange(o.info_, nullptr); + return *this; + } + + inline void RecordStorageChanged(size_t size, size_t capacity) { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordStorageChangedSlow(info_, size, capacity); + } + + inline void RecordRehash(size_t total_probe_length) { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordRehashSlow(info_, total_probe_length); + } + + inline void RecordReservation(size_t target_capacity) { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordReservationSlow(info_, target_capacity); + } + + inline void RecordClearedReservation() { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordClearedReservationSlow(info_); + } + + inline void RecordInsert(size_t hash, size_t distance_from_desired) { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordInsertSlow(info_, hash, distance_from_desired); + } + + inline void RecordErase() { + if (ABSL_PREDICT_TRUE(info_ == nullptr)) return; + RecordEraseSlow(info_); + } + + friend inline void swap(HashtablezInfoHandle& lhs, + HashtablezInfoHandle& rhs) { + std::swap(lhs.info_, rhs.info_); + } + + private: + friend class HashtablezInfoHandlePeer; + HashtablezInfo* info_; +}; +#else +// Ensure that when Hashtablez is turned off at compile time, HashtablezInfo can +// be removed by the linker, in order to reduce the binary size. +class HashtablezInfoHandle { + public: + explicit HashtablezInfoHandle() = default; + explicit HashtablezInfoHandle(std::nullptr_t) {} + + inline void RecordStorageChanged(size_t /*size*/, size_t /*capacity*/) {} + inline void RecordRehash(size_t /*total_probe_length*/) {} + inline void RecordReservation(size_t /*target_capacity*/) {} + inline void RecordClearedReservation() {} + inline void RecordInsert(size_t /*hash*/, size_t /*distance_from_desired*/) {} + inline void RecordErase() {} + + friend inline void swap(HashtablezInfoHandle& /*lhs*/, + HashtablezInfoHandle& /*rhs*/) {} +}; +#endif // defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) +extern ABSL_PER_THREAD_TLS_KEYWORD SamplingState global_next_sample; +#endif // defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + +// Returns an RAII sampling handle that manages registration and unregistation +// with the global sampler. +inline HashtablezInfoHandle Sample( + size_t inline_element_size ABSL_ATTRIBUTE_UNUSED) { +#if defined(ABSL_INTERNAL_HASHTABLEZ_SAMPLE) + if (ABSL_PREDICT_TRUE(--global_next_sample.next_sample > 0)) { + return HashtablezInfoHandle(nullptr); + } + return HashtablezInfoHandle( + SampleSlow(global_next_sample, inline_element_size)); +#else + return HashtablezInfoHandle(nullptr); +#endif // !ABSL_PER_THREAD_TLS +} + +using HashtablezSampler = + ::absl::profiling_internal::SampleRecorder; + +// Returns a global Sampler. +HashtablezSampler& GlobalHashtablezSampler(); + +using HashtablezConfigListener = void (*)(); +void SetHashtablezConfigListener(HashtablezConfigListener l); + +// Enables or disables sampling for Swiss tables. +bool IsHashtablezEnabled(); +void SetHashtablezEnabled(bool enabled); +void SetHashtablezEnabledInternal(bool enabled); + +// Sets the rate at which Swiss tables will be sampled. +int32_t GetHashtablezSampleParameter(); +void SetHashtablezSampleParameter(int32_t rate); +void SetHashtablezSampleParameterInternal(int32_t rate); + +// Sets a soft max for the number of samples that will be kept. +int32_t GetHashtablezMaxSamples(); +void SetHashtablezMaxSamples(int32_t max); +void SetHashtablezMaxSamplesInternal(int32_t max); + +// Configuration override. +// This allows process-wide sampling without depending on order of +// initialization of static storage duration objects. +// The definition of this constant is weak, which allows us to inject a +// different value for it at link time. +extern "C" bool ABSL_INTERNAL_C_SYMBOL(AbslContainerInternalSampleEverything)(); + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_HASHTABLEZ_SAMPLER_H_ diff --git a/src/absl/container/internal/hashtablez_sampler_force_weak_definition.cc b/src/absl/container/internal/hashtablez_sampler_force_weak_definition.cc new file mode 100644 index 0000000..ed35a7e --- /dev/null +++ b/src/absl/container/internal/hashtablez_sampler_force_weak_definition.cc @@ -0,0 +1,31 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/container/internal/hashtablez_sampler.h" + +#include "absl/base/attributes.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// See hashtablez_sampler.h for details. +extern "C" ABSL_ATTRIBUTE_WEAK bool ABSL_INTERNAL_C_SYMBOL( + AbslContainerInternalSampleEverything)() { + return false; +} + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/container/internal/inlined_vector.h b/src/absl/container/internal/inlined_vector.h new file mode 100644 index 0000000..6d49415 --- /dev/null +++ b/src/absl/container/internal/inlined_vector.h @@ -0,0 +1,953 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_INLINED_VECTOR_INTERNAL_H_ +#define ABSL_CONTAINER_INTERNAL_INLINED_VECTOR_INTERNAL_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/macros.h" +#include "absl/container/internal/compressed_tuple.h" +#include "absl/memory/memory.h" +#include "absl/meta/type_traits.h" +#include "absl/types/span.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace inlined_vector_internal { + +// GCC does not deal very well with the below code +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic push +// #pragma GCC diagnostic ignored "-Warray-bounds" +#endif + +template +using AllocatorTraits = std::allocator_traits; +template +using ValueType = typename AllocatorTraits::value_type; +template +using SizeType = typename AllocatorTraits::size_type; +template +using Pointer = typename AllocatorTraits::pointer; +template +using ConstPointer = typename AllocatorTraits::const_pointer; +template +using SizeType = typename AllocatorTraits::size_type; +template +using DifferenceType = typename AllocatorTraits::difference_type; +template +using Reference = ValueType&; +template +using ConstReference = const ValueType&; +template +using Iterator = Pointer; +template +using ConstIterator = ConstPointer; +template +using ReverseIterator = typename std::reverse_iterator>; +template +using ConstReverseIterator = typename std::reverse_iterator>; +template +using MoveIterator = typename std::move_iterator>; + +template +using IsAtLeastForwardIterator = std::is_convertible< + typename std::iterator_traits::iterator_category, + std::forward_iterator_tag>; + +template +using IsMemcpyOk = + absl::conjunction>>, + absl::is_trivially_copy_constructible>, + absl::is_trivially_copy_assignable>, + absl::is_trivially_destructible>>; + +template +struct TypeIdentity { + using type = T; +}; + +// Used for function arguments in template functions to prevent ADL by forcing +// callers to explicitly specify the template parameter. +template +using NoTypeDeduction = typename TypeIdentity::type; + +template >::value> +struct DestroyAdapter; + +template +struct DestroyAdapter { + static void DestroyElements(A& allocator, Pointer destroy_first, + SizeType destroy_size) { + for (SizeType i = destroy_size; i != 0;) { + --i; + AllocatorTraits::destroy(allocator, destroy_first + i); + } + } +}; + +template +struct DestroyAdapter { + static void DestroyElements(A& allocator, Pointer destroy_first, + SizeType destroy_size) { + static_cast(allocator); + static_cast(destroy_first); + static_cast(destroy_size); + } +}; + +template +struct Allocation { + Pointer data; + SizeType capacity; +}; + +template ) > ABSL_INTERNAL_DEFAULT_NEW_ALIGNMENT)> +struct MallocAdapter { + static Allocation Allocate(A& allocator, SizeType requested_capacity) { + return {AllocatorTraits::allocate(allocator, requested_capacity), + requested_capacity}; + } + + static void Deallocate(A& allocator, Pointer pointer, + SizeType capacity) { + AllocatorTraits::deallocate(allocator, pointer, capacity); + } +}; + +template +void ConstructElements(NoTypeDeduction& allocator, + Pointer construct_first, ValueAdapter& values, + SizeType construct_size) { + for (SizeType i = 0; i < construct_size; ++i) { + ABSL_INTERNAL_TRY { values.ConstructNext(allocator, construct_first + i); } + ABSL_INTERNAL_CATCH_ANY { + DestroyAdapter::DestroyElements(allocator, construct_first, i); + ABSL_INTERNAL_RETHROW; + } + } +} + +template +void AssignElements(Pointer assign_first, ValueAdapter& values, + SizeType assign_size) { + for (SizeType i = 0; i < assign_size; ++i) { + values.AssignNext(assign_first + i); + } +} + +template +struct StorageView { + Pointer data; + SizeType size; + SizeType capacity; +}; + +template +class IteratorValueAdapter { + public: + explicit IteratorValueAdapter(const Iterator& it) : it_(it) {} + + void ConstructNext(A& allocator, Pointer construct_at) { + AllocatorTraits::construct(allocator, construct_at, *it_); + ++it_; + } + + void AssignNext(Pointer assign_at) { + *assign_at = *it_; + ++it_; + } + + private: + Iterator it_; +}; + +template +class CopyValueAdapter { + public: + explicit CopyValueAdapter(ConstPointer p) : ptr_(p) {} + + void ConstructNext(A& allocator, Pointer construct_at) { + AllocatorTraits::construct(allocator, construct_at, *ptr_); + } + + void AssignNext(Pointer assign_at) { *assign_at = *ptr_; } + + private: + ConstPointer ptr_; +}; + +template +class DefaultValueAdapter { + public: + explicit DefaultValueAdapter() {} + + void ConstructNext(A& allocator, Pointer construct_at) { + AllocatorTraits::construct(allocator, construct_at); + } + + void AssignNext(Pointer assign_at) { *assign_at = ValueType(); } +}; + +template +class AllocationTransaction { + public: + explicit AllocationTransaction(A& allocator) + : allocator_data_(allocator, nullptr), capacity_(0) {} + + ~AllocationTransaction() { + if (DidAllocate()) { + MallocAdapter::Deallocate(GetAllocator(), GetData(), GetCapacity()); + } + } + + AllocationTransaction(const AllocationTransaction&) = delete; + void operator=(const AllocationTransaction&) = delete; + + A& GetAllocator() { return allocator_data_.template get<0>(); } + Pointer& GetData() { return allocator_data_.template get<1>(); } + SizeType& GetCapacity() { return capacity_; } + + bool DidAllocate() { return GetData() != nullptr; } + + Pointer Allocate(SizeType requested_capacity) { + Allocation result = + MallocAdapter::Allocate(GetAllocator(), requested_capacity); + GetData() = result.data; + GetCapacity() = result.capacity; + return result.data; + } + + ABSL_MUST_USE_RESULT Allocation Release() && { + Allocation result = {GetData(), GetCapacity()}; + Reset(); + return result; + } + + private: + void Reset() { + GetData() = nullptr; + GetCapacity() = 0; + } + + container_internal::CompressedTuple> allocator_data_; + SizeType capacity_; +}; + +template +class ConstructionTransaction { + public: + explicit ConstructionTransaction(A& allocator) + : allocator_data_(allocator, nullptr), size_(0) {} + + ~ConstructionTransaction() { + if (DidConstruct()) { + DestroyAdapter::DestroyElements(GetAllocator(), GetData(), GetSize()); + } + } + + ConstructionTransaction(const ConstructionTransaction&) = delete; + void operator=(const ConstructionTransaction&) = delete; + + A& GetAllocator() { return allocator_data_.template get<0>(); } + Pointer& GetData() { return allocator_data_.template get<1>(); } + SizeType& GetSize() { return size_; } + + bool DidConstruct() { return GetData() != nullptr; } + template + void Construct(Pointer data, ValueAdapter& values, SizeType size) { + ConstructElements(GetAllocator(), data, values, size); + GetData() = data; + GetSize() = size; + } + void Commit() && { + GetData() = nullptr; + GetSize() = 0; + } + + private: + container_internal::CompressedTuple> allocator_data_; + SizeType size_; +}; + +template +class Storage { + public: + static SizeType NextCapacity(SizeType current_capacity) { + return current_capacity * 2; + } + + static SizeType ComputeCapacity(SizeType current_capacity, + SizeType requested_capacity) { + return (std::max)(NextCapacity(current_capacity), requested_capacity); + } + + // --------------------------------------------------------------------------- + // Storage Constructors and Destructor + // --------------------------------------------------------------------------- + + Storage() : metadata_(A(), /* size and is_allocated */ 0u) {} + + explicit Storage(const A& allocator) + : metadata_(allocator, /* size and is_allocated */ 0u) {} + + ~Storage() { + if (GetSizeAndIsAllocated() == 0) { + // Empty and not allocated; nothing to do. + } else if (IsMemcpyOk::value) { + // No destructors need to be run; just deallocate if necessary. + DeallocateIfAllocated(); + } else { + DestroyContents(); + } + } + + // --------------------------------------------------------------------------- + // Storage Member Accessors + // --------------------------------------------------------------------------- + + SizeType& GetSizeAndIsAllocated() { return metadata_.template get<1>(); } + + const SizeType& GetSizeAndIsAllocated() const { + return metadata_.template get<1>(); + } + + SizeType GetSize() const { return GetSizeAndIsAllocated() >> 1; } + + bool GetIsAllocated() const { return GetSizeAndIsAllocated() & 1; } + + Pointer GetAllocatedData() { return data_.allocated.allocated_data; } + + ConstPointer GetAllocatedData() const { + return data_.allocated.allocated_data; + } + + Pointer GetInlinedData() { + return reinterpret_cast>( + std::addressof(data_.inlined.inlined_data[0])); + } + + ConstPointer GetInlinedData() const { + return reinterpret_cast>( + std::addressof(data_.inlined.inlined_data[0])); + } + + SizeType GetAllocatedCapacity() const { + return data_.allocated.allocated_capacity; + } + + SizeType GetInlinedCapacity() const { return static_cast>(N); } + + StorageView MakeStorageView() { + return GetIsAllocated() ? StorageView{GetAllocatedData(), GetSize(), + GetAllocatedCapacity()} + : StorageView{GetInlinedData(), GetSize(), + GetInlinedCapacity()}; + } + + A& GetAllocator() { return metadata_.template get<0>(); } + + const A& GetAllocator() const { return metadata_.template get<0>(); } + + // --------------------------------------------------------------------------- + // Storage Member Mutators + // --------------------------------------------------------------------------- + + ABSL_ATTRIBUTE_NOINLINE void InitFrom(const Storage& other); + + template + void Initialize(ValueAdapter values, SizeType new_size); + + template + void Assign(ValueAdapter values, SizeType new_size); + + template + void Resize(ValueAdapter values, SizeType new_size); + + template + Iterator Insert(ConstIterator pos, ValueAdapter values, + SizeType insert_count); + + template + Reference EmplaceBack(Args&&... args); + + Iterator Erase(ConstIterator from, ConstIterator to); + + void Reserve(SizeType requested_capacity); + + void ShrinkToFit(); + + void Swap(Storage* other_storage_ptr); + + void SetIsAllocated() { + GetSizeAndIsAllocated() |= static_cast>(1); + } + + void UnsetIsAllocated() { + GetSizeAndIsAllocated() &= ((std::numeric_limits>::max)() - 1); + } + + void SetSize(SizeType size) { + GetSizeAndIsAllocated() = + (size << 1) | static_cast>(GetIsAllocated()); + } + + void SetAllocatedSize(SizeType size) { + GetSizeAndIsAllocated() = (size << 1) | static_cast>(1); + } + + void SetInlinedSize(SizeType size) { + GetSizeAndIsAllocated() = size << static_cast>(1); + } + + void AddSize(SizeType count) { + GetSizeAndIsAllocated() += count << static_cast>(1); + } + + void SubtractSize(SizeType count) { + ABSL_HARDENING_ASSERT(count <= GetSize()); + + GetSizeAndIsAllocated() -= count << static_cast>(1); + } + + void SetAllocation(Allocation allocation) { + data_.allocated.allocated_data = allocation.data; + data_.allocated.allocated_capacity = allocation.capacity; + } + + void MemcpyFrom(const Storage& other_storage) { + ABSL_HARDENING_ASSERT(IsMemcpyOk::value || + other_storage.GetIsAllocated()); + + GetSizeAndIsAllocated() = other_storage.GetSizeAndIsAllocated(); + data_ = other_storage.data_; + } + + void DeallocateIfAllocated() { + if (GetIsAllocated()) { + MallocAdapter::Deallocate(GetAllocator(), GetAllocatedData(), + GetAllocatedCapacity()); + } + } + + private: + ABSL_ATTRIBUTE_NOINLINE void DestroyContents(); + + using Metadata = container_internal::CompressedTuple>; + + struct Allocated { + Pointer allocated_data; + SizeType allocated_capacity; + }; + + struct Inlined { + alignas(ValueType) char inlined_data[sizeof(ValueType[N])]; + }; + + union Data { + Allocated allocated; + Inlined inlined; + }; + + template + ABSL_ATTRIBUTE_NOINLINE Reference EmplaceBackSlow(Args&&... args); + + Metadata metadata_; + Data data_; +}; + +template +void Storage::DestroyContents() { + Pointer data = GetIsAllocated() ? GetAllocatedData() : GetInlinedData(); + DestroyAdapter::DestroyElements(GetAllocator(), data, GetSize()); + DeallocateIfAllocated(); +} + +template +void Storage::InitFrom(const Storage& other) { + const SizeType n = other.GetSize(); + ABSL_HARDENING_ASSERT(n > 0); // Empty sources handled handled in caller. + ConstPointer src; + Pointer dst; + if (!other.GetIsAllocated()) { + dst = GetInlinedData(); + src = other.GetInlinedData(); + } else { + // Because this is only called from the `InlinedVector` constructors, it's + // safe to take on the allocation with size `0`. If `ConstructElements(...)` + // throws, deallocation will be automatically handled by `~Storage()`. + SizeType requested_capacity = ComputeCapacity(GetInlinedCapacity(), n); + Allocation allocation = + MallocAdapter::Allocate(GetAllocator(), requested_capacity); + SetAllocation(allocation); + dst = allocation.data; + src = other.GetAllocatedData(); + } + if (IsMemcpyOk::value) { + std::memcpy(reinterpret_cast(dst), + reinterpret_cast(src), n * sizeof(ValueType)); + } else { + auto values = IteratorValueAdapter>(src); + ConstructElements(GetAllocator(), dst, values, n); + } + GetSizeAndIsAllocated() = other.GetSizeAndIsAllocated(); +} + +template +template +auto Storage::Initialize(ValueAdapter values, SizeType new_size) + -> void { + // Only callable from constructors! + ABSL_HARDENING_ASSERT(!GetIsAllocated()); + ABSL_HARDENING_ASSERT(GetSize() == 0); + + Pointer construct_data; + if (new_size > GetInlinedCapacity()) { + // Because this is only called from the `InlinedVector` constructors, it's + // safe to take on the allocation with size `0`. If `ConstructElements(...)` + // throws, deallocation will be automatically handled by `~Storage()`. + SizeType requested_capacity = + ComputeCapacity(GetInlinedCapacity(), new_size); + Allocation allocation = + MallocAdapter::Allocate(GetAllocator(), requested_capacity); + construct_data = allocation.data; + SetAllocation(allocation); + SetIsAllocated(); + } else { + construct_data = GetInlinedData(); + } + + ConstructElements(GetAllocator(), construct_data, values, new_size); + + // Since the initial size was guaranteed to be `0` and the allocated bit is + // already correct for either case, *adding* `new_size` gives us the correct + // result faster than setting it directly. + AddSize(new_size); +} + +template +template +auto Storage::Assign(ValueAdapter values, SizeType new_size) + -> void { + StorageView storage_view = MakeStorageView(); + + AllocationTransaction allocation_tx(GetAllocator()); + + absl::Span> assign_loop; + absl::Span> construct_loop; + absl::Span> destroy_loop; + + if (new_size > storage_view.capacity) { + SizeType requested_capacity = + ComputeCapacity(storage_view.capacity, new_size); + construct_loop = {allocation_tx.Allocate(requested_capacity), new_size}; + destroy_loop = {storage_view.data, storage_view.size}; + } else if (new_size > storage_view.size) { + assign_loop = {storage_view.data, storage_view.size}; + construct_loop = {storage_view.data + storage_view.size, + new_size - storage_view.size}; + } else { + assign_loop = {storage_view.data, new_size}; + destroy_loop = {storage_view.data + new_size, storage_view.size - new_size}; + } + + AssignElements(assign_loop.data(), values, assign_loop.size()); + + ConstructElements(GetAllocator(), construct_loop.data(), values, + construct_loop.size()); + + DestroyAdapter::DestroyElements(GetAllocator(), destroy_loop.data(), + destroy_loop.size()); + + if (allocation_tx.DidAllocate()) { + DeallocateIfAllocated(); + SetAllocation(std::move(allocation_tx).Release()); + SetIsAllocated(); + } + + SetSize(new_size); +} + +template +template +auto Storage::Resize(ValueAdapter values, SizeType new_size) + -> void { + StorageView storage_view = MakeStorageView(); + Pointer const base = storage_view.data; + const SizeType size = storage_view.size; + A& alloc = GetAllocator(); + if (new_size <= size) { + // Destroy extra old elements. + DestroyAdapter::DestroyElements(alloc, base + new_size, size - new_size); + } else if (new_size <= storage_view.capacity) { + // Construct new elements in place. + ConstructElements(alloc, base + size, values, new_size - size); + } else { + // Steps: + // a. Allocate new backing store. + // b. Construct new elements in new backing store. + // c. Move existing elements from old backing store to new backing store. + // d. Destroy all elements in old backing store. + // Use transactional wrappers for the first two steps so we can roll + // back if necessary due to exceptions. + AllocationTransaction allocation_tx(alloc); + SizeType requested_capacity = + ComputeCapacity(storage_view.capacity, new_size); + Pointer new_data = allocation_tx.Allocate(requested_capacity); + + ConstructionTransaction construction_tx(alloc); + construction_tx.Construct(new_data + size, values, new_size - size); + + IteratorValueAdapter> move_values( + (MoveIterator(base))); + ConstructElements(alloc, new_data, move_values, size); + + DestroyAdapter::DestroyElements(alloc, base, size); + std::move(construction_tx).Commit(); + DeallocateIfAllocated(); + SetAllocation(std::move(allocation_tx).Release()); + SetIsAllocated(); + } + SetSize(new_size); +} + +template +template +auto Storage::Insert(ConstIterator pos, ValueAdapter values, + SizeType insert_count) -> Iterator { + StorageView storage_view = MakeStorageView(); + + SizeType insert_index = + std::distance(ConstIterator(storage_view.data), pos); + SizeType insert_end_index = insert_index + insert_count; + SizeType new_size = storage_view.size + insert_count; + + if (new_size > storage_view.capacity) { + AllocationTransaction allocation_tx(GetAllocator()); + ConstructionTransaction construction_tx(GetAllocator()); + ConstructionTransaction move_construction_tx(GetAllocator()); + + IteratorValueAdapter> move_values( + MoveIterator(storage_view.data)); + + SizeType requested_capacity = + ComputeCapacity(storage_view.capacity, new_size); + Pointer new_data = allocation_tx.Allocate(requested_capacity); + + construction_tx.Construct(new_data + insert_index, values, insert_count); + + move_construction_tx.Construct(new_data, move_values, insert_index); + + ConstructElements(GetAllocator(), new_data + insert_end_index, + move_values, storage_view.size - insert_index); + + DestroyAdapter::DestroyElements(GetAllocator(), storage_view.data, + storage_view.size); + + std::move(construction_tx).Commit(); + std::move(move_construction_tx).Commit(); + DeallocateIfAllocated(); + SetAllocation(std::move(allocation_tx).Release()); + + SetAllocatedSize(new_size); + return Iterator(new_data + insert_index); + } else { + SizeType move_construction_destination_index = + (std::max)(insert_end_index, storage_view.size); + + ConstructionTransaction move_construction_tx(GetAllocator()); + + IteratorValueAdapter> move_construction_values( + MoveIterator(storage_view.data + + (move_construction_destination_index - insert_count))); + absl::Span> move_construction = { + storage_view.data + move_construction_destination_index, + new_size - move_construction_destination_index}; + + Pointer move_assignment_values = storage_view.data + insert_index; + absl::Span> move_assignment = { + storage_view.data + insert_end_index, + move_construction_destination_index - insert_end_index}; + + absl::Span> insert_assignment = {move_assignment_values, + move_construction.size()}; + + absl::Span> insert_construction = { + insert_assignment.data() + insert_assignment.size(), + insert_count - insert_assignment.size()}; + + move_construction_tx.Construct(move_construction.data(), + move_construction_values, + move_construction.size()); + + for (Pointer + destination = move_assignment.data() + move_assignment.size(), + last_destination = move_assignment.data(), + source = move_assignment_values + move_assignment.size(); + ;) { + --destination; + --source; + if (destination < last_destination) break; + *destination = std::move(*source); + } + + AssignElements(insert_assignment.data(), values, + insert_assignment.size()); + + ConstructElements(GetAllocator(), insert_construction.data(), values, + insert_construction.size()); + + std::move(move_construction_tx).Commit(); + + AddSize(insert_count); + return Iterator(storage_view.data + insert_index); + } +} + +template +template +auto Storage::EmplaceBack(Args&&... args) -> Reference { + StorageView storage_view = MakeStorageView(); + const SizeType n = storage_view.size; + if (ABSL_PREDICT_TRUE(n != storage_view.capacity)) { + // Fast path; new element fits. + Pointer last_ptr = storage_view.data + n; + AllocatorTraits::construct(GetAllocator(), last_ptr, + std::forward(args)...); + AddSize(1); + return *last_ptr; + } + // TODO(b/173712035): Annotate with musttail attribute to prevent regression. + return EmplaceBackSlow(std::forward(args)...); +} + +template +template +auto Storage::EmplaceBackSlow(Args&&... args) -> Reference { + StorageView storage_view = MakeStorageView(); + AllocationTransaction allocation_tx(GetAllocator()); + IteratorValueAdapter> move_values( + MoveIterator(storage_view.data)); + SizeType requested_capacity = NextCapacity(storage_view.capacity); + Pointer construct_data = allocation_tx.Allocate(requested_capacity); + Pointer last_ptr = construct_data + storage_view.size; + + // Construct new element. + AllocatorTraits::construct(GetAllocator(), last_ptr, + std::forward(args)...); + // Move elements from old backing store to new backing store. + ABSL_INTERNAL_TRY { + ConstructElements(GetAllocator(), allocation_tx.GetData(), move_values, + storage_view.size); + } + ABSL_INTERNAL_CATCH_ANY { + AllocatorTraits::destroy(GetAllocator(), last_ptr); + ABSL_INTERNAL_RETHROW; + } + // Destroy elements in old backing store. + DestroyAdapter::DestroyElements(GetAllocator(), storage_view.data, + storage_view.size); + + DeallocateIfAllocated(); + SetAllocation(std::move(allocation_tx).Release()); + SetIsAllocated(); + AddSize(1); + return *last_ptr; +} + +template +auto Storage::Erase(ConstIterator from, ConstIterator to) + -> Iterator { + StorageView storage_view = MakeStorageView(); + + SizeType erase_size = std::distance(from, to); + SizeType erase_index = + std::distance(ConstIterator(storage_view.data), from); + SizeType erase_end_index = erase_index + erase_size; + + IteratorValueAdapter> move_values( + MoveIterator(storage_view.data + erase_end_index)); + + AssignElements(storage_view.data + erase_index, move_values, + storage_view.size - erase_end_index); + + DestroyAdapter::DestroyElements( + GetAllocator(), storage_view.data + (storage_view.size - erase_size), + erase_size); + + SubtractSize(erase_size); + return Iterator(storage_view.data + erase_index); +} + +template +auto Storage::Reserve(SizeType requested_capacity) -> void { + StorageView storage_view = MakeStorageView(); + + if (ABSL_PREDICT_FALSE(requested_capacity <= storage_view.capacity)) return; + + AllocationTransaction allocation_tx(GetAllocator()); + + IteratorValueAdapter> move_values( + MoveIterator(storage_view.data)); + + SizeType new_requested_capacity = + ComputeCapacity(storage_view.capacity, requested_capacity); + Pointer new_data = allocation_tx.Allocate(new_requested_capacity); + + ConstructElements(GetAllocator(), new_data, move_values, + storage_view.size); + + DestroyAdapter::DestroyElements(GetAllocator(), storage_view.data, + storage_view.size); + + DeallocateIfAllocated(); + SetAllocation(std::move(allocation_tx).Release()); + SetIsAllocated(); +} + +template +auto Storage::ShrinkToFit() -> void { + // May only be called on allocated instances! + ABSL_HARDENING_ASSERT(GetIsAllocated()); + + StorageView storage_view{GetAllocatedData(), GetSize(), + GetAllocatedCapacity()}; + + if (ABSL_PREDICT_FALSE(storage_view.size == storage_view.capacity)) return; + + AllocationTransaction allocation_tx(GetAllocator()); + + IteratorValueAdapter> move_values( + MoveIterator(storage_view.data)); + + Pointer construct_data; + if (storage_view.size > GetInlinedCapacity()) { + SizeType requested_capacity = storage_view.size; + construct_data = allocation_tx.Allocate(requested_capacity); + if (allocation_tx.GetCapacity() >= storage_view.capacity) { + // Already using the smallest available heap allocation. + return; + } + } else { + construct_data = GetInlinedData(); + } + + ABSL_INTERNAL_TRY { + ConstructElements(GetAllocator(), construct_data, move_values, + storage_view.size); + } + ABSL_INTERNAL_CATCH_ANY { + SetAllocation({storage_view.data, storage_view.capacity}); + ABSL_INTERNAL_RETHROW; + } + + DestroyAdapter::DestroyElements(GetAllocator(), storage_view.data, + storage_view.size); + + MallocAdapter::Deallocate(GetAllocator(), storage_view.data, + storage_view.capacity); + + if (allocation_tx.DidAllocate()) { + SetAllocation(std::move(allocation_tx).Release()); + } else { + UnsetIsAllocated(); + } +} + +template +auto Storage::Swap(Storage* other_storage_ptr) -> void { + using std::swap; + ABSL_HARDENING_ASSERT(this != other_storage_ptr); + + if (GetIsAllocated() && other_storage_ptr->GetIsAllocated()) { + swap(data_.allocated, other_storage_ptr->data_.allocated); + } else if (!GetIsAllocated() && !other_storage_ptr->GetIsAllocated()) { + Storage* small_ptr = this; + Storage* large_ptr = other_storage_ptr; + if (small_ptr->GetSize() > large_ptr->GetSize()) swap(small_ptr, large_ptr); + + for (SizeType i = 0; i < small_ptr->GetSize(); ++i) { + swap(small_ptr->GetInlinedData()[i], large_ptr->GetInlinedData()[i]); + } + + IteratorValueAdapter> move_values( + MoveIterator(large_ptr->GetInlinedData() + small_ptr->GetSize())); + + ConstructElements(large_ptr->GetAllocator(), + small_ptr->GetInlinedData() + small_ptr->GetSize(), + move_values, + large_ptr->GetSize() - small_ptr->GetSize()); + + DestroyAdapter::DestroyElements( + large_ptr->GetAllocator(), + large_ptr->GetInlinedData() + small_ptr->GetSize(), + large_ptr->GetSize() - small_ptr->GetSize()); + } else { + Storage* allocated_ptr = this; + Storage* inlined_ptr = other_storage_ptr; + if (!allocated_ptr->GetIsAllocated()) swap(allocated_ptr, inlined_ptr); + + StorageView allocated_storage_view{ + allocated_ptr->GetAllocatedData(), allocated_ptr->GetSize(), + allocated_ptr->GetAllocatedCapacity()}; + + IteratorValueAdapter> move_values( + MoveIterator(inlined_ptr->GetInlinedData())); + + ABSL_INTERNAL_TRY { + ConstructElements(inlined_ptr->GetAllocator(), + allocated_ptr->GetInlinedData(), move_values, + inlined_ptr->GetSize()); + } + ABSL_INTERNAL_CATCH_ANY { + allocated_ptr->SetAllocation(Allocation{ + allocated_storage_view.data, allocated_storage_view.capacity}); + ABSL_INTERNAL_RETHROW; + } + + DestroyAdapter::DestroyElements(inlined_ptr->GetAllocator(), + inlined_ptr->GetInlinedData(), + inlined_ptr->GetSize()); + + inlined_ptr->SetAllocation(Allocation{allocated_storage_view.data, + allocated_storage_view.capacity}); + } + + swap(GetSizeAndIsAllocated(), other_storage_ptr->GetSizeAndIsAllocated()); + swap(GetAllocator(), other_storage_ptr->GetAllocator()); +} + +// End ignore "array-bounds" +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic pop +#endif + +} // namespace inlined_vector_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_INLINED_VECTOR_INTERNAL_H_ diff --git a/src/absl/container/internal/layout.h b/src/absl/container/internal/layout.h new file mode 100644 index 0000000..a59a243 --- /dev/null +++ b/src/absl/container/internal/layout.h @@ -0,0 +1,743 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// MOTIVATION AND TUTORIAL +// +// If you want to put in a single heap allocation N doubles followed by M ints, +// it's easy if N and M are known at compile time. +// +// struct S { +// double a[N]; +// int b[M]; +// }; +// +// S* p = new S; +// +// But what if N and M are known only in run time? Class template Layout to the +// rescue! It's a portable generalization of the technique known as struct hack. +// +// // This object will tell us everything we need to know about the memory +// // layout of double[N] followed by int[M]. It's structurally identical to +// // size_t[2] that stores N and M. It's very cheap to create. +// const Layout layout(N, M); +// +// // Allocate enough memory for both arrays. `AllocSize()` tells us how much +// // memory is needed. We are free to use any allocation function we want as +// // long as it returns aligned memory. +// std::unique_ptr p(new unsigned char[layout.AllocSize()]); +// +// // Obtain the pointer to the array of doubles. +// // Equivalent to `reinterpret_cast(p.get())`. +// // +// // We could have written layout.Pointer<0>(p) instead. If all the types are +// // unique you can use either form, but if some types are repeated you must +// // use the index form. +// double* a = layout.Pointer(p.get()); +// +// // Obtain the pointer to the array of ints. +// // Equivalent to `reinterpret_cast(p.get() + N * 8)`. +// int* b = layout.Pointer(p); +// +// If we are unable to specify sizes of all fields, we can pass as many sizes as +// we can to `Partial()`. In return, it'll allow us to access the fields whose +// locations and sizes can be computed from the provided information. +// `Partial()` comes in handy when the array sizes are embedded into the +// allocation. +// +// // size_t[1] containing N, size_t[1] containing M, double[N], int[M]. +// using L = Layout; +// +// unsigned char* Allocate(size_t n, size_t m) { +// const L layout(1, 1, n, m); +// unsigned char* p = new unsigned char[layout.AllocSize()]; +// *layout.Pointer<0>(p) = n; +// *layout.Pointer<1>(p) = m; +// return p; +// } +// +// void Use(unsigned char* p) { +// // First, extract N and M. +// // Specify that the first array has only one element. Using `prefix` we +// // can access the first two arrays but not more. +// constexpr auto prefix = L::Partial(1); +// size_t n = *prefix.Pointer<0>(p); +// size_t m = *prefix.Pointer<1>(p); +// +// // Now we can get pointers to the payload. +// const L layout(1, 1, n, m); +// double* a = layout.Pointer(p); +// int* b = layout.Pointer(p); +// } +// +// The layout we used above combines fixed-size with dynamically-sized fields. +// This is quite common. Layout is optimized for this use case and generates +// optimal code. All computations that can be performed at compile time are +// indeed performed at compile time. +// +// Efficiency tip: The order of fields matters. In `Layout` try to +// ensure that `alignof(T1) >= ... >= alignof(TN)`. This way you'll have no +// padding in between arrays. +// +// You can manually override the alignment of an array by wrapping the type in +// `Aligned`. `Layout<..., Aligned, ...>` has exactly the same API +// and behavior as `Layout<..., T, ...>` except that the first element of the +// array of `T` is aligned to `N` (the rest of the elements follow without +// padding). `N` cannot be less than `alignof(T)`. +// +// `AllocSize()` and `Pointer()` are the most basic methods for dealing with +// memory layouts. Check out the reference or code below to discover more. +// +// EXAMPLE +// +// // Immutable move-only string with sizeof equal to sizeof(void*). The +// // string size and the characters are kept in the same heap allocation. +// class CompactString { +// public: +// CompactString(const char* s = "") { +// const size_t size = strlen(s); +// // size_t[1] followed by char[size + 1]. +// const L layout(1, size + 1); +// p_.reset(new unsigned char[layout.AllocSize()]); +// // If running under ASAN, mark the padding bytes, if any, to catch +// // memory errors. +// layout.PoisonPadding(p_.get()); +// // Store the size in the allocation. +// *layout.Pointer(p_.get()) = size; +// // Store the characters in the allocation. +// memcpy(layout.Pointer(p_.get()), s, size + 1); +// } +// +// size_t size() const { +// // Equivalent to reinterpret_cast(*p). +// return *L::Partial().Pointer(p_.get()); +// } +// +// const char* c_str() const { +// // Equivalent to reinterpret_cast(p.get() + sizeof(size_t)). +// // The argument in Partial(1) specifies that we have size_t[1] in front +// // of the characters. +// return L::Partial(1).Pointer(p_.get()); +// } +// +// private: +// // Our heap allocation contains a size_t followed by an array of chars. +// using L = Layout; +// std::unique_ptr p_; +// }; +// +// int main() { +// CompactString s = "hello"; +// assert(s.size() == 5); +// assert(strcmp(s.c_str(), "hello") == 0); +// } +// +// DOCUMENTATION +// +// The interface exported by this file consists of: +// - class `Layout<>` and its public members. +// - The public members of class `internal_layout::LayoutImpl<>`. That class +// isn't intended to be used directly, and its name and template parameter +// list are internal implementation details, but the class itself provides +// most of the functionality in this file. See comments on its members for +// detailed documentation. +// +// `Layout::Partial(count1,..., countm)` (where `m` <= `n`) returns a +// `LayoutImpl<>` object. `Layout layout(count1,..., countn)` +// creates a `Layout` object, which exposes the same functionality by inheriting +// from `LayoutImpl<>`. + +#ifndef ABSL_CONTAINER_INTERNAL_LAYOUT_H_ +#define ABSL_CONTAINER_INTERNAL_LAYOUT_H_ + +#include +#include +#include + +#include +#include +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/meta/type_traits.h" +#include "absl/strings/str_cat.h" +#include "absl/types/span.h" +#include "absl/utility/utility.h" + +#ifdef ABSL_HAVE_ADDRESS_SANITIZER +#include +#endif + +#if defined(__GXX_RTTI) +#define ABSL_INTERNAL_HAS_CXA_DEMANGLE +#endif + +#ifdef ABSL_INTERNAL_HAS_CXA_DEMANGLE +#include +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// A type wrapper that instructs `Layout` to use the specific alignment for the +// array. `Layout<..., Aligned, ...>` has exactly the same API +// and behavior as `Layout<..., T, ...>` except that the first element of the +// array of `T` is aligned to `N` (the rest of the elements follow without +// padding). +// +// Requires: `N >= alignof(T)` and `N` is a power of 2. +template +struct Aligned; + +namespace internal_layout { + +template +struct NotAligned {}; + +template +struct NotAligned> { + static_assert(sizeof(T) == 0, "Aligned cannot be const-qualified"); +}; + +template +using IntToSize = size_t; + +template +using TypeToSize = size_t; + +template +struct Type : NotAligned { + using type = T; +}; + +template +struct Type> { + using type = T; +}; + +template +struct SizeOf : NotAligned, std::integral_constant {}; + +template +struct SizeOf> : std::integral_constant {}; + +// Note: workaround for https://gcc.gnu.org/PR88115 +template +struct AlignOf : NotAligned { + static constexpr size_t value = alignof(T); +}; + +template +struct AlignOf> { + static_assert(N % alignof(T) == 0, + "Custom alignment can't be lower than the type's alignment"); + static constexpr size_t value = N; +}; + +// Does `Ts...` contain `T`? +template +using Contains = absl::disjunction...>; + +template +using CopyConst = + typename std::conditional::value, const To, To>::type; + +// Note: We're not qualifying this with absl:: because it doesn't compile under +// MSVC. +template +using SliceType = Span; + +// This namespace contains no types. It prevents functions defined in it from +// being found by ADL. +namespace adl_barrier { + +template +constexpr size_t Find(Needle, Needle, Ts...) { + static_assert(!Contains(), "Duplicate element type"); + return 0; +} + +template +constexpr size_t Find(Needle, T, Ts...) { + return adl_barrier::Find(Needle(), Ts()...) + 1; +} + +constexpr bool IsPow2(size_t n) { return !(n & (n - 1)); } + +// Returns `q * m` for the smallest `q` such that `q * m >= n`. +// Requires: `m` is a power of two. It's enforced by IsLegalElementType below. +constexpr size_t Align(size_t n, size_t m) { return (n + m - 1) & ~(m - 1); } + +constexpr size_t Min(size_t a, size_t b) { return b < a ? b : a; } + +constexpr size_t Max(size_t a) { return a; } + +template +constexpr size_t Max(size_t a, size_t b, Ts... rest) { + return adl_barrier::Max(b < a ? a : b, rest...); +} + +template +std::string TypeName() { + std::string out; + int status = 0; + char* demangled = nullptr; +#ifdef ABSL_INTERNAL_HAS_CXA_DEMANGLE + demangled = abi::__cxa_demangle(typeid(T).name(), nullptr, nullptr, &status); +#endif + if (status == 0 && demangled != nullptr) { // Demangling succeeded. + absl::StrAppend(&out, "<", demangled, ">"); + free(demangled); + } else { +#if defined(__GXX_RTTI) || defined(_CPPRTTI) + absl::StrAppend(&out, "<", typeid(T).name(), ">"); +#endif + } + return out; +} + +} // namespace adl_barrier + +template +using EnableIf = typename std::enable_if::type; + +// Can `T` be a template argument of `Layout`? +template +using IsLegalElementType = std::integral_constant< + bool, !std::is_reference::value && !std::is_volatile::value && + !std::is_reference::type>::value && + !std::is_volatile::type>::value && + adl_barrier::IsPow2(AlignOf::value)>; + +template +class LayoutImpl; + +// Public base class of `Layout` and the result type of `Layout::Partial()`. +// +// `Elements...` contains all template arguments of `Layout` that created this +// instance. +// +// `SizeSeq...` is `[0, NumSizes)` where `NumSizes` is the number of arguments +// passed to `Layout::Partial()` or `Layout::Layout()`. +// +// `OffsetSeq...` is `[0, NumOffsets)` where `NumOffsets` is +// `Min(sizeof...(Elements), NumSizes + 1)` (the number of arrays for which we +// can compute offsets). +template +class LayoutImpl, absl::index_sequence, + absl::index_sequence> { + private: + static_assert(sizeof...(Elements) > 0, "At least one field is required"); + static_assert(absl::conjunction...>::value, + "Invalid element type (see IsLegalElementType)"); + + enum { + NumTypes = sizeof...(Elements), + NumSizes = sizeof...(SizeSeq), + NumOffsets = sizeof...(OffsetSeq), + }; + + // These are guaranteed by `Layout`. + static_assert(NumOffsets == adl_barrier::Min(NumTypes, NumSizes + 1), + "Internal error"); + static_assert(NumTypes > 0, "Internal error"); + + // Returns the index of `T` in `Elements...`. Results in a compilation error + // if `Elements...` doesn't contain exactly one instance of `T`. + template + static constexpr size_t ElementIndex() { + static_assert(Contains, Type::type>...>(), + "Type not found"); + return adl_barrier::Find(Type(), + Type::type>()...); + } + + template + using ElementAlignment = + AlignOf>::type>; + + public: + // Element types of all arrays packed in a tuple. + using ElementTypes = std::tuple::type...>; + + // Element type of the Nth array. + template + using ElementType = typename std::tuple_element::type; + + constexpr explicit LayoutImpl(IntToSize... sizes) + : size_{sizes...} {} + + // Alignment of the layout, equal to the strictest alignment of all elements. + // All pointers passed to the methods of layout must be aligned to this value. + static constexpr size_t Alignment() { + return adl_barrier::Max(AlignOf::value...); + } + + // Offset in bytes of the Nth array. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // assert(x.Offset<0>() == 0); // The ints starts from 0. + // assert(x.Offset<1>() == 16); // The doubles starts from 16. + // + // Requires: `N <= NumSizes && N < sizeof...(Ts)`. + template = 0> + constexpr size_t Offset() const { + return 0; + } + + template = 0> + constexpr size_t Offset() const { + static_assert(N < NumOffsets, "Index out of bounds"); + return adl_barrier::Align( + Offset() + SizeOf>::value * size_[N - 1], + ElementAlignment::value); + } + + // Offset in bytes of the array with the specified element type. There must + // be exactly one such array and its zero-based index must be at most + // `NumSizes`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // assert(x.Offset() == 0); // The ints starts from 0. + // assert(x.Offset() == 16); // The doubles starts from 16. + template + constexpr size_t Offset() const { + return Offset()>(); + } + + // Offsets in bytes of all arrays for which the offsets are known. + constexpr std::array Offsets() const { + return {{Offset()...}}; + } + + // The number of elements in the Nth array. This is the Nth argument of + // `Layout::Partial()` or `Layout::Layout()` (zero-based). + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // assert(x.Size<0>() == 3); + // assert(x.Size<1>() == 4); + // + // Requires: `N < NumSizes`. + template + constexpr size_t Size() const { + static_assert(N < NumSizes, "Index out of bounds"); + return size_[N]; + } + + // The number of elements in the array with the specified element type. + // There must be exactly one such array and its zero-based index must be + // at most `NumSizes`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // assert(x.Size() == 3); + // assert(x.Size() == 4); + template + constexpr size_t Size() const { + return Size()>(); + } + + // The number of elements of all arrays for which they are known. + constexpr std::array Sizes() const { + return {{Size()...}}; + } + + // Pointer to the beginning of the Nth array. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // int* ints = x.Pointer<0>(p); + // double* doubles = x.Pointer<1>(p); + // + // Requires: `N <= NumSizes && N < sizeof...(Ts)`. + // Requires: `p` is aligned to `Alignment()`. + template + CopyConst>* Pointer(Char* p) const { + using C = typename std::remove_const::type; + static_assert( + std::is_same() || std::is_same() || + std::is_same(), + "The argument must be a pointer to [const] [signed|unsigned] char"); + constexpr size_t alignment = Alignment(); + (void)alignment; + assert(reinterpret_cast(p) % alignment == 0); + return reinterpret_cast>*>(p + Offset()); + } + + // Pointer to the beginning of the array with the specified element type. + // There must be exactly one such array and its zero-based index must be at + // most `NumSizes`. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // int* ints = x.Pointer(p); + // double* doubles = x.Pointer(p); + // + // Requires: `p` is aligned to `Alignment()`. + template + CopyConst* Pointer(Char* p) const { + return Pointer()>(p); + } + + // Pointers to all arrays for which pointers are known. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // + // int* ints; + // double* doubles; + // std::tie(ints, doubles) = x.Pointers(p); + // + // Requires: `p` is aligned to `Alignment()`. + // + // Note: We're not using ElementType alias here because it does not compile + // under MSVC. + template + std::tuple::type>*...> + Pointers(Char* p) const { + return std::tuple>*...>( + Pointer(p)...); + } + + // The Nth array. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // Span ints = x.Slice<0>(p); + // Span doubles = x.Slice<1>(p); + // + // Requires: `N < NumSizes`. + // Requires: `p` is aligned to `Alignment()`. + template + SliceType>> Slice(Char* p) const { + return SliceType>>(Pointer(p), Size()); + } + + // The array with the specified element type. There must be exactly one + // such array and its zero-based index must be less than `NumSizes`. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // Span ints = x.Slice(p); + // Span doubles = x.Slice(p); + // + // Requires: `p` is aligned to `Alignment()`. + template + SliceType> Slice(Char* p) const { + return Slice()>(p); + } + + // All arrays with known sizes. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; + // + // Span ints; + // Span doubles; + // std::tie(ints, doubles) = x.Slices(p); + // + // Requires: `p` is aligned to `Alignment()`. + // + // Note: We're not using ElementType alias here because it does not compile + // under MSVC. + template + std::tuple::type>>...> + Slices(Char* p) const { + // Workaround for https://gcc.gnu.org/bugzilla/show_bug.cgi?id=63875 (fixed + // in 6.1). + (void)p; + return std::tuple>>...>( + Slice(p)...); + } + + // The size of the allocation that fits all arrays. + // + // // int[3], 4 bytes of padding, double[4]. + // Layout x(3, 4); + // unsigned char* p = new unsigned char[x.AllocSize()]; // 48 bytes + // + // Requires: `NumSizes == sizeof...(Ts)`. + constexpr size_t AllocSize() const { + static_assert(NumTypes == NumSizes, "You must specify sizes of all fields"); + return Offset() + + SizeOf>::value * size_[NumTypes - 1]; + } + + // If built with --config=asan, poisons padding bytes (if any) in the + // allocation. The pointer must point to a memory block at least + // `AllocSize()` bytes in length. + // + // `Char` must be `[const] [signed|unsigned] char`. + // + // Requires: `p` is aligned to `Alignment()`. + template = 0> + void PoisonPadding(const Char* p) const { + Pointer<0>(p); // verify the requirements on `Char` and `p` + } + + template = 0> + void PoisonPadding(const Char* p) const { + static_assert(N < NumOffsets, "Index out of bounds"); + (void)p; +#ifdef ABSL_HAVE_ADDRESS_SANITIZER + PoisonPadding(p); + // The `if` is an optimization. It doesn't affect the observable behaviour. + if (ElementAlignment::value % ElementAlignment::value) { + size_t start = + Offset() + SizeOf>::value * size_[N - 1]; + ASAN_POISON_MEMORY_REGION(p + start, Offset() - start); + } +#endif + } + + // Human-readable description of the memory layout. Useful for debugging. + // Slow. + // + // // char[5], 3 bytes of padding, int[3], 4 bytes of padding, followed + // // by an unknown number of doubles. + // auto x = Layout::Partial(5, 3); + // assert(x.DebugString() == + // "@0(1)[5]; @8(4)[3]; @24(8)"); + // + // Each field is in the following format: @offset(sizeof)[size] ( + // may be missing depending on the target platform). For example, + // @8(4)[3] means that at offset 8 we have an array of ints, where each + // int is 4 bytes, and we have 3 of those ints. The size of the last field may + // be missing (as in the example above). Only fields with known offsets are + // described. Type names may differ across platforms: one compiler might + // produce "unsigned*" where another produces "unsigned int *". + std::string DebugString() const { + const auto offsets = Offsets(); + const size_t sizes[] = {SizeOf>::value...}; + const std::string types[] = { + adl_barrier::TypeName>()...}; + std::string res = absl::StrCat("@0", types[0], "(", sizes[0], ")"); + for (size_t i = 0; i != NumOffsets - 1; ++i) { + absl::StrAppend(&res, "[", size_[i], "]; @", offsets[i + 1], types[i + 1], + "(", sizes[i + 1], ")"); + } + // NumSizes is a constant that may be zero. Some compilers cannot see that + // inside the if statement "size_[NumSizes - 1]" must be valid. + int last = static_cast(NumSizes) - 1; + if (NumTypes == NumSizes && last >= 0) { + absl::StrAppend(&res, "[", size_[last], "]"); + } + return res; + } + + private: + // Arguments of `Layout::Partial()` or `Layout::Layout()`. + size_t size_[NumSizes > 0 ? NumSizes : 1]; +}; + +template +using LayoutType = LayoutImpl< + std::tuple, absl::make_index_sequence, + absl::make_index_sequence>; + +} // namespace internal_layout + +// Descriptor of arrays of various types and sizes laid out in memory one after +// another. See the top of the file for documentation. +// +// Check out the public API of internal_layout::LayoutImpl above. The type is +// internal to the library but its methods are public, and they are inherited +// by `Layout`. +template +class Layout : public internal_layout::LayoutType { + public: + static_assert(sizeof...(Ts) > 0, "At least one field is required"); + static_assert( + absl::conjunction...>::value, + "Invalid element type (see IsLegalElementType)"); + + // The result type of `Partial()` with `NumSizes` arguments. + template + using PartialType = internal_layout::LayoutType; + + // `Layout` knows the element types of the arrays we want to lay out in + // memory but not the number of elements in each array. + // `Partial(size1, ..., sizeN)` allows us to specify the latter. The + // resulting immutable object can be used to obtain pointers to the + // individual arrays. + // + // It's allowed to pass fewer array sizes than the number of arrays. E.g., + // if all you need is to the offset of the second array, you only need to + // pass one argument -- the number of elements in the first array. + // + // // int[3] followed by 4 bytes of padding and an unknown number of + // // doubles. + // auto x = Layout::Partial(3); + // // doubles start at byte 16. + // assert(x.Offset<1>() == 16); + // + // If you know the number of elements in all arrays, you can still call + // `Partial()` but it's more convenient to use the constructor of `Layout`. + // + // Layout x(3, 5); + // + // Note: The sizes of the arrays must be specified in number of elements, + // not in bytes. + // + // Requires: `sizeof...(Sizes) <= sizeof...(Ts)`. + // Requires: all arguments are convertible to `size_t`. + template + static constexpr PartialType Partial(Sizes&&... sizes) { + static_assert(sizeof...(Sizes) <= sizeof...(Ts), ""); + return PartialType(absl::forward(sizes)...); + } + + // Creates a layout with the sizes of all arrays specified. If you know + // only the sizes of the first N arrays (where N can be zero), you can use + // `Partial()` defined above. The constructor is essentially equivalent to + // calling `Partial()` and passing in all array sizes; the constructor is + // provided as a convenient abbreviation. + // + // Note: The sizes of the arrays must be specified in number of elements, + // not in bytes. + constexpr explicit Layout(internal_layout::TypeToSize... sizes) + : internal_layout::LayoutType(sizes...) {} +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_LAYOUT_H_ diff --git a/src/absl/container/internal/node_slot_policy.h b/src/absl/container/internal/node_slot_policy.h new file mode 100644 index 0000000..baba574 --- /dev/null +++ b/src/absl/container/internal/node_slot_policy.h @@ -0,0 +1,92 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Adapts a policy for nodes. +// +// The node policy should model: +// +// struct Policy { +// // Returns a new node allocated and constructed using the allocator, using +// // the specified arguments. +// template +// value_type* new_element(Alloc* alloc, Args&&... args) const; +// +// // Destroys and deallocates node using the allocator. +// template +// void delete_element(Alloc* alloc, value_type* node) const; +// }; +// +// It may also optionally define `value()` and `apply()`. For documentation on +// these, see hash_policy_traits.h. + +#ifndef ABSL_CONTAINER_INTERNAL_NODE_SLOT_POLICY_H_ +#define ABSL_CONTAINER_INTERNAL_NODE_SLOT_POLICY_H_ + +#include +#include +#include +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +struct node_slot_policy { + static_assert(std::is_lvalue_reference::value, ""); + + using slot_type = typename std::remove_cv< + typename std::remove_reference::type>::type*; + + template + static void construct(Alloc* alloc, slot_type* slot, Args&&... args) { + *slot = Policy::new_element(alloc, std::forward(args)...); + } + + template + static void destroy(Alloc* alloc, slot_type* slot) { + Policy::delete_element(alloc, *slot); + } + + template + static void transfer(Alloc*, slot_type* new_slot, slot_type* old_slot) { + *new_slot = *old_slot; + } + + static size_t space_used(const slot_type* slot) { + if (slot == nullptr) return Policy::element_space_used(nullptr); + return Policy::element_space_used(*slot); + } + + static Reference element(slot_type* slot) { return **slot; } + + template + static auto value(T* elem) -> decltype(P::value(elem)) { + return P::value(elem); + } + + template + static auto apply(Ts&&... ts) -> decltype(P::apply(std::forward(ts)...)) { + return P::apply(std::forward(ts)...); + } +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_NODE_SLOT_POLICY_H_ diff --git a/src/absl/container/internal/raw_hash_map.h b/src/absl/container/internal/raw_hash_map.h new file mode 100644 index 0000000..c7df2ef --- /dev/null +++ b/src/absl/container/internal/raw_hash_map.h @@ -0,0 +1,198 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_RAW_HASH_MAP_H_ +#define ABSL_CONTAINER_INTERNAL_RAW_HASH_MAP_H_ + +#include +#include +#include + +#include "absl/base/internal/throw_delegate.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/raw_hash_set.h" // IWYU pragma: export + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +class raw_hash_map : public raw_hash_set { + // P is Policy. It's passed as a template argument to support maps that have + // incomplete types as values, as in unordered_map. + // MappedReference<> may be a non-reference type. + template + using MappedReference = decltype(P::value( + std::addressof(std::declval()))); + + // MappedConstReference<> may be a non-reference type. + template + using MappedConstReference = decltype(P::value( + std::addressof(std::declval()))); + + using KeyArgImpl = + KeyArg::value && IsTransparent::value>; + + public: + using key_type = typename Policy::key_type; + using mapped_type = typename Policy::mapped_type; + template + using key_arg = typename KeyArgImpl::template type; + + static_assert(!std::is_reference::value, ""); + + // TODO(b/187807849): Evaluate whether to support reference mapped_type and + // remove this assertion if/when it is supported. + static_assert(!std::is_reference::value, ""); + + using iterator = typename raw_hash_map::raw_hash_set::iterator; + using const_iterator = typename raw_hash_map::raw_hash_set::const_iterator; + + raw_hash_map() {} + using raw_hash_map::raw_hash_set::raw_hash_set; + + // The last two template parameters ensure that both arguments are rvalues + // (lvalue arguments are handled by the overloads below). This is necessary + // for supporting bitfield arguments. + // + // union { int n : 1; }; + // flat_hash_map m; + // m.insert_or_assign(n, n); + template + std::pair insert_or_assign(key_arg&& k, V&& v) { + return insert_or_assign_impl(std::forward(k), std::forward(v)); + } + + template + std::pair insert_or_assign(key_arg&& k, const V& v) { + return insert_or_assign_impl(std::forward(k), v); + } + + template + std::pair insert_or_assign(const key_arg& k, V&& v) { + return insert_or_assign_impl(k, std::forward(v)); + } + + template + std::pair insert_or_assign(const key_arg& k, const V& v) { + return insert_or_assign_impl(k, v); + } + + template + iterator insert_or_assign(const_iterator, key_arg&& k, V&& v) { + return insert_or_assign(std::forward(k), std::forward(v)).first; + } + + template + iterator insert_or_assign(const_iterator, key_arg&& k, const V& v) { + return insert_or_assign(std::forward(k), v).first; + } + + template + iterator insert_or_assign(const_iterator, const key_arg& k, V&& v) { + return insert_or_assign(k, std::forward(v)).first; + } + + template + iterator insert_or_assign(const_iterator, const key_arg& k, const V& v) { + return insert_or_assign(k, v).first; + } + + // All `try_emplace()` overloads make the same guarantees regarding rvalue + // arguments as `std::unordered_map::try_emplace()`, namely that these + // functions will not move from rvalue arguments if insertions do not happen. + template ::value, int>::type = 0, + K* = nullptr> + std::pair try_emplace(key_arg&& k, Args&&... args) { + return try_emplace_impl(std::forward(k), std::forward(args)...); + } + + template ::value, int>::type = 0> + std::pair try_emplace(const key_arg& k, Args&&... args) { + return try_emplace_impl(k, std::forward(args)...); + } + + template + iterator try_emplace(const_iterator, key_arg&& k, Args&&... args) { + return try_emplace(std::forward(k), std::forward(args)...).first; + } + + template + iterator try_emplace(const_iterator, const key_arg& k, Args&&... args) { + return try_emplace(k, std::forward(args)...).first; + } + + template + MappedReference

at(const key_arg& key) { + auto it = this->find(key); + if (it == this->end()) { + base_internal::ThrowStdOutOfRange( + "absl::container_internal::raw_hash_map<>::at"); + } + return Policy::value(&*it); + } + + template + MappedConstReference

at(const key_arg& key) const { + auto it = this->find(key); + if (it == this->end()) { + base_internal::ThrowStdOutOfRange( + "absl::container_internal::raw_hash_map<>::at"); + } + return Policy::value(&*it); + } + + template + MappedReference

operator[](key_arg&& key) { + return Policy::value(&*try_emplace(std::forward(key)).first); + } + + template + MappedReference

operator[](const key_arg& key) { + return Policy::value(&*try_emplace(key).first); + } + + private: + template + std::pair insert_or_assign_impl(K&& k, V&& v) { + auto res = this->find_or_prepare_insert(k); + if (res.second) + this->emplace_at(res.first, std::forward(k), std::forward(v)); + else + Policy::value(&*this->iterator_at(res.first)) = std::forward(v); + return {this->iterator_at(res.first), res.second}; + } + + template + std::pair try_emplace_impl(K&& k, Args&&... args) { + auto res = this->find_or_prepare_insert(k); + if (res.second) + this->emplace_at(res.first, std::piecewise_construct, + std::forward_as_tuple(std::forward(k)), + std::forward_as_tuple(std::forward(args)...)); + return {this->iterator_at(res.first), res.second}; + } +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_RAW_HASH_MAP_H_ diff --git a/src/absl/container/internal/raw_hash_set.cc b/src/absl/container/internal/raw_hash_set.cc new file mode 100644 index 0000000..c63a2e0 --- /dev/null +++ b/src/absl/container/internal/raw_hash_set.cc @@ -0,0 +1,71 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/container/internal/raw_hash_set.h" + +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// A single block of empty control bytes for tables without any slots allocated. +// This enables removing a branch in the hot path of find(). +alignas(16) ABSL_CONST_INIT ABSL_DLL const ctrl_t kEmptyGroup[16] = { + ctrl_t::kSentinel, ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, + ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, + ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, + ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty, ctrl_t::kEmpty}; + +#ifdef ABSL_INTERNAL_NEED_REDUNDANT_CONSTEXPR_DECL +constexpr size_t Group::kWidth; +#endif + +// Returns "random" seed. +inline size_t RandomSeed() { +#ifdef ABSL_HAVE_THREAD_LOCAL + static thread_local size_t counter = 0; + size_t value = ++counter; +#else // ABSL_HAVE_THREAD_LOCAL + static std::atomic counter(0); + size_t value = counter.fetch_add(1, std::memory_order_relaxed); +#endif // ABSL_HAVE_THREAD_LOCAL + return value ^ static_cast(reinterpret_cast(&counter)); +} + +bool ShouldInsertBackwards(size_t hash, const ctrl_t* ctrl) { + // To avoid problems with weak hashes and single bit tests, we use % 13. + // TODO(kfm,sbenza): revisit after we do unconditional mixing + return (H1(hash, ctrl) ^ RandomSeed()) % 13 > 6; +} + +void ConvertDeletedToEmptyAndFullToDeleted(ctrl_t* ctrl, size_t capacity) { + assert(ctrl[capacity] == ctrl_t::kSentinel); + assert(IsValidCapacity(capacity)); + for (ctrl_t* pos = ctrl; pos < ctrl + capacity; pos += Group::kWidth) { + Group{pos}.ConvertSpecialToEmptyAndFullToDeleted(pos); + } + // Copy the cloned ctrl bytes. + std::memcpy(ctrl + capacity + 1, ctrl, NumClonedBytes()); + ctrl[capacity] = ctrl_t::kSentinel; +} +// Extern template instantiotion for inline function. +template FindInfo find_first_non_full(const ctrl_t*, size_t, size_t); + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/container/internal/raw_hash_set.h b/src/absl/container/internal/raw_hash_set.h new file mode 100644 index 0000000..e875e24 --- /dev/null +++ b/src/absl/container/internal/raw_hash_set.h @@ -0,0 +1,2369 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// An open-addressing +// hashtable with quadratic probing. +// +// This is a low level hashtable on top of which different interfaces can be +// implemented, like flat_hash_set, node_hash_set, string_hash_set, etc. +// +// The table interface is similar to that of std::unordered_set. Notable +// differences are that most member functions support heterogeneous keys when +// BOTH the hash and eq functions are marked as transparent. They do so by +// providing a typedef called `is_transparent`. +// +// When heterogeneous lookup is enabled, functions that take key_type act as if +// they have an overload set like: +// +// iterator find(const key_type& key); +// template +// iterator find(const K& key); +// +// size_type erase(const key_type& key); +// template +// size_type erase(const K& key); +// +// std::pair equal_range(const key_type& key); +// template +// std::pair equal_range(const K& key); +// +// When heterogeneous lookup is disabled, only the explicit `key_type` overloads +// exist. +// +// find() also supports passing the hash explicitly: +// +// iterator find(const key_type& key, size_t hash); +// template +// iterator find(const U& key, size_t hash); +// +// In addition the pointer to element and iterator stability guarantees are +// weaker: all iterators and pointers are invalidated after a new element is +// inserted. +// +// IMPLEMENTATION DETAILS +// +// # Table Layout +// +// A raw_hash_set's backing array consists of control bytes followed by slots +// that may or may not contain objects. +// +// The layout of the backing array, for `capacity` slots, is thus, as a +// pseudo-struct: +// +// struct BackingArray { +// // Control bytes for the "real" slots. +// ctrl_t ctrl[capacity]; +// // Always `ctrl_t::kSentinel`. This is used by iterators to find when to +// // stop and serves no other purpose. +// ctrl_t sentinel; +// // A copy of the first `kWidth - 1` elements of `ctrl`. This is used so +// // that if a probe sequence picks a value near the end of `ctrl`, +// // `Group` will have valid control bytes to look at. +// ctrl_t clones[kWidth - 1]; +// // The actual slot data. +// slot_type slots[capacity]; +// }; +// +// The length of this array is computed by `AllocSize()` below. +// +// Control bytes (`ctrl_t`) are bytes (collected into groups of a +// platform-specific size) that define the state of the corresponding slot in +// the slot array. Group manipulation is tightly optimized to be as efficient +// as possible: SSE and friends on x86, clever bit operations on other arches. +// +// Group 1 Group 2 Group 3 +// +---------------+---------------+---------------+ +// | | | | | | | | | | | | | | | | | | | | | | | | | +// +---------------+---------------+---------------+ +// +// Each control byte is either a special value for empty slots, deleted slots +// (sometimes called *tombstones*), and a special end-of-table marker used by +// iterators, or, if occupied, seven bits (H2) from the hash of the value in the +// corresponding slot. +// +// Storing control bytes in a separate array also has beneficial cache effects, +// since more logical slots will fit into a cache line. +// +// # Hashing +// +// We compute two separate hashes, `H1` and `H2`, from the hash of an object. +// `H1(hash(x))` is an index into `slots`, and essentially the starting point +// for the probe sequence. `H2(hash(x))` is a 7-bit value used to filter out +// objects that cannot possibly be the one we are looking for. +// +// # Table operations. +// +// The key operations are `insert`, `find`, and `erase`. +// +// Since `insert` and `erase` are implemented in terms of `find`, we describe +// `find` first. To `find` a value `x`, we compute `hash(x)`. From +// `H1(hash(x))` and the capacity, we construct a `probe_seq` that visits every +// group of slots in some interesting order. +// +// We now walk through these indices. At each index, we select the entire group +// starting with that index and extract potential candidates: occupied slots +// with a control byte equal to `H2(hash(x))`. If we find an empty slot in the +// group, we stop and return an error. Each candidate slot `y` is compared with +// `x`; if `x == y`, we are done and return `&y`; otherwise we contine to the +// next probe index. Tombstones effectively behave like full slots that never +// match the value we're looking for. +// +// The `H2` bits ensure when we compare a slot to an object with `==`, we are +// likely to have actually found the object. That is, the chance is low that +// `==` is called and returns `false`. Thus, when we search for an object, we +// are unlikely to call `==` many times. This likelyhood can be analyzed as +// follows (assuming that H2 is a random enough hash function). +// +// Let's assume that there are `k` "wrong" objects that must be examined in a +// probe sequence. For example, when doing a `find` on an object that is in the +// table, `k` is the number of objects between the start of the probe sequence +// and the final found object (not including the final found object). The +// expected number of objects with an H2 match is then `k/128`. Measurements +// and analysis indicate that even at high load factors, `k` is less than 32, +// meaning that the number of "false positive" comparisons we must perform is +// less than 1/8 per `find`. + +// `insert` is implemented in terms of `unchecked_insert`, which inserts a +// value presumed to not be in the table (violating this requirement will cause +// the table to behave erratically). Given `x` and its hash `hash(x)`, to insert +// it, we construct a `probe_seq` once again, and use it to find the first +// group with an unoccupied (empty *or* deleted) slot. We place `x` into the +// first such slot in the group and mark it as full with `x`'s H2. +// +// To `insert`, we compose `unchecked_insert` with `find`. We compute `h(x)` and +// perform a `find` to see if it's already present; if it is, we're done. If +// it's not, we may decide the table is getting overcrowded (i.e. the load +// factor is greater than 7/8 for big tables; `is_small()` tables use a max load +// factor of 1); in this case, we allocate a bigger array, `unchecked_insert` +// each element of the table into the new array (we know that no insertion here +// will insert an already-present value), and discard the old backing array. At +// this point, we may `unchecked_insert` the value `x`. +// +// Below, `unchecked_insert` is partly implemented by `prepare_insert`, which +// presents a viable, initialized slot pointee to the caller. +// +// `erase` is implemented in terms of `erase_at`, which takes an index to a +// slot. Given an offset, we simply create a tombstone and destroy its contents. +// If we can prove that the slot would not appear in a probe sequence, we can +// make the slot as empty, instead. We can prove this by observing that if a +// group has any empty slots, it has never been full (assuming we never create +// an empty slot in a group with no empties, which this heuristic guarantees we +// never do) and find would stop at this group anyways (since it does not probe +// beyond groups with empties). +// +// `erase` is `erase_at` composed with `find`: if we +// have a value `x`, we can perform a `find`, and then `erase_at` the resulting +// slot. +// +// To iterate, we simply traverse the array, skipping empty and deleted slots +// and stopping when we hit a `kSentinel`. + +#ifndef ABSL_CONTAINER_INTERNAL_RAW_HASH_SET_H_ +#define ABSL_CONTAINER_INTERNAL_RAW_HASH_SET_H_ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/base/internal/endian.h" +#include "absl/base/internal/prefetch.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" +#include "absl/container/internal/common.h" +#include "absl/container/internal/compressed_tuple.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/hash_policy_traits.h" +#include "absl/container/internal/hashtable_debug_hooks.h" +#include "absl/container/internal/hashtablez_sampler.h" +#include "absl/memory/memory.h" +#include "absl/meta/type_traits.h" +#include "absl/numeric/bits.h" +#include "absl/utility/utility.h" + +#ifdef ABSL_INTERNAL_HAVE_SSE2 +#include +#endif + +#ifdef ABSL_INTERNAL_HAVE_SSSE3 +#include +#endif + +#ifdef _MSC_VER +#include +#endif + +#ifdef ABSL_INTERNAL_HAVE_ARM_NEON +#include +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +template +void SwapAlloc(AllocType& lhs, AllocType& rhs, + std::true_type /* propagate_on_container_swap */) { + using std::swap; + swap(lhs, rhs); +} +template +void SwapAlloc(AllocType& /*lhs*/, AllocType& /*rhs*/, + std::false_type /* propagate_on_container_swap */) {} + +// The state for a probe sequence. +// +// Currently, the sequence is a triangular progression of the form +// +// p(i) := Width * (i^2 + i)/2 + hash (mod mask + 1) +// +// The use of `Width` ensures that each probe step does not overlap groups; +// the sequence effectively outputs the addresses of *groups* (although not +// necessarily aligned to any boundary). The `Group` machinery allows us +// to check an entire group with minimal branching. +// +// Wrapping around at `mask + 1` is important, but not for the obvious reason. +// As described above, the first few entries of the control byte array +// are mirrored at the end of the array, which `Group` will find and use +// for selecting candidates. However, when those candidates' slots are +// actually inspected, there are no corresponding slots for the cloned bytes, +// so we need to make sure we've treated those offsets as "wrapping around". +// +// It turns out that this probe sequence visits every group exactly once if the +// number of groups is a power of two, since (i^2+i)/2 is a bijection in +// Z/(2^m). See https://en.wikipedia.org/wiki/Quadratic_probing +template +class probe_seq { + public: + // Creates a new probe sequence using `hash` as the initial value of the + // sequence and `mask` (usually the capacity of the table) as the mask to + // apply to each value in the progression. + probe_seq(size_t hash, size_t mask) { + assert(((mask + 1) & mask) == 0 && "not a mask"); + mask_ = mask; + offset_ = hash & mask_; + } + + // The offset within the table, i.e., the value `p(i)` above. + size_t offset() const { return offset_; } + size_t offset(size_t i) const { return (offset_ + i) & mask_; } + + void next() { + index_ += Width; + offset_ += index_; + offset_ &= mask_; + } + // 0-based probe index, a multiple of `Width`. + size_t index() const { return index_; } + + private: + size_t mask_; + size_t offset_; + size_t index_ = 0; +}; + +template +struct RequireUsableKey { + template + std::pair< + decltype(std::declval()(std::declval())), + decltype(std::declval()(std::declval(), + std::declval()))>* + operator()(const PassedKey&, const Args&...) const; +}; + +template +struct IsDecomposable : std::false_type {}; + +template +struct IsDecomposable< + absl::void_t(), + std::declval()...))>, + Policy, Hash, Eq, Ts...> : std::true_type {}; + +// TODO(alkis): Switch to std::is_nothrow_swappable when gcc/clang supports it. +template +constexpr bool IsNoThrowSwappable(std::true_type = {} /* is_swappable */) { + using std::swap; + return noexcept(swap(std::declval(), std::declval())); +} +template +constexpr bool IsNoThrowSwappable(std::false_type /* is_swappable */) { + return false; +} + +template +uint32_t TrailingZeros(T x) { + ABSL_ASSUME(x != 0); + return static_cast(countr_zero(x)); +} + +// An abstract bitmask, such as that emitted by a SIMD instruction. +// +// Specifically, this type implements a simple bitset whose representation is +// controlled by `SignificantBits` and `Shift`. `SignificantBits` is the number +// of abstract bits in the bitset, while `Shift` is the log-base-two of the +// width of an abstract bit in the representation. +// This mask provides operations for any number of real bits set in an abstract +// bit. To add iteration on top of that, implementation must guarantee no more +// than one real bit is set in an abstract bit. +template +class NonIterableBitMask { + public: + explicit NonIterableBitMask(T mask) : mask_(mask) {} + + explicit operator bool() const { return this->mask_ != 0; } + + // Returns the index of the lowest *abstract* bit set in `self`. + uint32_t LowestBitSet() const { + return container_internal::TrailingZeros(mask_) >> Shift; + } + + // Returns the index of the highest *abstract* bit set in `self`. + uint32_t HighestBitSet() const { + return static_cast((bit_width(mask_) - 1) >> Shift); + } + + // Return the number of trailing zero *abstract* bits. + uint32_t TrailingZeros() const { + return container_internal::TrailingZeros(mask_) >> Shift; + } + + // Return the number of leading zero *abstract* bits. + uint32_t LeadingZeros() const { + constexpr int total_significant_bits = SignificantBits << Shift; + constexpr int extra_bits = sizeof(T) * 8 - total_significant_bits; + return static_cast(countl_zero(mask_ << extra_bits)) >> Shift; + } + + T mask_; +}; + +// Mask that can be iterable +// +// For example, when `SignificantBits` is 16 and `Shift` is zero, this is just +// an ordinary 16-bit bitset occupying the low 16 bits of `mask`. When +// `SignificantBits` is 8 and `Shift` is 3, abstract bits are represented as +// the bytes `0x00` and `0x80`, and it occupies all 64 bits of the bitmask. +// +// For example: +// for (int i : BitMask(0b101)) -> yields 0, 2 +// for (int i : BitMask(0x0000000080800000)) -> yields 2, 3 +template +class BitMask : public NonIterableBitMask { + using Base = NonIterableBitMask; + static_assert(std::is_unsigned::value, ""); + static_assert(Shift == 0 || Shift == 3, ""); + + public: + explicit BitMask(T mask) : Base(mask) {} + // BitMask is an iterator over the indices of its abstract bits. + using value_type = int; + using iterator = BitMask; + using const_iterator = BitMask; + + BitMask& operator++() { + this->mask_ &= (this->mask_ - 1); + return *this; + } + + uint32_t operator*() const { return Base::LowestBitSet(); } + + BitMask begin() const { return *this; } + BitMask end() const { return BitMask(0); } + + private: + friend bool operator==(const BitMask& a, const BitMask& b) { + return a.mask_ == b.mask_; + } + friend bool operator!=(const BitMask& a, const BitMask& b) { + return a.mask_ != b.mask_; + } +}; + +using h2_t = uint8_t; + +// The values here are selected for maximum performance. See the static asserts +// below for details. + +// A `ctrl_t` is a single control byte, which can have one of four +// states: empty, deleted, full (which has an associated seven-bit h2_t value) +// and the sentinel. They have the following bit patterns: +// +// empty: 1 0 0 0 0 0 0 0 +// deleted: 1 1 1 1 1 1 1 0 +// full: 0 h h h h h h h // h represents the hash bits. +// sentinel: 1 1 1 1 1 1 1 1 +// +// These values are specifically tuned for SSE-flavored SIMD. +// The static_asserts below detail the source of these choices. +// +// We use an enum class so that when strict aliasing is enabled, the compiler +// knows ctrl_t doesn't alias other types. +enum class ctrl_t : int8_t { + kEmpty = -128, // 0b10000000 + kDeleted = -2, // 0b11111110 + kSentinel = -1, // 0b11111111 +}; +static_assert( + (static_cast(ctrl_t::kEmpty) & + static_cast(ctrl_t::kDeleted) & + static_cast(ctrl_t::kSentinel) & 0x80) != 0, + "Special markers need to have the MSB to make checking for them efficient"); +static_assert( + ctrl_t::kEmpty < ctrl_t::kSentinel && ctrl_t::kDeleted < ctrl_t::kSentinel, + "ctrl_t::kEmpty and ctrl_t::kDeleted must be smaller than " + "ctrl_t::kSentinel to make the SIMD test of IsEmptyOrDeleted() efficient"); +static_assert( + ctrl_t::kSentinel == static_cast(-1), + "ctrl_t::kSentinel must be -1 to elide loading it from memory into SIMD " + "registers (pcmpeqd xmm, xmm)"); +static_assert(ctrl_t::kEmpty == static_cast(-128), + "ctrl_t::kEmpty must be -128 to make the SIMD check for its " + "existence efficient (psignb xmm, xmm)"); +static_assert( + (~static_cast(ctrl_t::kEmpty) & + ~static_cast(ctrl_t::kDeleted) & + static_cast(ctrl_t::kSentinel) & 0x7F) != 0, + "ctrl_t::kEmpty and ctrl_t::kDeleted must share an unset bit that is not " + "shared by ctrl_t::kSentinel to make the scalar test for " + "MaskEmptyOrDeleted() efficient"); +static_assert(ctrl_t::kDeleted == static_cast(-2), + "ctrl_t::kDeleted must be -2 to make the implementation of " + "ConvertSpecialToEmptyAndFullToDeleted efficient"); + +ABSL_DLL extern const ctrl_t kEmptyGroup[16]; + +// Returns a pointer to a control byte group that can be used by empty tables. +inline ctrl_t* EmptyGroup() { + // Const must be cast away here; no uses of this function will actually write + // to it, because it is only used for empty tables. + return const_cast(kEmptyGroup); +} + +// Mixes a randomly generated per-process seed with `hash` and `ctrl` to +// randomize insertion order within groups. +bool ShouldInsertBackwards(size_t hash, const ctrl_t* ctrl); + +// Returns a per-table, hash salt, which changes on resize. This gets mixed into +// H1 to randomize iteration order per-table. +// +// The seed consists of the ctrl_ pointer, which adds enough entropy to ensure +// non-determinism of iteration order in most cases. +inline size_t PerTableSalt(const ctrl_t* ctrl) { + // The low bits of the pointer have little or no entropy because of + // alignment. We shift the pointer to try to use higher entropy bits. A + // good number seems to be 12 bits, because that aligns with page size. + return reinterpret_cast(ctrl) >> 12; +} +// Extracts the H1 portion of a hash: 57 bits mixed with a per-table salt. +inline size_t H1(size_t hash, const ctrl_t* ctrl) { + return (hash >> 7) ^ PerTableSalt(ctrl); +} + +// Extracts the H2 portion of a hash: the 7 bits not used for H1. +// +// These are used as an occupied control byte. +inline h2_t H2(size_t hash) { return hash & 0x7F; } + +// Helpers for checking the state of a control byte. +inline bool IsEmpty(ctrl_t c) { return c == ctrl_t::kEmpty; } +inline bool IsFull(ctrl_t c) { return c >= static_cast(0); } +inline bool IsDeleted(ctrl_t c) { return c == ctrl_t::kDeleted; } +inline bool IsEmptyOrDeleted(ctrl_t c) { return c < ctrl_t::kSentinel; } + +#ifdef ABSL_INTERNAL_HAVE_SSE2 +// Quick reference guide for intrinsics used below: +// +// * __m128i: An XMM (128-bit) word. +// +// * _mm_setzero_si128: Returns a zero vector. +// * _mm_set1_epi8: Returns a vector with the same i8 in each lane. +// +// * _mm_subs_epi8: Saturating-subtracts two i8 vectors. +// * _mm_and_si128: Ands two i128s together. +// * _mm_or_si128: Ors two i128s together. +// * _mm_andnot_si128: And-nots two i128s together. +// +// * _mm_cmpeq_epi8: Component-wise compares two i8 vectors for equality, +// filling each lane with 0x00 or 0xff. +// * _mm_cmpgt_epi8: Same as above, but using > rather than ==. +// +// * _mm_loadu_si128: Performs an unaligned load of an i128. +// * _mm_storeu_si128: Performs an unaligned store of an i128. +// +// * _mm_sign_epi8: Retains, negates, or zeroes each i8 lane of the first +// argument if the corresponding lane of the second +// argument is positive, negative, or zero, respectively. +// * _mm_movemask_epi8: Selects the sign bit out of each i8 lane and produces a +// bitmask consisting of those bits. +// * _mm_shuffle_epi8: Selects i8s from the first argument, using the low +// four bits of each i8 lane in the second argument as +// indices. + +// https://github.com/abseil/abseil-cpp/issues/209 +// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=87853 +// _mm_cmpgt_epi8 is broken under GCC with -funsigned-char +// Work around this by using the portable implementation of Group +// when using -funsigned-char under GCC. +inline __m128i _mm_cmpgt_epi8_fixed(__m128i a, __m128i b) { +// dd: in the latest gcc we get warnings about overflow here. +// the linked bug was fixed in gcc 9 and fixes were backported +// to gcc 7 and 8 patch releases. To fix the warning, I just +// added a check for the GCC version. +#if defined(__GNUC__) && (__GNUC__ < 9) && !defined(__clang__) + if (std::is_unsigned::value) { + const __m128i mask = _mm_set1_epi8(0x80); + const __m128i diff = _mm_subs_epi8(b, a); + return _mm_cmpeq_epi8(_mm_and_si128(diff, mask), mask); + } +#endif + return _mm_cmpgt_epi8(a, b); +} + +struct GroupSse2Impl { + static constexpr size_t kWidth = 16; // the number of slots per group + + explicit GroupSse2Impl(const ctrl_t* pos) { + ctrl = _mm_loadu_si128(reinterpret_cast(pos)); + } + + // Returns a bitmask representing the positions of slots that match hash. + BitMask Match(h2_t hash) const { + auto match = _mm_set1_epi8(hash); + return BitMask( + static_cast(_mm_movemask_epi8(_mm_cmpeq_epi8(match, ctrl)))); + } + + // Returns a bitmask representing the positions of empty slots. + NonIterableBitMask MaskEmpty() const { +#ifdef ABSL_INTERNAL_HAVE_SSSE3 + // This only works because ctrl_t::kEmpty is -128. + return NonIterableBitMask( + static_cast(_mm_movemask_epi8(_mm_sign_epi8(ctrl, ctrl)))); +#else + auto match = _mm_set1_epi8(static_cast(ctrl_t::kEmpty)); + return NonIterableBitMask( + static_cast(_mm_movemask_epi8(_mm_cmpeq_epi8(match, ctrl)))); +#endif + } + + // Returns a bitmask representing the positions of empty or deleted slots. + NonIterableBitMask MaskEmptyOrDeleted() const { + auto special = _mm_set1_epi8(static_cast(ctrl_t::kSentinel)); + return NonIterableBitMask(static_cast( + _mm_movemask_epi8(_mm_cmpgt_epi8_fixed(special, ctrl)))); + } + + // Returns the number of trailing empty or deleted elements in the group. + uint32_t CountLeadingEmptyOrDeleted() const { + auto special = _mm_set1_epi8(static_cast(ctrl_t::kSentinel)); + return TrailingZeros(static_cast( + _mm_movemask_epi8(_mm_cmpgt_epi8_fixed(special, ctrl)) + 1)); + } + + void ConvertSpecialToEmptyAndFullToDeleted(ctrl_t* dst) const { + auto msbs = _mm_set1_epi8(static_cast(-128)); + auto x126 = _mm_set1_epi8(126); +#ifdef ABSL_INTERNAL_HAVE_SSSE3 + auto res = _mm_or_si128(_mm_shuffle_epi8(x126, ctrl), msbs); +#else + auto zero = _mm_setzero_si128(); + auto special_mask = _mm_cmpgt_epi8_fixed(zero, ctrl); + auto res = _mm_or_si128(msbs, _mm_andnot_si128(special_mask, x126)); +#endif + _mm_storeu_si128(reinterpret_cast<__m128i*>(dst), res); + } + + __m128i ctrl; +}; +#endif // ABSL_INTERNAL_RAW_HASH_SET_HAVE_SSE2 + +#if defined(ABSL_INTERNAL_HAVE_ARM_NEON) && defined(ABSL_IS_LITTLE_ENDIAN) +struct GroupAArch64Impl { + static constexpr size_t kWidth = 8; + + explicit GroupAArch64Impl(const ctrl_t* pos) { + ctrl = vld1_u8(reinterpret_cast(pos)); + } + + BitMask Match(h2_t hash) const { + uint8x8_t dup = vdup_n_u8(hash); + auto mask = vceq_u8(ctrl, dup); + constexpr uint64_t msbs = 0x8080808080808080ULL; + return BitMask( + vget_lane_u64(vreinterpret_u64_u8(mask), 0) & msbs); + } + + NonIterableBitMask MaskEmpty() const { + uint64_t mask = + vget_lane_u64(vreinterpret_u64_u8( + vceq_s8(vdup_n_s8(static_cast(ctrl_t::kEmpty)), + vreinterpret_s8_u8(ctrl))), + 0); + return NonIterableBitMask(mask); + } + + NonIterableBitMask MaskEmptyOrDeleted() const { + uint64_t mask = + vget_lane_u64(vreinterpret_u64_u8(vcgt_s8( + vdup_n_s8(static_cast(ctrl_t::kSentinel)), + vreinterpret_s8_u8(ctrl))), + 0); + return NonIterableBitMask(mask); + } + + uint32_t CountLeadingEmptyOrDeleted() const { + uint64_t mask = vget_lane_u64(vreinterpret_u64_u8(ctrl), 0); + // ctrl | ~(ctrl >> 7) will have the lowest bit set to zero for kEmpty and + // kDeleted. We lower all other bits and count number of trailing zeros. + // Clang and GCC optimize countr_zero to rbit+clz without any check for 0, + // so we should be fine. + constexpr uint64_t bits = 0x0101010101010101ULL; + return countr_zero((mask | ~(mask >> 7)) & bits) >> 3; + } + + void ConvertSpecialToEmptyAndFullToDeleted(ctrl_t* dst) const { + uint64_t mask = vget_lane_u64(vreinterpret_u64_u8(ctrl), 0); + constexpr uint64_t msbs = 0x8080808080808080ULL; + constexpr uint64_t lsbs = 0x0101010101010101ULL; + auto x = mask & msbs; + auto res = (~x + (x >> 7)) & ~lsbs; + little_endian::Store64(dst, res); + } + + uint8x8_t ctrl; +}; +#endif // ABSL_INTERNAL_HAVE_ARM_NEON && ABSL_IS_LITTLE_ENDIAN + +struct GroupPortableImpl { + static constexpr size_t kWidth = 8; + + explicit GroupPortableImpl(const ctrl_t* pos) + : ctrl(little_endian::Load64(pos)) {} + + BitMask Match(h2_t hash) const { + // For the technique, see: + // http://graphics.stanford.edu/~seander/bithacks.html##ValueInWord + // (Determine if a word has a byte equal to n). + // + // Caveat: there are false positives but: + // - they only occur if there is a real match + // - they never occur on ctrl_t::kEmpty, ctrl_t::kDeleted, ctrl_t::kSentinel + // - they will be handled gracefully by subsequent checks in code + // + // Example: + // v = 0x1716151413121110 + // hash = 0x12 + // retval = (v - lsbs) & ~v & msbs = 0x0000000080800000 + constexpr uint64_t msbs = 0x8080808080808080ULL; + constexpr uint64_t lsbs = 0x0101010101010101ULL; + auto x = ctrl ^ (lsbs * hash); + return BitMask((x - lsbs) & ~x & msbs); + } + + NonIterableBitMask MaskEmpty() const { + constexpr uint64_t msbs = 0x8080808080808080ULL; + return NonIterableBitMask((ctrl & (~ctrl << 6)) & + msbs); + } + + NonIterableBitMask MaskEmptyOrDeleted() const { + constexpr uint64_t msbs = 0x8080808080808080ULL; + return NonIterableBitMask((ctrl & (~ctrl << 7)) & + msbs); + } + + uint32_t CountLeadingEmptyOrDeleted() const { + // ctrl | ~(ctrl >> 7) will have the lowest bit set to zero for kEmpty and + // kDeleted. We lower all other bits and count number of trailing zeros. + constexpr uint64_t bits = 0x0101010101010101ULL; + return countr_zero((ctrl | ~(ctrl >> 7)) & bits) >> 3; + } + + void ConvertSpecialToEmptyAndFullToDeleted(ctrl_t* dst) const { + constexpr uint64_t msbs = 0x8080808080808080ULL; + constexpr uint64_t lsbs = 0x0101010101010101ULL; + auto x = ctrl & msbs; + auto res = (~x + (x >> 7)) & ~lsbs; + little_endian::Store64(dst, res); + } + + uint64_t ctrl; +}; + +#ifdef ABSL_INTERNAL_HAVE_SSE2 +using Group = GroupSse2Impl; +#elif defined(ABSL_INTERNAL_HAVE_ARM_NEON) && defined(ABSL_IS_LITTLE_ENDIAN) +using Group = GroupAArch64Impl; +#else +using Group = GroupPortableImpl; +#endif + +// Returns he number of "cloned control bytes". +// +// This is the number of control bytes that are present both at the beginning +// of the control byte array and at the end, such that we can create a +// `Group::kWidth`-width probe window starting from any control byte. +constexpr size_t NumClonedBytes() { return Group::kWidth - 1; } + +template +class raw_hash_set; + +// Returns whether `n` is a valid capacity (i.e., number of slots). +// +// A valid capacity is a non-zero integer `2^m - 1`. +inline bool IsValidCapacity(size_t n) { return ((n + 1) & n) == 0 && n > 0; } + +// Applies the following mapping to every byte in the control array: +// * kDeleted -> kEmpty +// * kEmpty -> kEmpty +// * _ -> kDeleted +// PRECONDITION: +// IsValidCapacity(capacity) +// ctrl[capacity] == ctrl_t::kSentinel +// ctrl[i] != ctrl_t::kSentinel for all i < capacity +void ConvertDeletedToEmptyAndFullToDeleted(ctrl_t* ctrl, size_t capacity); + +// Converts `n` into the next valid capacity, per `IsValidCapacity`. +inline size_t NormalizeCapacity(size_t n) { + return n ? ~size_t{} >> countl_zero(n) : 1; +} + +// General notes on capacity/growth methods below: +// - We use 7/8th as maximum load factor. For 16-wide groups, that gives an +// average of two empty slots per group. +// - For (capacity+1) >= Group::kWidth, growth is 7/8*capacity. +// - For (capacity+1) < Group::kWidth, growth == capacity. In this case, we +// never need to probe (the whole table fits in one group) so we don't need a +// load factor less than 1. + +// Given `capacity`, applies the load factor; i.e., it returns the maximum +// number of values we should put into the table before a resizing rehash. +inline size_t CapacityToGrowth(size_t capacity) { + assert(IsValidCapacity(capacity)); + // `capacity*7/8` + if (Group::kWidth == 8 && capacity == 7) { + // x-x/8 does not work when x==7. + return 6; + } + return capacity - capacity / 8; +} + +// Given `growth`, "unapplies" the load factor to find how large the capacity +// should be to stay within the load factor. +// +// This might not be a valid capacity and `NormalizeCapacity()` should be +// called on this. +inline size_t GrowthToLowerboundCapacity(size_t growth) { + // `growth*8/7` + if (Group::kWidth == 8 && growth == 7) { + // x+(x-1)/7 does not work when x==7. + return 8; + } + return growth + static_cast((static_cast(growth) - 1) / 7); +} + +template +size_t SelectBucketCountForIterRange(InputIter first, InputIter last, + size_t bucket_count) { + if (bucket_count != 0) { + return bucket_count; + } + using InputIterCategory = + typename std::iterator_traits::iterator_category; + if (std::is_base_of::value) { + return GrowthToLowerboundCapacity( + static_cast(std::distance(first, last))); + } + return 0; +} + +#define ABSL_INTERNAL_ASSERT_IS_FULL(ctrl, msg) \ + ABSL_HARDENING_ASSERT((ctrl != nullptr && IsFull(*ctrl)) && msg) + +inline void AssertIsValid(ctrl_t* ctrl) { + ABSL_HARDENING_ASSERT( + (ctrl == nullptr || IsFull(*ctrl)) && + "Invalid operation on iterator. The element might have " + "been erased, the table might have rehashed, or this may " + "be an end() iterator."); +} + +struct FindInfo { + size_t offset; + size_t probe_length; +}; + +// Whether a table is "small". A small table fits entirely into a probing +// group, i.e., has a capacity < `Group::kWidth`. +// +// In small mode we are able to use the whole capacity. The extra control +// bytes give us at least one "empty" control byte to stop the iteration. +// This is important to make 1 a valid capacity. +// +// In small mode only the first `capacity` control bytes after the sentinel +// are valid. The rest contain dummy ctrl_t::kEmpty values that do not +// represent a real slot. This is important to take into account on +// `find_first_non_full()`, where we never try +// `ShouldInsertBackwards()` for small tables. +inline bool is_small(size_t capacity) { return capacity < Group::kWidth - 1; } + +// Begins a probing operation on `ctrl`, using `hash`. +inline probe_seq probe(const ctrl_t* ctrl, size_t hash, + size_t capacity) { + return probe_seq(H1(hash, ctrl), capacity); +} + +// Probes an array of control bits using a probe sequence derived from `hash`, +// and returns the offset corresponding to the first deleted or empty slot. +// +// Behavior when the entire table is full is undefined. +// +// NOTE: this function must work with tables having both empty and deleted +// slots in the same group. Such tables appear during `erase()`. +template +inline FindInfo find_first_non_full(const ctrl_t* ctrl, size_t hash, + size_t capacity) { + auto seq = probe(ctrl, hash, capacity); + while (true) { + Group g{ctrl + seq.offset()}; + auto mask = g.MaskEmptyOrDeleted(); + if (mask) { +#if !defined(NDEBUG) + // We want to add entropy even when ASLR is not enabled. + // In debug build we will randomly insert in either the front or back of + // the group. + // TODO(kfm,sbenza): revisit after we do unconditional mixing + if (!is_small(capacity) && ShouldInsertBackwards(hash, ctrl)) { + return {seq.offset(mask.HighestBitSet()), seq.index()}; + } +#endif + return {seq.offset(mask.LowestBitSet()), seq.index()}; + } + seq.next(); + assert(seq.index() <= capacity && "full table!"); + } +} + +// Extern template for inline function keep possibility of inlining. +// When compiler decided to not inline, no symbols will be added to the +// corresponding translation unit. +extern template FindInfo find_first_non_full(const ctrl_t*, size_t, size_t); + +// Sets `ctrl` to `{kEmpty, kSentinel, ..., kEmpty}`, marking the entire +// array as marked as empty. +inline void ResetCtrl(size_t capacity, ctrl_t* ctrl, const void* slot, + size_t slot_size) { + std::memset(ctrl, static_cast(ctrl_t::kEmpty), + capacity + 1 + NumClonedBytes()); + ctrl[capacity] = ctrl_t::kSentinel; + SanitizerPoisonMemoryRegion(slot, slot_size * capacity); +} + +// Sets `ctrl[i]` to `h`. +// +// Unlike setting it directly, this function will perform bounds checks and +// mirror the value to the cloned tail if necessary. +inline void SetCtrl(size_t i, ctrl_t h, size_t capacity, ctrl_t* ctrl, + const void* slot, size_t slot_size) { + assert(i < capacity); + + auto* slot_i = static_cast(slot) + i * slot_size; + if (IsFull(h)) { + SanitizerUnpoisonMemoryRegion(slot_i, slot_size); + } else { + SanitizerPoisonMemoryRegion(slot_i, slot_size); + } + + ctrl[i] = h; + ctrl[((i - NumClonedBytes()) & capacity) + (NumClonedBytes() & capacity)] = h; +} + +// Overload for setting to an occupied `h2_t` rather than a special `ctrl_t`. +inline void SetCtrl(size_t i, h2_t h, size_t capacity, ctrl_t* ctrl, + const void* slot, size_t slot_size) { + SetCtrl(i, static_cast(h), capacity, ctrl, slot, slot_size); +} + +// Given the capacity of a table, computes the offset (from the start of the +// backing allocation) at which the slots begin. +inline size_t SlotOffset(size_t capacity, size_t slot_align) { + assert(IsValidCapacity(capacity)); + const size_t num_control_bytes = capacity + 1 + NumClonedBytes(); + return (num_control_bytes + slot_align - 1) & (~slot_align + 1); +} + +// Given the capacity of a table, computes the total size of the backing +// array. +inline size_t AllocSize(size_t capacity, size_t slot_size, size_t slot_align) { + return SlotOffset(capacity, slot_align) + capacity * slot_size; +} + +// A SwissTable. +// +// Policy: a policy defines how to perform different operations on +// the slots of the hashtable (see hash_policy_traits.h for the full interface +// of policy). +// +// Hash: a (possibly polymorphic) functor that hashes keys of the hashtable. The +// functor should accept a key and return size_t as hash. For best performance +// it is important that the hash function provides high entropy across all bits +// of the hash. +// +// Eq: a (possibly polymorphic) functor that compares two keys for equality. It +// should accept two (of possibly different type) keys and return a bool: true +// if they are equal, false if they are not. If two keys compare equal, then +// their hash values as defined by Hash MUST be equal. +// +// Allocator: an Allocator +// [https://en.cppreference.com/w/cpp/named_req/Allocator] with which +// the storage of the hashtable will be allocated and the elements will be +// constructed and destroyed. +template +class raw_hash_set { + using PolicyTraits = hash_policy_traits; + using KeyArgImpl = + KeyArg::value && IsTransparent::value>; + + public: + using init_type = typename PolicyTraits::init_type; + using key_type = typename PolicyTraits::key_type; + // TODO(sbenza): Hide slot_type as it is an implementation detail. Needs user + // code fixes! + using slot_type = typename PolicyTraits::slot_type; + using allocator_type = Alloc; + using size_type = size_t; + using difference_type = ptrdiff_t; + using hasher = Hash; + using key_equal = Eq; + using policy_type = Policy; + using value_type = typename PolicyTraits::value_type; + using reference = value_type&; + using const_reference = const value_type&; + using pointer = typename absl::allocator_traits< + allocator_type>::template rebind_traits::pointer; + using const_pointer = typename absl::allocator_traits< + allocator_type>::template rebind_traits::const_pointer; + + // Alias used for heterogeneous lookup functions. + // `key_arg` evaluates to `K` when the functors are transparent and to + // `key_type` otherwise. It permits template argument deduction on `K` for the + // transparent case. + template + using key_arg = typename KeyArgImpl::template type; + + private: + // Give an early error when key_type is not hashable/eq. + auto KeyTypeCanBeHashed(const Hash& h, const key_type& k) -> decltype(h(k)); + auto KeyTypeCanBeEq(const Eq& eq, const key_type& k) -> decltype(eq(k, k)); + + using AllocTraits = absl::allocator_traits; + using SlotAlloc = typename absl::allocator_traits< + allocator_type>::template rebind_alloc; + using SlotAllocTraits = typename absl::allocator_traits< + allocator_type>::template rebind_traits; + + static_assert(std::is_lvalue_reference::value, + "Policy::element() must return a reference"); + + template + struct SameAsElementReference + : std::is_same::type>::type, + typename std::remove_cv< + typename std::remove_reference::type>::type> {}; + + // An enabler for insert(T&&): T must be convertible to init_type or be the + // same as [cv] value_type [ref]. + // Note: we separate SameAsElementReference into its own type to avoid using + // reference unless we need to. MSVC doesn't seem to like it in some + // cases. + template + using RequiresInsertable = typename std::enable_if< + absl::disjunction, + SameAsElementReference>::value, + int>::type; + + // RequiresNotInit is a workaround for gcc prior to 7.1. + // See https://godbolt.org/g/Y4xsUh. + template + using RequiresNotInit = + typename std::enable_if::value, int>::type; + + template + using IsDecomposable = IsDecomposable; + + public: + static_assert(std::is_same::value, + "Allocators with custom pointer types are not supported"); + static_assert(std::is_same::value, + "Allocators with custom pointer types are not supported"); + + class iterator { + friend class raw_hash_set; + + public: + using iterator_category = std::forward_iterator_tag; + using value_type = typename raw_hash_set::value_type; + using reference = + absl::conditional_t; + using pointer = absl::remove_reference_t*; + using difference_type = typename raw_hash_set::difference_type; + + iterator() {} + + // PRECONDITION: not an end() iterator. + reference operator*() const { + ABSL_INTERNAL_ASSERT_IS_FULL(ctrl_, + "operator*() called on invalid iterator."); + return PolicyTraits::element(slot_); + } + + // PRECONDITION: not an end() iterator. + pointer operator->() const { + ABSL_INTERNAL_ASSERT_IS_FULL(ctrl_, + "operator-> called on invalid iterator."); + return &operator*(); + } + + // PRECONDITION: not an end() iterator. + iterator& operator++() { + ABSL_INTERNAL_ASSERT_IS_FULL(ctrl_, + "operator++ called on invalid iterator."); + ++ctrl_; + ++slot_; + skip_empty_or_deleted(); + return *this; + } + // PRECONDITION: not an end() iterator. + iterator operator++(int) { + auto tmp = *this; + ++*this; + return tmp; + } + + friend bool operator==(const iterator& a, const iterator& b) { + AssertIsValid(a.ctrl_); + AssertIsValid(b.ctrl_); + return a.ctrl_ == b.ctrl_; + } + friend bool operator!=(const iterator& a, const iterator& b) { + return !(a == b); + } + + private: + iterator(ctrl_t* ctrl, slot_type* slot) : ctrl_(ctrl), slot_(slot) { + // This assumption helps the compiler know that any non-end iterator is + // not equal to any end iterator. + ABSL_ASSUME(ctrl != nullptr); + } + + // Fixes up `ctrl_` to point to a full by advancing it and `slot_` until + // they reach one. + // + // If a sentinel is reached, we null both of them out instead. + void skip_empty_or_deleted() { + while (IsEmptyOrDeleted(*ctrl_)) { + uint32_t shift = Group{ctrl_}.CountLeadingEmptyOrDeleted(); + ctrl_ += shift; + slot_ += shift; + } + if (ABSL_PREDICT_FALSE(*ctrl_ == ctrl_t::kSentinel)) ctrl_ = nullptr; + } + + ctrl_t* ctrl_ = nullptr; + // To avoid uninitialized member warnings, put slot_ in an anonymous union. + // The member is not initialized on singleton and end iterators. + union { + slot_type* slot_; + }; + }; + + class const_iterator { + friend class raw_hash_set; + + public: + using iterator_category = typename iterator::iterator_category; + using value_type = typename raw_hash_set::value_type; + using reference = typename raw_hash_set::const_reference; + using pointer = typename raw_hash_set::const_pointer; + using difference_type = typename raw_hash_set::difference_type; + + const_iterator() {} + // Implicit construction from iterator. + const_iterator(iterator i) : inner_(std::move(i)) {} + + reference operator*() const { return *inner_; } + pointer operator->() const { return inner_.operator->(); } + + const_iterator& operator++() { + ++inner_; + return *this; + } + const_iterator operator++(int) { return inner_++; } + + friend bool operator==(const const_iterator& a, const const_iterator& b) { + return a.inner_ == b.inner_; + } + friend bool operator!=(const const_iterator& a, const const_iterator& b) { + return !(a == b); + } + + private: + const_iterator(const ctrl_t* ctrl, const slot_type* slot) + : inner_(const_cast(ctrl), const_cast(slot)) {} + + iterator inner_; + }; + + using node_type = node_handle, Alloc>; + using insert_return_type = InsertReturnType; + + raw_hash_set() noexcept( + std::is_nothrow_default_constructible::value&& + std::is_nothrow_default_constructible::value&& + std::is_nothrow_default_constructible::value) {} + + explicit raw_hash_set(size_t bucket_count, const hasher& hash = hasher(), + const key_equal& eq = key_equal(), + const allocator_type& alloc = allocator_type()) + : ctrl_(EmptyGroup()), + settings_(0, HashtablezInfoHandle(), hash, eq, alloc) { + if (bucket_count) { + capacity_ = NormalizeCapacity(bucket_count); + initialize_slots(); + } + } + + raw_hash_set(size_t bucket_count, const hasher& hash, + const allocator_type& alloc) + : raw_hash_set(bucket_count, hash, key_equal(), alloc) {} + + raw_hash_set(size_t bucket_count, const allocator_type& alloc) + : raw_hash_set(bucket_count, hasher(), key_equal(), alloc) {} + + explicit raw_hash_set(const allocator_type& alloc) + : raw_hash_set(0, hasher(), key_equal(), alloc) {} + + template + raw_hash_set(InputIter first, InputIter last, size_t bucket_count = 0, + const hasher& hash = hasher(), const key_equal& eq = key_equal(), + const allocator_type& alloc = allocator_type()) + : raw_hash_set(SelectBucketCountForIterRange(first, last, bucket_count), + hash, eq, alloc) { + insert(first, last); + } + + template + raw_hash_set(InputIter first, InputIter last, size_t bucket_count, + const hasher& hash, const allocator_type& alloc) + : raw_hash_set(first, last, bucket_count, hash, key_equal(), alloc) {} + + template + raw_hash_set(InputIter first, InputIter last, size_t bucket_count, + const allocator_type& alloc) + : raw_hash_set(first, last, bucket_count, hasher(), key_equal(), alloc) {} + + template + raw_hash_set(InputIter first, InputIter last, const allocator_type& alloc) + : raw_hash_set(first, last, 0, hasher(), key_equal(), alloc) {} + + // Instead of accepting std::initializer_list as the first + // argument like std::unordered_set does, we have two overloads + // that accept std::initializer_list and std::initializer_list. + // This is advantageous for performance. + // + // // Turns {"abc", "def"} into std::initializer_list, then + // // copies the strings into the set. + // std::unordered_set s = {"abc", "def"}; + // + // // Turns {"abc", "def"} into std::initializer_list, then + // // copies the strings into the set. + // absl::flat_hash_set s = {"abc", "def"}; + // + // The same trick is used in insert(). + // + // The enabler is necessary to prevent this constructor from triggering where + // the copy constructor is meant to be called. + // + // absl::flat_hash_set a, b{a}; + // + // RequiresNotInit is a workaround for gcc prior to 7.1. + template = 0, RequiresInsertable = 0> + raw_hash_set(std::initializer_list init, size_t bucket_count = 0, + const hasher& hash = hasher(), const key_equal& eq = key_equal(), + const allocator_type& alloc = allocator_type()) + : raw_hash_set(init.begin(), init.end(), bucket_count, hash, eq, alloc) {} + + raw_hash_set(std::initializer_list init, size_t bucket_count = 0, + const hasher& hash = hasher(), const key_equal& eq = key_equal(), + const allocator_type& alloc = allocator_type()) + : raw_hash_set(init.begin(), init.end(), bucket_count, hash, eq, alloc) {} + + template = 0, RequiresInsertable = 0> + raw_hash_set(std::initializer_list init, size_t bucket_count, + const hasher& hash, const allocator_type& alloc) + : raw_hash_set(init, bucket_count, hash, key_equal(), alloc) {} + + raw_hash_set(std::initializer_list init, size_t bucket_count, + const hasher& hash, const allocator_type& alloc) + : raw_hash_set(init, bucket_count, hash, key_equal(), alloc) {} + + template = 0, RequiresInsertable = 0> + raw_hash_set(std::initializer_list init, size_t bucket_count, + const allocator_type& alloc) + : raw_hash_set(init, bucket_count, hasher(), key_equal(), alloc) {} + + raw_hash_set(std::initializer_list init, size_t bucket_count, + const allocator_type& alloc) + : raw_hash_set(init, bucket_count, hasher(), key_equal(), alloc) {} + + template = 0, RequiresInsertable = 0> + raw_hash_set(std::initializer_list init, const allocator_type& alloc) + : raw_hash_set(init, 0, hasher(), key_equal(), alloc) {} + + raw_hash_set(std::initializer_list init, + const allocator_type& alloc) + : raw_hash_set(init, 0, hasher(), key_equal(), alloc) {} + + raw_hash_set(const raw_hash_set& that) + : raw_hash_set(that, AllocTraits::select_on_container_copy_construction( + that.alloc_ref())) {} + + raw_hash_set(const raw_hash_set& that, const allocator_type& a) + : raw_hash_set(0, that.hash_ref(), that.eq_ref(), a) { + reserve(that.size()); + // Because the table is guaranteed to be empty, we can do something faster + // than a full `insert`. + for (const auto& v : that) { + const size_t hash = PolicyTraits::apply(HashElement{hash_ref()}, v); + auto target = find_first_non_full(ctrl_, hash, capacity_); + SetCtrl(target.offset, H2(hash), capacity_, ctrl_, slots_, + sizeof(slot_type)); + emplace_at(target.offset, v); + infoz().RecordInsert(hash, target.probe_length); + } + size_ = that.size(); + growth_left() -= that.size(); + } + + raw_hash_set(raw_hash_set&& that) noexcept( + std::is_nothrow_copy_constructible::value&& + std::is_nothrow_copy_constructible::value&& + std::is_nothrow_copy_constructible::value) + : ctrl_(absl::exchange(that.ctrl_, EmptyGroup())), + slots_(absl::exchange(that.slots_, nullptr)), + size_(absl::exchange(that.size_, 0)), + capacity_(absl::exchange(that.capacity_, 0)), + // Hash, equality and allocator are copied instead of moved because + // `that` must be left valid. If Hash is std::function, moving it + // would create a nullptr functor that cannot be called. + settings_(absl::exchange(that.growth_left(), 0), + absl::exchange(that.infoz(), HashtablezInfoHandle()), + that.hash_ref(), that.eq_ref(), that.alloc_ref()) {} + + raw_hash_set(raw_hash_set&& that, const allocator_type& a) + : ctrl_(EmptyGroup()), + slots_(nullptr), + size_(0), + capacity_(0), + settings_(0, HashtablezInfoHandle(), that.hash_ref(), that.eq_ref(), + a) { + if (a == that.alloc_ref()) { + std::swap(ctrl_, that.ctrl_); + std::swap(slots_, that.slots_); + std::swap(size_, that.size_); + std::swap(capacity_, that.capacity_); + std::swap(growth_left(), that.growth_left()); + std::swap(infoz(), that.infoz()); + } else { + reserve(that.size()); + // Note: this will copy elements of dense_set and unordered_set instead of + // moving them. This can be fixed if it ever becomes an issue. + for (auto& elem : that) insert(std::move(elem)); + } + } + + raw_hash_set& operator=(const raw_hash_set& that) { + raw_hash_set tmp(that, + AllocTraits::propagate_on_container_copy_assignment::value + ? that.alloc_ref() + : alloc_ref()); + swap(tmp); + return *this; + } + + raw_hash_set& operator=(raw_hash_set&& that) noexcept( + absl::allocator_traits::is_always_equal::value&& + std::is_nothrow_move_assignable::value&& + std::is_nothrow_move_assignable::value) { + // TODO(sbenza): We should only use the operations from the noexcept clause + // to make sure we actually adhere to that contract. + return move_assign( + std::move(that), + typename AllocTraits::propagate_on_container_move_assignment()); + } + + ~raw_hash_set() { destroy_slots(); } + + iterator begin() { + auto it = iterator_at(0); + it.skip_empty_or_deleted(); + return it; + } + iterator end() { return {}; } + + const_iterator begin() const { + return const_cast(this)->begin(); + } + const_iterator end() const { return {}; } + const_iterator cbegin() const { return begin(); } + const_iterator cend() const { return end(); } + + bool empty() const { return !size(); } + size_t size() const { return size_; } + size_t capacity() const { return capacity_; } + size_t max_size() const { return (std::numeric_limits::max)(); } + + ABSL_ATTRIBUTE_REINITIALIZES void clear() { + // Iterating over this container is O(bucket_count()). When bucket_count() + // is much greater than size(), iteration becomes prohibitively expensive. + // For clear() it is more important to reuse the allocated array when the + // container is small because allocation takes comparatively long time + // compared to destruction of the elements of the container. So we pick the + // largest bucket_count() threshold for which iteration is still fast and + // past that we simply deallocate the array. + if (capacity_ > 127) { + destroy_slots(); + + infoz().RecordClearedReservation(); + } else if (capacity_) { + for (size_t i = 0; i != capacity_; ++i) { + if (IsFull(ctrl_[i])) { + PolicyTraits::destroy(&alloc_ref(), slots_ + i); + } + } + size_ = 0; + ResetCtrl(capacity_, ctrl_, slots_, sizeof(slot_type)); + reset_growth_left(); + } + assert(empty()); + infoz().RecordStorageChanged(0, capacity_); + } + + // This overload kicks in when the argument is an rvalue of insertable and + // decomposable type other than init_type. + // + // flat_hash_map m; + // m.insert(std::make_pair("abc", 42)); + // TODO(cheshire): A type alias T2 is introduced as a workaround for the nvcc + // bug. + template = 0, class T2 = T, + typename std::enable_if::value, int>::type = 0, + T* = nullptr> + std::pair insert(T&& value) { + return emplace(std::forward(value)); + } + + // This overload kicks in when the argument is a bitfield or an lvalue of + // insertable and decomposable type. + // + // union { int n : 1; }; + // flat_hash_set s; + // s.insert(n); + // + // flat_hash_set s; + // const char* p = "hello"; + // s.insert(p); + // + // TODO(romanp): Once we stop supporting gcc 5.1 and below, replace + // RequiresInsertable with RequiresInsertable. + // We are hitting this bug: https://godbolt.org/g/1Vht4f. + template < + class T, RequiresInsertable = 0, + typename std::enable_if::value, int>::type = 0> + std::pair insert(const T& value) { + return emplace(value); + } + + // This overload kicks in when the argument is an rvalue of init_type. Its + // purpose is to handle brace-init-list arguments. + // + // flat_hash_map s; + // s.insert({"abc", 42}); + std::pair insert(init_type&& value) { + return emplace(std::move(value)); + } + + // TODO(cheshire): A type alias T2 is introduced as a workaround for the nvcc + // bug. + template = 0, class T2 = T, + typename std::enable_if::value, int>::type = 0, + T* = nullptr> + iterator insert(const_iterator, T&& value) { + return insert(std::forward(value)).first; + } + + // TODO(romanp): Once we stop supporting gcc 5.1 and below, replace + // RequiresInsertable with RequiresInsertable. + // We are hitting this bug: https://godbolt.org/g/1Vht4f. + template < + class T, RequiresInsertable = 0, + typename std::enable_if::value, int>::type = 0> + iterator insert(const_iterator, const T& value) { + return insert(value).first; + } + + iterator insert(const_iterator, init_type&& value) { + return insert(std::move(value)).first; + } + + template + void insert(InputIt first, InputIt last) { + for (; first != last; ++first) emplace(*first); + } + + template = 0, RequiresInsertable = 0> + void insert(std::initializer_list ilist) { + insert(ilist.begin(), ilist.end()); + } + + void insert(std::initializer_list ilist) { + insert(ilist.begin(), ilist.end()); + } + + insert_return_type insert(node_type&& node) { + if (!node) return {end(), false, node_type()}; + const auto& elem = PolicyTraits::element(CommonAccess::GetSlot(node)); + auto res = PolicyTraits::apply( + InsertSlot{*this, std::move(*CommonAccess::GetSlot(node))}, + elem); + if (res.second) { + CommonAccess::Reset(&node); + return {res.first, true, node_type()}; + } else { + return {res.first, false, std::move(node)}; + } + } + + iterator insert(const_iterator, node_type&& node) { + auto res = insert(std::move(node)); + node = std::move(res.node); + return res.position; + } + + // This overload kicks in if we can deduce the key from args. This enables us + // to avoid constructing value_type if an entry with the same key already + // exists. + // + // For example: + // + // flat_hash_map m = {{"abc", "def"}}; + // // Creates no std::string copies and makes no heap allocations. + // m.emplace("abc", "xyz"); + template ::value, int>::type = 0> + std::pair emplace(Args&&... args) { + return PolicyTraits::apply(EmplaceDecomposable{*this}, + std::forward(args)...); + } + + // This overload kicks in if we cannot deduce the key from args. It constructs + // value_type unconditionally and then either moves it into the table or + // destroys. + template ::value, int>::type = 0> + std::pair emplace(Args&&... args) { + alignas(slot_type) unsigned char raw[sizeof(slot_type)]; + slot_type* slot = reinterpret_cast(&raw); + + PolicyTraits::construct(&alloc_ref(), slot, std::forward(args)...); + const auto& elem = PolicyTraits::element(slot); + return PolicyTraits::apply(InsertSlot{*this, std::move(*slot)}, elem); + } + + template + iterator emplace_hint(const_iterator, Args&&... args) { + return emplace(std::forward(args)...).first; + } + + // Extension API: support for lazy emplace. + // + // Looks up key in the table. If found, returns the iterator to the element. + // Otherwise calls `f` with one argument of type `raw_hash_set::constructor`. + // + // `f` must abide by several restrictions: + // - it MUST call `raw_hash_set::constructor` with arguments as if a + // `raw_hash_set::value_type` is constructed, + // - it MUST NOT access the container before the call to + // `raw_hash_set::constructor`, and + // - it MUST NOT erase the lazily emplaced element. + // Doing any of these is undefined behavior. + // + // For example: + // + // std::unordered_set s; + // // Makes ArenaStr even if "abc" is in the map. + // s.insert(ArenaString(&arena, "abc")); + // + // flat_hash_set s; + // // Makes ArenaStr only if "abc" is not in the map. + // s.lazy_emplace("abc", [&](const constructor& ctor) { + // ctor(&arena, "abc"); + // }); + // + // WARNING: This API is currently experimental. If there is a way to implement + // the same thing with the rest of the API, prefer that. + class constructor { + friend class raw_hash_set; + + public: + template + void operator()(Args&&... args) const { + assert(*slot_); + PolicyTraits::construct(alloc_, *slot_, std::forward(args)...); + *slot_ = nullptr; + } + + private: + constructor(allocator_type* a, slot_type** slot) : alloc_(a), slot_(slot) {} + + allocator_type* alloc_; + slot_type** slot_; + }; + + template + iterator lazy_emplace(const key_arg& key, F&& f) { + auto res = find_or_prepare_insert(key); + if (res.second) { + slot_type* slot = slots_ + res.first; + std::forward(f)(constructor(&alloc_ref(), &slot)); + assert(!slot); + } + return iterator_at(res.first); + } + + // Extension API: support for heterogeneous keys. + // + // std::unordered_set s; + // // Turns "abc" into std::string. + // s.erase("abc"); + // + // flat_hash_set s; + // // Uses "abc" directly without copying it into std::string. + // s.erase("abc"); + template + size_type erase(const key_arg& key) { + auto it = find(key); + if (it == end()) return 0; + erase(it); + return 1; + } + + // Erases the element pointed to by `it`. Unlike `std::unordered_set::erase`, + // this method returns void to reduce algorithmic complexity to O(1). The + // iterator is invalidated, so any increment should be done before calling + // erase. In order to erase while iterating across a map, use the following + // idiom (which also works for standard containers): + // + // for (auto it = m.begin(), end = m.end(); it != end;) { + // // `erase()` will invalidate `it`, so advance `it` first. + // auto copy_it = it++; + // if () { + // m.erase(copy_it); + // } + // } + void erase(const_iterator cit) { erase(cit.inner_); } + + // This overload is necessary because otherwise erase(const K&) would be + // a better match if non-const iterator is passed as an argument. + void erase(iterator it) { + ABSL_INTERNAL_ASSERT_IS_FULL(it.ctrl_, + "erase() called on invalid iterator."); + PolicyTraits::destroy(&alloc_ref(), it.slot_); + erase_meta_only(it); + } + + iterator erase(const_iterator first, const_iterator last) { + while (first != last) { + erase(first++); + } + return last.inner_; + } + + // Moves elements from `src` into `this`. + // If the element already exists in `this`, it is left unmodified in `src`. + template + void merge(raw_hash_set& src) { // NOLINT + assert(this != &src); + for (auto it = src.begin(), e = src.end(); it != e;) { + auto next = std::next(it); + if (PolicyTraits::apply(InsertSlot{*this, std::move(*it.slot_)}, + PolicyTraits::element(it.slot_)) + .second) { + src.erase_meta_only(it); + } + it = next; + } + } + + template + void merge(raw_hash_set&& src) { + merge(src); + } + + node_type extract(const_iterator position) { + ABSL_INTERNAL_ASSERT_IS_FULL(position.inner_.ctrl_, + "extract() called on invalid iterator."); + auto node = + CommonAccess::Transfer(alloc_ref(), position.inner_.slot_); + erase_meta_only(position); + return node; + } + + template < + class K = key_type, + typename std::enable_if::value, int>::type = 0> + node_type extract(const key_arg& key) { + auto it = find(key); + return it == end() ? node_type() : extract(const_iterator{it}); + } + + void swap(raw_hash_set& that) noexcept( + IsNoThrowSwappable() && IsNoThrowSwappable() && + IsNoThrowSwappable( + typename AllocTraits::propagate_on_container_swap{})) { + using std::swap; + swap(ctrl_, that.ctrl_); + swap(slots_, that.slots_); + swap(size_, that.size_); + swap(capacity_, that.capacity_); + swap(growth_left(), that.growth_left()); + swap(hash_ref(), that.hash_ref()); + swap(eq_ref(), that.eq_ref()); + swap(infoz(), that.infoz()); + SwapAlloc(alloc_ref(), that.alloc_ref(), + typename AllocTraits::propagate_on_container_swap{}); + } + + void rehash(size_t n) { + if (n == 0 && capacity_ == 0) return; + if (n == 0 && size_ == 0) { + destroy_slots(); + infoz().RecordStorageChanged(0, 0); + infoz().RecordClearedReservation(); + return; + } + + // bitor is a faster way of doing `max` here. We will round up to the next + // power-of-2-minus-1, so bitor is good enough. + auto m = NormalizeCapacity(n | GrowthToLowerboundCapacity(size())); + // n == 0 unconditionally rehashes as per the standard. + if (n == 0 || m > capacity_) { + resize(m); + + // This is after resize, to ensure that we have completed the allocation + // and have potentially sampled the hashtable. + infoz().RecordReservation(n); + } + } + + void reserve(size_t n) { + if (n > size() + growth_left()) { + size_t m = GrowthToLowerboundCapacity(n); + resize(NormalizeCapacity(m)); + + // This is after resize, to ensure that we have completed the allocation + // and have potentially sampled the hashtable. + infoz().RecordReservation(n); + } + } + + // Extension API: support for heterogeneous keys. + // + // std::unordered_set s; + // // Turns "abc" into std::string. + // s.count("abc"); + // + // ch_set s; + // // Uses "abc" directly without copying it into std::string. + // s.count("abc"); + template + size_t count(const key_arg& key) const { + return find(key) == end() ? 0 : 1; + } + + // Issues CPU prefetch instructions for the memory needed to find or insert + // a key. Like all lookup functions, this support heterogeneous keys. + // + // NOTE: This is a very low level operation and should not be used without + // specific benchmarks indicating its importance. + template + void prefetch(const key_arg& key) const { + (void)key; + // Avoid probing if we won't be able to prefetch the addresses received. +#ifdef ABSL_INTERNAL_HAVE_PREFETCH + prefetch_heap_block(); + auto seq = probe(ctrl_, hash_ref()(key), capacity_); + base_internal::PrefetchT0(ctrl_ + seq.offset()); + base_internal::PrefetchT0(slots_ + seq.offset()); +#endif // ABSL_INTERNAL_HAVE_PREFETCH + } + + // The API of find() has two extensions. + // + // 1. The hash can be passed by the user. It must be equal to the hash of the + // key. + // + // 2. The type of the key argument doesn't have to be key_type. This is so + // called heterogeneous key support. + template + iterator find(const key_arg& key, size_t hash) { + auto seq = probe(ctrl_, hash, capacity_); + while (true) { + Group g{ctrl_ + seq.offset()}; + for (uint32_t i : g.Match(H2(hash))) { + if (ABSL_PREDICT_TRUE(PolicyTraits::apply( + EqualElement{key, eq_ref()}, + PolicyTraits::element(slots_ + seq.offset(i))))) + return iterator_at(seq.offset(i)); + } + if (ABSL_PREDICT_TRUE(g.MaskEmpty())) return end(); + seq.next(); + assert(seq.index() <= capacity_ && "full table!"); + } + } + template + iterator find(const key_arg& key) { + prefetch_heap_block(); + return find(key, hash_ref()(key)); + } + + template + const_iterator find(const key_arg& key, size_t hash) const { + return const_cast(this)->find(key, hash); + } + template + const_iterator find(const key_arg& key) const { + prefetch_heap_block(); + return find(key, hash_ref()(key)); + } + + template + bool contains(const key_arg& key) const { + return find(key) != end(); + } + + template + std::pair equal_range(const key_arg& key) { + auto it = find(key); + if (it != end()) return {it, std::next(it)}; + return {it, it}; + } + template + std::pair equal_range( + const key_arg& key) const { + auto it = find(key); + if (it != end()) return {it, std::next(it)}; + return {it, it}; + } + + size_t bucket_count() const { return capacity_; } + float load_factor() const { + return capacity_ ? static_cast(size()) / capacity_ : 0.0; + } + float max_load_factor() const { return 1.0f; } + void max_load_factor(float) { + // Does nothing. + } + + hasher hash_function() const { return hash_ref(); } + key_equal key_eq() const { return eq_ref(); } + allocator_type get_allocator() const { return alloc_ref(); } + + friend bool operator==(const raw_hash_set& a, const raw_hash_set& b) { + if (a.size() != b.size()) return false; + const raw_hash_set* outer = &a; + const raw_hash_set* inner = &b; + if (outer->capacity() > inner->capacity()) std::swap(outer, inner); + for (const value_type& elem : *outer) + if (!inner->has_element(elem)) return false; + return true; + } + + friend bool operator!=(const raw_hash_set& a, const raw_hash_set& b) { + return !(a == b); + } + + template + friend typename std::enable_if::value, + H>::type + AbslHashValue(H h, const raw_hash_set& s) { + return H::combine(H::combine_unordered(std::move(h), s.begin(), s.end()), + s.size()); + } + + friend void swap(raw_hash_set& a, + raw_hash_set& b) noexcept(noexcept(a.swap(b))) { + a.swap(b); + } + + private: + template + friend struct absl::container_internal::hashtable_debug_internal:: + HashtableDebugAccess; + + struct FindElement { + template + const_iterator operator()(const K& key, Args&&...) const { + return s.find(key); + } + const raw_hash_set& s; + }; + + struct HashElement { + template + size_t operator()(const K& key, Args&&...) const { + return h(key); + } + const hasher& h; + }; + + template + struct EqualElement { + template + bool operator()(const K2& lhs, Args&&...) const { + return eq(lhs, rhs); + } + const K1& rhs; + const key_equal& eq; + }; + + struct EmplaceDecomposable { + template + std::pair operator()(const K& key, Args&&... args) const { + auto res = s.find_or_prepare_insert(key); + if (res.second) { + s.emplace_at(res.first, std::forward(args)...); + } + return {s.iterator_at(res.first), res.second}; + } + raw_hash_set& s; + }; + + template + struct InsertSlot { + template + std::pair operator()(const K& key, Args&&...) && { + auto res = s.find_or_prepare_insert(key); + if (res.second) { + PolicyTraits::transfer(&s.alloc_ref(), s.slots_ + res.first, &slot); + } else if (do_destroy) { + PolicyTraits::destroy(&s.alloc_ref(), &slot); + } + return {s.iterator_at(res.first), res.second}; + } + raw_hash_set& s; + // Constructed slot. Either moved into place or destroyed. + slot_type&& slot; + }; + + // Erases, but does not destroy, the value pointed to by `it`. + // + // This merely updates the pertinent control byte. This can be used in + // conjunction with Policy::transfer to move the object to another place. + void erase_meta_only(const_iterator it) { + assert(IsFull(*it.inner_.ctrl_) && "erasing a dangling iterator"); + --size_; + const size_t index = static_cast(it.inner_.ctrl_ - ctrl_); + const size_t index_before = (index - Group::kWidth) & capacity_; + const auto empty_after = Group(it.inner_.ctrl_).MaskEmpty(); + const auto empty_before = Group(ctrl_ + index_before).MaskEmpty(); + + // We count how many consecutive non empties we have to the right and to the + // left of `it`. If the sum is >= kWidth then there is at least one probe + // window that might have seen a full group. + bool was_never_full = + empty_before && empty_after && + static_cast(empty_after.TrailingZeros() + + empty_before.LeadingZeros()) < Group::kWidth; + + SetCtrl(index, was_never_full ? ctrl_t::kEmpty : ctrl_t::kDeleted, + capacity_, ctrl_, slots_, sizeof(slot_type)); + growth_left() += was_never_full; + infoz().RecordErase(); + } + + // Allocates a backing array for `self` and initializes its control bytes. + // This reads `capacity_` and updates all other fields based on the result of + // the allocation. + // + // This does not free the currently held array; `capacity_` must be nonzero. + void initialize_slots() { + assert(capacity_); + // Folks with custom allocators often make unwarranted assumptions about the + // behavior of their classes vis-a-vis trivial destructability and what + // calls they will or wont make. Avoid sampling for people with custom + // allocators to get us out of this mess. This is not a hard guarantee but + // a workaround while we plan the exact guarantee we want to provide. + // + // People are often sloppy with the exact type of their allocator (sometimes + // it has an extra const or is missing the pair, but rebinds made it work + // anyway). To avoid the ambiguity, we work off SlotAlloc which we have + // bound more carefully. + if (std::is_same>::value && + slots_ == nullptr) { + infoz() = Sample(sizeof(slot_type)); + } + + char* mem = static_cast(Allocate( + &alloc_ref(), + AllocSize(capacity_, sizeof(slot_type), alignof(slot_type)))); + ctrl_ = reinterpret_cast(mem); + slots_ = reinterpret_cast( + mem + SlotOffset(capacity_, alignof(slot_type))); + ResetCtrl(capacity_, ctrl_, slots_, sizeof(slot_type)); + reset_growth_left(); + infoz().RecordStorageChanged(size_, capacity_); + } + + // Destroys all slots in the backing array, frees the backing array, and + // clears all top-level book-keeping data. + // + // This essentially implements `map = raw_hash_set();`. + void destroy_slots() { + if (!capacity_) return; + for (size_t i = 0; i != capacity_; ++i) { + if (IsFull(ctrl_[i])) { + PolicyTraits::destroy(&alloc_ref(), slots_ + i); + } + } + + // Unpoison before returning the memory to the allocator. + SanitizerUnpoisonMemoryRegion(slots_, sizeof(slot_type) * capacity_); + Deallocate( + &alloc_ref(), ctrl_, + AllocSize(capacity_, sizeof(slot_type), alignof(slot_type))); + ctrl_ = EmptyGroup(); + slots_ = nullptr; + size_ = 0; + capacity_ = 0; + growth_left() = 0; + } + + void resize(size_t new_capacity) { + assert(IsValidCapacity(new_capacity)); + auto* old_ctrl = ctrl_; + auto* old_slots = slots_; + const size_t old_capacity = capacity_; + capacity_ = new_capacity; + initialize_slots(); + + size_t total_probe_length = 0; + for (size_t i = 0; i != old_capacity; ++i) { + if (IsFull(old_ctrl[i])) { + size_t hash = PolicyTraits::apply(HashElement{hash_ref()}, + PolicyTraits::element(old_slots + i)); + auto target = find_first_non_full(ctrl_, hash, capacity_); + size_t new_i = target.offset; + total_probe_length += target.probe_length; + SetCtrl(new_i, H2(hash), capacity_, ctrl_, slots_, sizeof(slot_type)); + PolicyTraits::transfer(&alloc_ref(), slots_ + new_i, old_slots + i); + } + } + if (old_capacity) { + SanitizerUnpoisonMemoryRegion(old_slots, + sizeof(slot_type) * old_capacity); + Deallocate( + &alloc_ref(), old_ctrl, + AllocSize(old_capacity, sizeof(slot_type), alignof(slot_type))); + } + infoz().RecordRehash(total_probe_length); + } + + // Prunes control bytes to remove as many tombstones as possible. + // + // See the comment on `rehash_and_grow_if_necessary()`. + void drop_deletes_without_resize() ABSL_ATTRIBUTE_NOINLINE { + assert(IsValidCapacity(capacity_)); + assert(!is_small(capacity_)); + // Algorithm: + // - mark all DELETED slots as EMPTY + // - mark all FULL slots as DELETED + // - for each slot marked as DELETED + // hash = Hash(element) + // target = find_first_non_full(hash) + // if target is in the same group + // mark slot as FULL + // else if target is EMPTY + // transfer element to target + // mark slot as EMPTY + // mark target as FULL + // else if target is DELETED + // swap current element with target element + // mark target as FULL + // repeat procedure for current slot with moved from element (target) + ConvertDeletedToEmptyAndFullToDeleted(ctrl_, capacity_); + alignas(slot_type) unsigned char raw[sizeof(slot_type)]; + size_t total_probe_length = 0; + slot_type* slot = reinterpret_cast(&raw); + for (size_t i = 0; i != capacity_; ++i) { + if (!IsDeleted(ctrl_[i])) continue; + const size_t hash = PolicyTraits::apply( + HashElement{hash_ref()}, PolicyTraits::element(slots_ + i)); + const FindInfo target = find_first_non_full(ctrl_, hash, capacity_); + const size_t new_i = target.offset; + total_probe_length += target.probe_length; + + // Verify if the old and new i fall within the same group wrt the hash. + // If they do, we don't need to move the object as it falls already in the + // best probe we can. + const size_t probe_offset = probe(ctrl_, hash, capacity_).offset(); + const auto probe_index = [probe_offset, this](size_t pos) { + return ((pos - probe_offset) & capacity_) / Group::kWidth; + }; + + // Element doesn't move. + if (ABSL_PREDICT_TRUE(probe_index(new_i) == probe_index(i))) { + SetCtrl(i, H2(hash), capacity_, ctrl_, slots_, sizeof(slot_type)); + continue; + } + if (IsEmpty(ctrl_[new_i])) { + // Transfer element to the empty spot. + // SetCtrl poisons/unpoisons the slots so we have to call it at the + // right time. + SetCtrl(new_i, H2(hash), capacity_, ctrl_, slots_, sizeof(slot_type)); + PolicyTraits::transfer(&alloc_ref(), slots_ + new_i, slots_ + i); + SetCtrl(i, ctrl_t::kEmpty, capacity_, ctrl_, slots_, sizeof(slot_type)); + } else { + assert(IsDeleted(ctrl_[new_i])); + SetCtrl(new_i, H2(hash), capacity_, ctrl_, slots_, sizeof(slot_type)); + // Until we are done rehashing, DELETED marks previously FULL slots. + // Swap i and new_i elements. + PolicyTraits::transfer(&alloc_ref(), slot, slots_ + i); + PolicyTraits::transfer(&alloc_ref(), slots_ + i, slots_ + new_i); + PolicyTraits::transfer(&alloc_ref(), slots_ + new_i, slot); + --i; // repeat + } + } + reset_growth_left(); + infoz().RecordRehash(total_probe_length); + } + + // Called whenever the table *might* need to conditionally grow. + // + // This function is an optimization opportunity to perform a rehash even when + // growth is unnecessary, because vacating tombstones is beneficial for + // performance in the long-run. + void rehash_and_grow_if_necessary() { + if (capacity_ == 0) { + resize(1); + } else if (capacity_ > Group::kWidth && + // Do these calcuations in 64-bit to avoid overflow. + size() * uint64_t{32} <= capacity_ * uint64_t{25}) { + // Squash DELETED without growing if there is enough capacity. + // + // Rehash in place if the current size is <= 25/32 of capacity_. + // Rationale for such a high factor: 1) drop_deletes_without_resize() is + // faster than resize, and 2) it takes quite a bit of work to add + // tombstones. In the worst case, seems to take approximately 4 + // insert/erase pairs to create a single tombstone and so if we are + // rehashing because of tombstones, we can afford to rehash-in-place as + // long as we are reclaiming at least 1/8 the capacity without doing more + // than 2X the work. (Where "work" is defined to be size() for rehashing + // or rehashing in place, and 1 for an insert or erase.) But rehashing in + // place is faster per operation than inserting or even doubling the size + // of the table, so we actually afford to reclaim even less space from a + // resize-in-place. The decision is to rehash in place if we can reclaim + // at about 1/8th of the usable capacity (specifically 3/28 of the + // capacity) which means that the total cost of rehashing will be a small + // fraction of the total work. + // + // Here is output of an experiment using the BM_CacheInSteadyState + // benchmark running the old case (where we rehash-in-place only if we can + // reclaim at least 7/16*capacity_) vs. this code (which rehashes in place + // if we can recover 3/32*capacity_). + // + // Note that although in the worst-case number of rehashes jumped up from + // 15 to 190, but the number of operations per second is almost the same. + // + // Abridged output of running BM_CacheInSteadyState benchmark from + // raw_hash_set_benchmark. N is the number of insert/erase operations. + // + // | OLD (recover >= 7/16 | NEW (recover >= 3/32) + // size | N/s LoadFactor NRehashes | N/s LoadFactor NRehashes + // 448 | 145284 0.44 18 | 140118 0.44 19 + // 493 | 152546 0.24 11 | 151417 0.48 28 + // 538 | 151439 0.26 11 | 151152 0.53 38 + // 583 | 151765 0.28 11 | 150572 0.57 50 + // 628 | 150241 0.31 11 | 150853 0.61 66 + // 672 | 149602 0.33 12 | 150110 0.66 90 + // 717 | 149998 0.35 12 | 149531 0.70 129 + // 762 | 149836 0.37 13 | 148559 0.74 190 + // 807 | 149736 0.39 14 | 151107 0.39 14 + // 852 | 150204 0.42 15 | 151019 0.42 15 + drop_deletes_without_resize(); + } else { + // Otherwise grow the container. + resize(capacity_ * 2 + 1); + } + } + + bool has_element(const value_type& elem) const { + size_t hash = PolicyTraits::apply(HashElement{hash_ref()}, elem); + auto seq = probe(ctrl_, hash, capacity_); + while (true) { + Group g{ctrl_ + seq.offset()}; + for (uint32_t i : g.Match(H2(hash))) { + if (ABSL_PREDICT_TRUE(PolicyTraits::element(slots_ + seq.offset(i)) == + elem)) + return true; + } + if (ABSL_PREDICT_TRUE(g.MaskEmpty())) return false; + seq.next(); + assert(seq.index() <= capacity_ && "full table!"); + } + return false; + } + + // TODO(alkis): Optimize this assuming *this and that don't overlap. + raw_hash_set& move_assign(raw_hash_set&& that, std::true_type) { + raw_hash_set tmp(std::move(that)); + swap(tmp); + return *this; + } + raw_hash_set& move_assign(raw_hash_set&& that, std::false_type) { + raw_hash_set tmp(std::move(that), alloc_ref()); + swap(tmp); + return *this; + } + + protected: + // Attempts to find `key` in the table; if it isn't found, returns a slot that + // the value can be inserted into, with the control byte already set to + // `key`'s H2. + template + std::pair find_or_prepare_insert(const K& key) { + prefetch_heap_block(); + auto hash = hash_ref()(key); + auto seq = probe(ctrl_, hash, capacity_); + while (true) { + Group g{ctrl_ + seq.offset()}; + for (uint32_t i : g.Match(H2(hash))) { + if (ABSL_PREDICT_TRUE(PolicyTraits::apply( + EqualElement{key, eq_ref()}, + PolicyTraits::element(slots_ + seq.offset(i))))) + return {seq.offset(i), false}; + } + if (ABSL_PREDICT_TRUE(g.MaskEmpty())) break; + seq.next(); + assert(seq.index() <= capacity_ && "full table!"); + } + return {prepare_insert(hash), true}; + } + + // Given the hash of a value not currently in the table, finds the next + // viable slot index to insert it at. + // + // REQUIRES: At least one non-full slot available. + size_t prepare_insert(size_t hash) ABSL_ATTRIBUTE_NOINLINE { + auto target = find_first_non_full(ctrl_, hash, capacity_); + if (ABSL_PREDICT_FALSE(growth_left() == 0 && + !IsDeleted(ctrl_[target.offset]))) { + rehash_and_grow_if_necessary(); + target = find_first_non_full(ctrl_, hash, capacity_); + } + ++size_; + growth_left() -= IsEmpty(ctrl_[target.offset]); + SetCtrl(target.offset, H2(hash), capacity_, ctrl_, slots_, + sizeof(slot_type)); + infoz().RecordInsert(hash, target.probe_length); + return target.offset; + } + + // Constructs the value in the space pointed by the iterator. This only works + // after an unsuccessful find_or_prepare_insert() and before any other + // modifications happen in the raw_hash_set. + // + // PRECONDITION: i is an index returned from find_or_prepare_insert(k), where + // k is the key decomposed from `forward(args)...`, and the bool + // returned by find_or_prepare_insert(k) was true. + // POSTCONDITION: *m.iterator_at(i) == value_type(forward(args)...). + template + void emplace_at(size_t i, Args&&... args) { + PolicyTraits::construct(&alloc_ref(), slots_ + i, + std::forward(args)...); + + assert(PolicyTraits::apply(FindElement{*this}, *iterator_at(i)) == + iterator_at(i) && + "constructed value does not match the lookup key"); + } + + iterator iterator_at(size_t i) { return {ctrl_ + i, slots_ + i}; } + const_iterator iterator_at(size_t i) const { return {ctrl_ + i, slots_ + i}; } + + private: + friend struct RawHashSetTestOnlyAccess; + + void reset_growth_left() { + growth_left() = CapacityToGrowth(capacity()) - size_; + } + + // The number of slots we can still fill without needing to rehash. + // + // This is stored separately due to tombstones: we do not include tombstones + // in the growth capacity, because we'd like to rehash when the table is + // otherwise filled with tombstones: otherwise, probe sequences might get + // unacceptably long without triggering a rehash. Callers can also force a + // rehash via the standard `rehash(0)`, which will recompute this value as a + // side-effect. + // + // See `CapacityToGrowth()`. + size_t& growth_left() { return settings_.template get<0>(); } + + // Prefetch the heap-allocated memory region to resolve potential TLB misses. + // This is intended to overlap with execution of calculating the hash for a + // key. + void prefetch_heap_block() const { + base_internal::PrefetchT2(ctrl_); + } + + HashtablezInfoHandle& infoz() { return settings_.template get<1>(); } + + hasher& hash_ref() { return settings_.template get<2>(); } + const hasher& hash_ref() const { return settings_.template get<2>(); } + key_equal& eq_ref() { return settings_.template get<3>(); } + const key_equal& eq_ref() const { return settings_.template get<3>(); } + allocator_type& alloc_ref() { return settings_.template get<4>(); } + const allocator_type& alloc_ref() const { + return settings_.template get<4>(); + } + + // TODO(alkis): Investigate removing some of these fields: + // - ctrl/slots can be derived from each other + // - size can be moved into the slot array + + // The control bytes (and, also, a pointer to the base of the backing array). + // + // This contains `capacity_ + 1 + NumClonedBytes()` entries, even + // when the table is empty (hence EmptyGroup). + ctrl_t* ctrl_ = EmptyGroup(); + // The beginning of the slots, located at `SlotOffset()` bytes after + // `ctrl_`. May be null for empty tables. + slot_type* slots_ = nullptr; + + // The number of filled slots. + size_t size_ = 0; + + // The total number of available slots. + size_t capacity_ = 0; + absl::container_internal::CompressedTuple + settings_{0u, HashtablezInfoHandle{}, hasher{}, key_equal{}, + allocator_type{}}; +}; + +// Erases all elements that satisfy the predicate `pred` from the container `c`. +template +typename raw_hash_set::size_type EraseIf( + Predicate& pred, raw_hash_set* c) { + const auto initial_size = c->size(); + for (auto it = c->begin(), last = c->end(); it != last;) { + if (pred(*it)) { + c->erase(it++); + } else { + ++it; + } + } + return initial_size - c->size(); +} + +namespace hashtable_debug_internal { +template +struct HashtableDebugAccess> { + using Traits = typename Set::PolicyTraits; + using Slot = typename Traits::slot_type; + + static size_t GetNumProbes(const Set& set, + const typename Set::key_type& key) { + size_t num_probes = 0; + size_t hash = set.hash_ref()(key); + auto seq = probe(set.ctrl_, hash, set.capacity_); + while (true) { + container_internal::Group g{set.ctrl_ + seq.offset()}; + for (uint32_t i : g.Match(container_internal::H2(hash))) { + if (Traits::apply( + typename Set::template EqualElement{ + key, set.eq_ref()}, + Traits::element(set.slots_ + seq.offset(i)))) + return num_probes; + ++num_probes; + } + if (g.MaskEmpty()) return num_probes; + seq.next(); + ++num_probes; + } + } + + static size_t AllocatedByteSize(const Set& c) { + size_t capacity = c.capacity_; + if (capacity == 0) return 0; + size_t m = AllocSize(capacity, sizeof(Slot), alignof(Slot)); + + size_t per_slot = Traits::space_used(static_cast(nullptr)); + if (per_slot != ~size_t{}) { + m += per_slot * c.size(); + } else { + for (size_t i = 0; i != capacity; ++i) { + if (container_internal::IsFull(c.ctrl_[i])) { + m += Traits::space_used(c.slots_ + i); + } + } + } + return m; + } + + static size_t LowerBoundAllocatedByteSize(size_t size) { + size_t capacity = GrowthToLowerboundCapacity(size); + if (capacity == 0) return 0; + size_t m = + AllocSize(NormalizeCapacity(capacity), sizeof(Slot), alignof(Slot)); + size_t per_slot = Traits::space_used(static_cast(nullptr)); + if (per_slot != ~size_t{}) { + m += per_slot * size; + } + return m; + } +}; + +} // namespace hashtable_debug_internal +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#undef ABSL_INTERNAL_ASSERT_IS_FULL + +#endif // ABSL_CONTAINER_INTERNAL_RAW_HASH_SET_H_ diff --git a/src/absl/container/internal/tracked.h b/src/absl/container/internal/tracked.h new file mode 100644 index 0000000..29f5829 --- /dev/null +++ b/src/absl/container/internal/tracked.h @@ -0,0 +1,83 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_CONTAINER_INTERNAL_TRACKED_H_ +#define ABSL_CONTAINER_INTERNAL_TRACKED_H_ + +#include + +#include +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { + +// A class that tracks its copies and moves so that it can be queried in tests. +template +class Tracked { + public: + Tracked() {} + // NOLINTNEXTLINE(runtime/explicit) + Tracked(const T& val) : val_(val) {} + Tracked(const Tracked& that) + : val_(that.val_), + num_moves_(that.num_moves_), + num_copies_(that.num_copies_) { + ++(*num_copies_); + } + Tracked(Tracked&& that) + : val_(std::move(that.val_)), + num_moves_(std::move(that.num_moves_)), + num_copies_(std::move(that.num_copies_)) { + ++(*num_moves_); + } + Tracked& operator=(const Tracked& that) { + val_ = that.val_; + num_moves_ = that.num_moves_; + num_copies_ = that.num_copies_; + ++(*num_copies_); + } + Tracked& operator=(Tracked&& that) { + val_ = std::move(that.val_); + num_moves_ = std::move(that.num_moves_); + num_copies_ = std::move(that.num_copies_); + ++(*num_moves_); + } + + const T& val() const { return val_; } + + friend bool operator==(const Tracked& a, const Tracked& b) { + return a.val_ == b.val_; + } + friend bool operator!=(const Tracked& a, const Tracked& b) { + return !(a == b); + } + + size_t num_copies() { return *num_copies_; } + size_t num_moves() { return *num_moves_; } + + private: + T val_; + std::shared_ptr num_moves_ = std::make_shared(0); + std::shared_ptr num_copies_ = std::make_shared(0); +}; + +} // namespace container_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_INTERNAL_TRACKED_H_ diff --git a/src/absl/container/node_hash_map.h b/src/absl/container/node_hash_map.h new file mode 100644 index 0000000..6868e63 --- /dev/null +++ b/src/absl/container/node_hash_map.h @@ -0,0 +1,604 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: node_hash_map.h +// ----------------------------------------------------------------------------- +// +// An `absl::node_hash_map` is an unordered associative container of +// unique keys and associated values designed to be a more efficient replacement +// for `std::unordered_map`. Like `unordered_map`, search, insertion, and +// deletion of map elements can be done as an `O(1)` operation. However, +// `node_hash_map` (and other unordered associative containers known as the +// collection of Abseil "Swiss tables") contain other optimizations that result +// in both memory and computation advantages. +// +// In most cases, your default choice for a hash map should be a map of type +// `flat_hash_map`. However, if you need pointer stability and cannot store +// a `flat_hash_map` with `unique_ptr` elements, a `node_hash_map` may be a +// valid alternative. As well, if you are migrating your code from using +// `std::unordered_map`, a `node_hash_map` provides a more straightforward +// migration, because it guarantees pointer stability. Consider migrating to +// `node_hash_map` and perhaps converting to a more efficient `flat_hash_map` +// upon further review. + +#ifndef ABSL_CONTAINER_NODE_HASH_MAP_H_ +#define ABSL_CONTAINER_NODE_HASH_MAP_H_ + +#include +#include +#include + +#include "absl/algorithm/container.h" +#include "absl/base/macros.h" +#include "absl/container/internal/container_memory.h" +#include "absl/container/internal/hash_function_defaults.h" // IWYU pragma: export +#include "absl/container/internal/node_slot_policy.h" +#include "absl/container/internal/raw_hash_map.h" // IWYU pragma: export +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { +template +class NodeHashMapPolicy; +} // namespace container_internal + +// ----------------------------------------------------------------------------- +// absl::node_hash_map +// ----------------------------------------------------------------------------- +// +// An `absl::node_hash_map` is an unordered associative container which +// has been optimized for both speed and memory footprint in most common use +// cases. Its interface is similar to that of `std::unordered_map` with +// the following notable differences: +// +// * Supports heterogeneous lookup, through `find()`, `operator[]()` and +// `insert()`, provided that the map is provided a compatible heterogeneous +// hashing function and equality operator. +// * Contains a `capacity()` member function indicating the number of element +// slots (open, deleted, and empty) within the hash map. +// * Returns `void` from the `erase(iterator)` overload. +// +// By default, `node_hash_map` uses the `absl::Hash` hashing framework. +// All fundamental and Abseil types that support the `absl::Hash` framework have +// a compatible equality operator for comparing insertions into `node_hash_map`. +// If your type is not yet supported by the `absl::Hash` framework, see +// absl/hash/hash.h for information on extending Abseil hashing to user-defined +// types. +// +// Using `absl::node_hash_map` at interface boundaries in dynamically loaded +// libraries (e.g. .dll, .so) is unsupported due to way `absl::Hash` values may +// be randomized across dynamically loaded libraries. +// +// Example: +// +// // Create a node hash map of three strings (that map to strings) +// absl::node_hash_map ducks = +// {{"a", "huey"}, {"b", "dewey"}, {"c", "louie"}}; +// +// // Insert a new element into the node hash map +// ducks.insert({"d", "donald"}}; +// +// // Force a rehash of the node hash map +// ducks.rehash(0); +// +// // Find the element with the key "b" +// std::string search_key = "b"; +// auto result = ducks.find(search_key); +// if (result != ducks.end()) { +// std::cout << "Result: " << result->second << std::endl; +// } +template , + class Eq = absl::container_internal::hash_default_eq, + class Alloc = std::allocator>> +class node_hash_map + : public absl::container_internal::raw_hash_map< + absl::container_internal::NodeHashMapPolicy, Hash, Eq, + Alloc> { + using Base = typename node_hash_map::raw_hash_map; + + public: + // Constructors and Assignment Operators + // + // A node_hash_map supports the same overload set as `std::unordered_map` + // for construction and assignment: + // + // * Default constructor + // + // // No allocation for the table's elements is made. + // absl::node_hash_map map1; + // + // * Initializer List constructor + // + // absl::node_hash_map map2 = + // {{1, "huey"}, {2, "dewey"}, {3, "louie"},}; + // + // * Copy constructor + // + // absl::node_hash_map map3(map2); + // + // * Copy assignment operator + // + // // Hash functor and Comparator are copied as well + // absl::node_hash_map map4; + // map4 = map3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::node_hash_map map5(std::move(map4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::node_hash_map map6; + // map6 = std::move(map5); + // + // * Range constructor + // + // std::vector> v = {{1, "a"}, {2, "b"}}; + // absl::node_hash_map map7(v.begin(), v.end()); + node_hash_map() {} + using Base::Base; + + // node_hash_map::begin() + // + // Returns an iterator to the beginning of the `node_hash_map`. + using Base::begin; + + // node_hash_map::cbegin() + // + // Returns a const iterator to the beginning of the `node_hash_map`. + using Base::cbegin; + + // node_hash_map::cend() + // + // Returns a const iterator to the end of the `node_hash_map`. + using Base::cend; + + // node_hash_map::end() + // + // Returns an iterator to the end of the `node_hash_map`. + using Base::end; + + // node_hash_map::capacity() + // + // Returns the number of element slots (assigned, deleted, and empty) + // available within the `node_hash_map`. + // + // NOTE: this member function is particular to `absl::node_hash_map` and is + // not provided in the `std::unordered_map` API. + using Base::capacity; + + // node_hash_map::empty() + // + // Returns whether or not the `node_hash_map` is empty. + using Base::empty; + + // node_hash_map::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `node_hash_map` under current memory constraints. This value can be thought + // of as the largest value of `std::distance(begin(), end())` for a + // `node_hash_map`. + using Base::max_size; + + // node_hash_map::size() + // + // Returns the number of elements currently within the `node_hash_map`. + using Base::size; + + // node_hash_map::clear() + // + // Removes all elements from the `node_hash_map`. Invalidates any references, + // pointers, or iterators referring to contained elements. + // + // NOTE: this operation may shrink the underlying buffer. To avoid shrinking + // the underlying buffer call `erase(begin(), end())`. + using Base::clear; + + // node_hash_map::erase() + // + // Erases elements within the `node_hash_map`. Erasing does not trigger a + // rehash. Overloads are listed below. + // + // void erase(const_iterator pos): + // + // Erases the element at `position` of the `node_hash_map`, returning + // `void`. + // + // NOTE: this return behavior is different than that of STL containers in + // general and `std::unordered_map` in particular. + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning an + // iterator pointing to `last`. + // + // size_type erase(const key_type& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // node_hash_map::insert() + // + // Inserts an element of the specified value into the `node_hash_map`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If rehashing occurs + // due to the insertion, all iterators are invalidated. Overloads are listed + // below. + // + // std::pair insert(const init_type& value): + // + // Inserts a value into the `node_hash_map`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a `bool` denoting whether the insertion took place. + // + // std::pair insert(T&& value): + // std::pair insert(init_type&& value): + // + // Inserts a moveable value into the `node_hash_map`. Returns a `std::pair` + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a `bool` denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const init_type& value): + // iterator insert(const_iterator hint, T&& value): + // iterator insert(const_iterator hint, init_type&& value); + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently, for `node_hash_map` we guarantee the + // first match is inserted. + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently within the initializer list, for + // `node_hash_map` we guarantee the first match is inserted. + using Base::insert; + + // node_hash_map::insert_or_assign() + // + // Inserts an element of the specified value into the `node_hash_map` provided + // that a value with the given key does not already exist, or replaces it with + // the element value if a key for that value already exists, returning an + // iterator pointing to the newly inserted element. If rehashing occurs due to + // the insertion, all iterators are invalidated. Overloads are listed + // below. + // + // std::pair insert_or_assign(const init_type& k, T&& obj): + // std::pair insert_or_assign(init_type&& k, T&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `node_hash_map`. + // + // iterator insert_or_assign(const_iterator hint, + // const init_type& k, T&& obj): + // iterator insert_or_assign(const_iterator hint, init_type&& k, T&& obj): + // + // Inserts/Assigns (or moves) the element of the specified key into the + // `node_hash_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + using Base::insert_or_assign; + + // node_hash_map::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `node_hash_map`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace; + + // node_hash_map::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `node_hash_map`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. Prefer `try_emplace()` unless your key is not + // copyable or moveable. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace_hint; + + // node_hash_map::try_emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `node_hash_map`, provided that no element with the given key + // already exists. Unlike `emplace()`, if an element with the given key + // already exists, we guarantee that no element is constructed. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + // Overloads are listed below. + // + // std::pair try_emplace(const key_type& k, Args&&... args): + // std::pair try_emplace(key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `node_hash_map`. + // + // iterator try_emplace(const_iterator hint, + // const key_type& k, Args&&... args): + // iterator try_emplace(const_iterator hint, key_type&& k, Args&&... args): + // + // Inserts (via copy or move) the element of the specified key into the + // `node_hash_map` using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. + // + // All `try_emplace()` overloads make the same guarantees regarding rvalue + // arguments as `std::unordered_map::try_emplace()`, namely that these + // functions will not move from rvalue arguments if insertions do not happen. + using Base::try_emplace; + + // node_hash_map::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the key,value pair of the element at the indicated position and + // returns a node handle owning that extracted data. + // + // node_type extract(const key_type& x): + // + // Extracts the key,value pair of the element with a key matching the passed + // key value and returns a node handle owning that extracted data. If the + // `node_hash_map` does not contain an element with a matching key, this + // function returns an empty node handle. + // + // NOTE: when compiled in an earlier version of C++ than C++17, + // `node_type::key()` returns a const reference to the key instead of a + // mutable reference. We cannot safely return a mutable reference without + // std::launder (which is not available before C++17). + using Base::extract; + + // node_hash_map::merge() + // + // Extracts elements from a given `source` node hash map into this + // `node_hash_map`. If the destination `node_hash_map` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // node_hash_map::swap(node_hash_map& other) + // + // Exchanges the contents of this `node_hash_map` with those of the `other` + // node hash map, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `node_hash_map` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + // + // `swap()` requires that the node hash map's hashing and key equivalence + // functions be Swappable, and are exchaged using unqualified calls to + // non-member `swap()`. If the map's allocator has + // `std::allocator_traits::propagate_on_container_swap::value` + // set to `true`, the allocators are also exchanged using an unqualified call + // to non-member `swap()`; otherwise, the allocators are not swapped. + using Base::swap; + + // node_hash_map::rehash(count) + // + // Rehashes the `node_hash_map`, setting the number of slots to be at least + // the passed value. If the new number of slots increases the load factor more + // than the current maximum load factor + // (`count` < `size()` / `max_load_factor()`), then the new number of slots + // will be at least `size()` / `max_load_factor()`. + // + // To force a rehash, pass rehash(0). + using Base::rehash; + + // node_hash_map::reserve(count) + // + // Sets the number of slots in the `node_hash_map` to the number needed to + // accommodate at least `count` total elements without exceeding the current + // maximum load factor, and may rehash the container if needed. + using Base::reserve; + + // node_hash_map::at() + // + // Returns a reference to the mapped value of the element with key equivalent + // to the passed key. + using Base::at; + + // node_hash_map::contains() + // + // Determines whether an element with a key comparing equal to the given `key` + // exists within the `node_hash_map`, returning `true` if so or `false` + // otherwise. + using Base::contains; + + // node_hash_map::count(const Key& key) const + // + // Returns the number of elements with a key comparing equal to the given + // `key` within the `node_hash_map`. note that this function will return + // either `1` or `0` since duplicate keys are not allowed within a + // `node_hash_map`. + using Base::count; + + // node_hash_map::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `node_hash_map`. + using Base::equal_range; + + // node_hash_map::find() + // + // Finds an element with the passed `key` within the `node_hash_map`. + using Base::find; + + // node_hash_map::operator[]() + // + // Returns a reference to the value mapped to the passed key within the + // `node_hash_map`, performing an `insert()` if the key does not already + // exist. If an insertion occurs and results in a rehashing of the container, + // all iterators are invalidated. Otherwise iterators are not affected and + // references are not invalidated. Overloads are listed below. + // + // T& operator[](const Key& key): + // + // Inserts an init_type object constructed in-place if the element with the + // given key does not exist. + // + // T& operator[](Key&& key): + // + // Inserts an init_type object constructed in-place provided that an element + // with the given key does not exist. + using Base::operator[]; + + // node_hash_map::bucket_count() + // + // Returns the number of "buckets" within the `node_hash_map`. + using Base::bucket_count; + + // node_hash_map::load_factor() + // + // Returns the current load factor of the `node_hash_map` (the average number + // of slots occupied with a value within the hash map). + using Base::load_factor; + + // node_hash_map::max_load_factor() + // + // Manages the maximum load factor of the `node_hash_map`. Overloads are + // listed below. + // + // float node_hash_map::max_load_factor() + // + // Returns the current maximum load factor of the `node_hash_map`. + // + // void node_hash_map::max_load_factor(float ml) + // + // Sets the maximum load factor of the `node_hash_map` to the passed value. + // + // NOTE: This overload is provided only for API compatibility with the STL; + // `node_hash_map` will ignore any set load factor and manage its rehashing + // internally as an implementation detail. + using Base::max_load_factor; + + // node_hash_map::get_allocator() + // + // Returns the allocator function associated with this `node_hash_map`. + using Base::get_allocator; + + // node_hash_map::hash_function() + // + // Returns the hashing function used to hash the keys within this + // `node_hash_map`. + using Base::hash_function; + + // node_hash_map::key_eq() + // + // Returns the function used for comparing keys equality. + using Base::key_eq; +}; + +// erase_if(node_hash_map<>, Pred) +// +// Erases all elements that satisfy the predicate `pred` from the container `c`. +// Returns the number of erased elements. +template +typename node_hash_map::size_type erase_if( + node_hash_map& c, Predicate pred) { + return container_internal::EraseIf(pred, &c); +} + +namespace container_internal { + +template +class NodeHashMapPolicy + : public absl::container_internal::node_slot_policy< + std::pair&, NodeHashMapPolicy> { + using value_type = std::pair; + + public: + using key_type = Key; + using mapped_type = Value; + using init_type = std::pair; + + template + static value_type* new_element(Allocator* alloc, Args&&... args) { + using PairAlloc = typename absl::allocator_traits< + Allocator>::template rebind_alloc; + PairAlloc pair_alloc(*alloc); + value_type* res = + absl::allocator_traits::allocate(pair_alloc, 1); + absl::allocator_traits::construct(pair_alloc, res, + std::forward(args)...); + return res; + } + + template + static void delete_element(Allocator* alloc, value_type* pair) { + using PairAlloc = typename absl::allocator_traits< + Allocator>::template rebind_alloc; + PairAlloc pair_alloc(*alloc); + absl::allocator_traits::destroy(pair_alloc, pair); + absl::allocator_traits::deallocate(pair_alloc, pair, 1); + } + + template + static decltype(absl::container_internal::DecomposePair( + std::declval(), std::declval()...)) + apply(F&& f, Args&&... args) { + return absl::container_internal::DecomposePair(std::forward(f), + std::forward(args)...); + } + + static size_t element_space_used(const value_type*) { + return sizeof(value_type); + } + + static Value& value(value_type* elem) { return elem->second; } + static const Value& value(const value_type* elem) { return elem->second; } +}; +} // namespace container_internal + +namespace container_algorithm_internal { + +// Specialization of trait in absl/algorithm/container.h +template +struct IsUnorderedContainer< + absl::node_hash_map> : std::true_type {}; + +} // namespace container_algorithm_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_NODE_HASH_MAP_H_ diff --git a/src/absl/container/node_hash_set.h b/src/absl/container/node_hash_set.h new file mode 100644 index 0000000..f2cc70c --- /dev/null +++ b/src/absl/container/node_hash_set.h @@ -0,0 +1,500 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: node_hash_set.h +// ----------------------------------------------------------------------------- +// +// An `absl::node_hash_set` is an unordered associative container designed to +// be a more efficient replacement for `std::unordered_set`. Like +// `unordered_set`, search, insertion, and deletion of set elements can be done +// as an `O(1)` operation. However, `node_hash_set` (and other unordered +// associative containers known as the collection of Abseil "Swiss tables") +// contain other optimizations that result in both memory and computation +// advantages. +// +// In most cases, your default choice for a hash table should be a map of type +// `flat_hash_map` or a set of type `flat_hash_set`. However, if you need +// pointer stability, a `node_hash_set` should be your preferred choice. As +// well, if you are migrating your code from using `std::unordered_set`, a +// `node_hash_set` should be an easy migration. Consider migrating to +// `node_hash_set` and perhaps converting to a more efficient `flat_hash_set` +// upon further review. + +#ifndef ABSL_CONTAINER_NODE_HASH_SET_H_ +#define ABSL_CONTAINER_NODE_HASH_SET_H_ + +#include + +#include "absl/algorithm/container.h" +#include "absl/base/macros.h" +#include "absl/container/internal/hash_function_defaults.h" // IWYU pragma: export +#include "absl/container/internal/node_slot_policy.h" +#include "absl/container/internal/raw_hash_set.h" // IWYU pragma: export +#include "absl/memory/memory.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace container_internal { +template +struct NodeHashSetPolicy; +} // namespace container_internal + +// ----------------------------------------------------------------------------- +// absl::node_hash_set +// ----------------------------------------------------------------------------- +// +// An `absl::node_hash_set` is an unordered associative container which +// has been optimized for both speed and memory footprint in most common use +// cases. Its interface is similar to that of `std::unordered_set` with the +// following notable differences: +// +// * Supports heterogeneous lookup, through `find()`, `operator[]()` and +// `insert()`, provided that the set is provided a compatible heterogeneous +// hashing function and equality operator. +// * Contains a `capacity()` member function indicating the number of element +// slots (open, deleted, and empty) within the hash set. +// * Returns `void` from the `erase(iterator)` overload. +// +// By default, `node_hash_set` uses the `absl::Hash` hashing framework. +// All fundamental and Abseil types that support the `absl::Hash` framework have +// a compatible equality operator for comparing insertions into `node_hash_set`. +// If your type is not yet supported by the `absl::Hash` framework, see +// absl/hash/hash.h for information on extending Abseil hashing to user-defined +// types. +// +// Using `absl::node_hash_set` at interface boundaries in dynamically loaded +// libraries (e.g. .dll, .so) is unsupported due to way `absl::Hash` values may +// be randomized across dynamically loaded libraries. +// +// Example: +// +// // Create a node hash set of three strings +// absl::node_hash_set ducks = +// {"huey", "dewey", "louie"}; +// +// // Insert a new element into the node hash set +// ducks.insert("donald"); +// +// // Force a rehash of the node hash set +// ducks.rehash(0); +// +// // See if "dewey" is present +// if (ducks.contains("dewey")) { +// std::cout << "We found dewey!" << std::endl; +// } +template , + class Eq = absl::container_internal::hash_default_eq, + class Alloc = std::allocator> +class node_hash_set + : public absl::container_internal::raw_hash_set< + absl::container_internal::NodeHashSetPolicy, Hash, Eq, Alloc> { + using Base = typename node_hash_set::raw_hash_set; + + public: + // Constructors and Assignment Operators + // + // A node_hash_set supports the same overload set as `std::unordered_set` + // for construction and assignment: + // + // * Default constructor + // + // // No allocation for the table's elements is made. + // absl::node_hash_set set1; + // + // * Initializer List constructor + // + // absl::node_hash_set set2 = + // {{"huey"}, {"dewey"}, {"louie"}}; + // + // * Copy constructor + // + // absl::node_hash_set set3(set2); + // + // * Copy assignment operator + // + // // Hash functor and Comparator are copied as well + // absl::node_hash_set set4; + // set4 = set3; + // + // * Move constructor + // + // // Move is guaranteed efficient + // absl::node_hash_set set5(std::move(set4)); + // + // * Move assignment operator + // + // // May be efficient if allocators are compatible + // absl::node_hash_set set6; + // set6 = std::move(set5); + // + // * Range constructor + // + // std::vector v = {"a", "b"}; + // absl::node_hash_set set7(v.begin(), v.end()); + node_hash_set() {} + using Base::Base; + + // node_hash_set::begin() + // + // Returns an iterator to the beginning of the `node_hash_set`. + using Base::begin; + + // node_hash_set::cbegin() + // + // Returns a const iterator to the beginning of the `node_hash_set`. + using Base::cbegin; + + // node_hash_set::cend() + // + // Returns a const iterator to the end of the `node_hash_set`. + using Base::cend; + + // node_hash_set::end() + // + // Returns an iterator to the end of the `node_hash_set`. + using Base::end; + + // node_hash_set::capacity() + // + // Returns the number of element slots (assigned, deleted, and empty) + // available within the `node_hash_set`. + // + // NOTE: this member function is particular to `absl::node_hash_set` and is + // not provided in the `std::unordered_set` API. + using Base::capacity; + + // node_hash_set::empty() + // + // Returns whether or not the `node_hash_set` is empty. + using Base::empty; + + // node_hash_set::max_size() + // + // Returns the largest theoretical possible number of elements within a + // `node_hash_set` under current memory constraints. This value can be thought + // of the largest value of `std::distance(begin(), end())` for a + // `node_hash_set`. + using Base::max_size; + + // node_hash_set::size() + // + // Returns the number of elements currently within the `node_hash_set`. + using Base::size; + + // node_hash_set::clear() + // + // Removes all elements from the `node_hash_set`. Invalidates any references, + // pointers, or iterators referring to contained elements. + // + // NOTE: this operation may shrink the underlying buffer. To avoid shrinking + // the underlying buffer call `erase(begin(), end())`. + using Base::clear; + + // node_hash_set::erase() + // + // Erases elements within the `node_hash_set`. Erasing does not trigger a + // rehash. Overloads are listed below. + // + // void erase(const_iterator pos): + // + // Erases the element at `position` of the `node_hash_set`, returning + // `void`. + // + // NOTE: this return behavior is different than that of STL containers in + // general and `std::unordered_set` in particular. + // + // iterator erase(const_iterator first, const_iterator last): + // + // Erases the elements in the open interval [`first`, `last`), returning an + // iterator pointing to `last`. + // + // size_type erase(const key_type& key): + // + // Erases the element with the matching key, if it exists, returning the + // number of elements erased (0 or 1). + using Base::erase; + + // node_hash_set::insert() + // + // Inserts an element of the specified value into the `node_hash_set`, + // returning an iterator pointing to the newly inserted element, provided that + // an element with the given key does not already exist. If rehashing occurs + // due to the insertion, all iterators are invalidated. Overloads are listed + // below. + // + // std::pair insert(const T& value): + // + // Inserts a value into the `node_hash_set`. Returns a pair consisting of an + // iterator to the inserted element (or to the element that prevented the + // insertion) and a bool denoting whether the insertion took place. + // + // std::pair insert(T&& value): + // + // Inserts a moveable value into the `node_hash_set`. Returns a pair + // consisting of an iterator to the inserted element (or to the element that + // prevented the insertion) and a bool denoting whether the insertion took + // place. + // + // iterator insert(const_iterator hint, const T& value): + // iterator insert(const_iterator hint, T&& value): + // + // Inserts a value, using the position of `hint` as a non-binding suggestion + // for where to begin the insertion search. Returns an iterator to the + // inserted element, or to the existing element that prevented the + // insertion. + // + // void insert(InputIterator first, InputIterator last): + // + // Inserts a range of values [`first`, `last`). + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently, for `node_hash_set` we guarantee the + // first match is inserted. + // + // void insert(std::initializer_list ilist): + // + // Inserts the elements within the initializer list `ilist`. + // + // NOTE: Although the STL does not specify which element may be inserted if + // multiple keys compare equivalently within the initializer list, for + // `node_hash_set` we guarantee the first match is inserted. + using Base::insert; + + // node_hash_set::emplace() + // + // Inserts an element of the specified value by constructing it in-place + // within the `node_hash_set`, provided that no element with the given key + // already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace; + + // node_hash_set::emplace_hint() + // + // Inserts an element of the specified value by constructing it in-place + // within the `node_hash_set`, using the position of `hint` as a non-binding + // suggestion for where to begin the insertion search, and only inserts + // provided that no element with the given key already exists. + // + // The element may be constructed even if there already is an element with the + // key in the container, in which case the newly constructed element will be + // destroyed immediately. + // + // If rehashing occurs due to the insertion, all iterators are invalidated. + using Base::emplace_hint; + + // node_hash_set::extract() + // + // Extracts the indicated element, erasing it in the process, and returns it + // as a C++17-compatible node handle. Overloads are listed below. + // + // node_type extract(const_iterator position): + // + // Extracts the element at the indicated position and returns a node handle + // owning that extracted data. + // + // node_type extract(const key_type& x): + // + // Extracts the element with the key matching the passed key value and + // returns a node handle owning that extracted data. If the `node_hash_set` + // does not contain an element with a matching key, this function returns an + // empty node handle. + using Base::extract; + + // node_hash_set::merge() + // + // Extracts elements from a given `source` node hash set into this + // `node_hash_set`. If the destination `node_hash_set` already contains an + // element with an equivalent key, that element is not extracted. + using Base::merge; + + // node_hash_set::swap(node_hash_set& other) + // + // Exchanges the contents of this `node_hash_set` with those of the `other` + // node hash set, avoiding invocation of any move, copy, or swap operations on + // individual elements. + // + // All iterators and references on the `node_hash_set` remain valid, excepting + // for the past-the-end iterator, which is invalidated. + // + // `swap()` requires that the node hash set's hashing and key equivalence + // functions be Swappable, and are exchaged using unqualified calls to + // non-member `swap()`. If the set's allocator has + // `std::allocator_traits::propagate_on_container_swap::value` + // set to `true`, the allocators are also exchanged using an unqualified call + // to non-member `swap()`; otherwise, the allocators are not swapped. + using Base::swap; + + // node_hash_set::rehash(count) + // + // Rehashes the `node_hash_set`, setting the number of slots to be at least + // the passed value. If the new number of slots increases the load factor more + // than the current maximum load factor + // (`count` < `size()` / `max_load_factor()`), then the new number of slots + // will be at least `size()` / `max_load_factor()`. + // + // To force a rehash, pass rehash(0). + // + // NOTE: unlike behavior in `std::unordered_set`, references are also + // invalidated upon a `rehash()`. + using Base::rehash; + + // node_hash_set::reserve(count) + // + // Sets the number of slots in the `node_hash_set` to the number needed to + // accommodate at least `count` total elements without exceeding the current + // maximum load factor, and may rehash the container if needed. + using Base::reserve; + + // node_hash_set::contains() + // + // Determines whether an element comparing equal to the given `key` exists + // within the `node_hash_set`, returning `true` if so or `false` otherwise. + using Base::contains; + + // node_hash_set::count(const Key& key) const + // + // Returns the number of elements comparing equal to the given `key` within + // the `node_hash_set`. note that this function will return either `1` or `0` + // since duplicate elements are not allowed within a `node_hash_set`. + using Base::count; + + // node_hash_set::equal_range() + // + // Returns a closed range [first, last], defined by a `std::pair` of two + // iterators, containing all elements with the passed key in the + // `node_hash_set`. + using Base::equal_range; + + // node_hash_set::find() + // + // Finds an element with the passed `key` within the `node_hash_set`. + using Base::find; + + // node_hash_set::bucket_count() + // + // Returns the number of "buckets" within the `node_hash_set`. Note that + // because a node hash set contains all elements within its internal storage, + // this value simply equals the current capacity of the `node_hash_set`. + using Base::bucket_count; + + // node_hash_set::load_factor() + // + // Returns the current load factor of the `node_hash_set` (the average number + // of slots occupied with a value within the hash set). + using Base::load_factor; + + // node_hash_set::max_load_factor() + // + // Manages the maximum load factor of the `node_hash_set`. Overloads are + // listed below. + // + // float node_hash_set::max_load_factor() + // + // Returns the current maximum load factor of the `node_hash_set`. + // + // void node_hash_set::max_load_factor(float ml) + // + // Sets the maximum load factor of the `node_hash_set` to the passed value. + // + // NOTE: This overload is provided only for API compatibility with the STL; + // `node_hash_set` will ignore any set load factor and manage its rehashing + // internally as an implementation detail. + using Base::max_load_factor; + + // node_hash_set::get_allocator() + // + // Returns the allocator function associated with this `node_hash_set`. + using Base::get_allocator; + + // node_hash_set::hash_function() + // + // Returns the hashing function used to hash the keys within this + // `node_hash_set`. + using Base::hash_function; + + // node_hash_set::key_eq() + // + // Returns the function used for comparing keys equality. + using Base::key_eq; +}; + +// erase_if(node_hash_set<>, Pred) +// +// Erases all elements that satisfy the predicate `pred` from the container `c`. +// Returns the number of erased elements. +template +typename node_hash_set::size_type erase_if( + node_hash_set& c, Predicate pred) { + return container_internal::EraseIf(pred, &c); +} + +namespace container_internal { + +template +struct NodeHashSetPolicy + : absl::container_internal::node_slot_policy> { + using key_type = T; + using init_type = T; + using constant_iterators = std::true_type; + + template + static T* new_element(Allocator* alloc, Args&&... args) { + using ValueAlloc = + typename absl::allocator_traits::template rebind_alloc; + ValueAlloc value_alloc(*alloc); + T* res = absl::allocator_traits::allocate(value_alloc, 1); + absl::allocator_traits::construct(value_alloc, res, + std::forward(args)...); + return res; + } + + template + static void delete_element(Allocator* alloc, T* elem) { + using ValueAlloc = + typename absl::allocator_traits::template rebind_alloc; + ValueAlloc value_alloc(*alloc); + absl::allocator_traits::destroy(value_alloc, elem); + absl::allocator_traits::deallocate(value_alloc, elem, 1); + } + + template + static decltype(absl::container_internal::DecomposeValue( + std::declval(), std::declval()...)) + apply(F&& f, Args&&... args) { + return absl::container_internal::DecomposeValue( + std::forward(f), std::forward(args)...); + } + + static size_t element_space_used(const T*) { return sizeof(T); } +}; +} // namespace container_internal + +namespace container_algorithm_internal { + +// Specialization of trait in absl/algorithm/container.h +template +struct IsUnorderedContainer> + : std::true_type {}; + +} // namespace container_algorithm_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_CONTAINER_NODE_HASH_SET_H_ diff --git a/src/absl/debugging/failure_signal_handler.cc b/src/absl/debugging/failure_signal_handler.cc new file mode 100644 index 0000000..affade3 --- /dev/null +++ b/src/absl/debugging/failure_signal_handler.cc @@ -0,0 +1,386 @@ +// +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#include "absl/debugging/failure_signal_handler.h" + +#include "absl/base/config.h" + +#ifdef _WIN32 +#include +#else +#include +#include +#endif + +#ifdef __APPLE__ +#include +#endif + +#ifdef ABSL_HAVE_MMAP +#include +#endif + +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/sysinfo.h" +#include "absl/debugging/internal/examine_stack.h" +#include "absl/debugging/stacktrace.h" + +#ifndef _WIN32 +#define ABSL_HAVE_SIGACTION +// Apple WatchOS and TVOS don't allow sigaltstack +#if !(defined(TARGET_OS_WATCH) && TARGET_OS_WATCH) && \ + !(defined(TARGET_OS_TV) && TARGET_OS_TV) && !defined(__QNX__) +#define ABSL_HAVE_SIGALTSTACK +#endif +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN + +ABSL_CONST_INIT static FailureSignalHandlerOptions fsh_options; + +// Resets the signal handler for signo to the default action for that +// signal, then raises the signal. +static void RaiseToDefaultHandler(int signo) { + signal(signo, SIG_DFL); + raise(signo); +} + +struct FailureSignalData { + const int signo; + const char* const as_string; +#ifdef ABSL_HAVE_SIGACTION + struct sigaction previous_action; + // StructSigaction is used to silence -Wmissing-field-initializers. + using StructSigaction = struct sigaction; + #define FSD_PREVIOUS_INIT FailureSignalData::StructSigaction() +#else + void (*previous_handler)(int); + #define FSD_PREVIOUS_INIT SIG_DFL +#endif +}; + +ABSL_CONST_INIT static FailureSignalData failure_signal_data[] = { + {SIGSEGV, "SIGSEGV", FSD_PREVIOUS_INIT}, + {SIGILL, "SIGILL", FSD_PREVIOUS_INIT}, + {SIGFPE, "SIGFPE", FSD_PREVIOUS_INIT}, + {SIGABRT, "SIGABRT", FSD_PREVIOUS_INIT}, + {SIGTERM, "SIGTERM", FSD_PREVIOUS_INIT}, +#ifndef _WIN32 + {SIGBUS, "SIGBUS", FSD_PREVIOUS_INIT}, + {SIGTRAP, "SIGTRAP", FSD_PREVIOUS_INIT}, +#endif +}; + +#undef FSD_PREVIOUS_INIT + +static void RaiseToPreviousHandler(int signo) { + // Search for the previous handler. + for (const auto& it : failure_signal_data) { + if (it.signo == signo) { +#ifdef ABSL_HAVE_SIGACTION + sigaction(signo, &it.previous_action, nullptr); +#else + signal(signo, it.previous_handler); +#endif + raise(signo); + return; + } + } + + // Not found, use the default handler. + RaiseToDefaultHandler(signo); +} + +namespace debugging_internal { + +const char* FailureSignalToString(int signo) { + for (const auto& it : failure_signal_data) { + if (it.signo == signo) { + return it.as_string; + } + } + return ""; +} + +} // namespace debugging_internal + +#ifdef ABSL_HAVE_SIGALTSTACK + +static bool SetupAlternateStackOnce() { +#if defined(__wasm__) || defined (__asjms__) + const size_t page_mask = getpagesize() - 1; +#else + const size_t page_mask = sysconf(_SC_PAGESIZE) - 1; +#endif + size_t stack_size = + (std::max(SIGSTKSZ, 65536) + page_mask) & ~page_mask; +#if defined(ABSL_HAVE_ADDRESS_SANITIZER) || \ + defined(ABSL_HAVE_MEMORY_SANITIZER) || defined(ABSL_HAVE_THREAD_SANITIZER) + // Account for sanitizer instrumentation requiring additional stack space. + stack_size *= 5; +#endif + + stack_t sigstk; + memset(&sigstk, 0, sizeof(sigstk)); + sigstk.ss_size = stack_size; + +#ifdef ABSL_HAVE_MMAP +#ifndef MAP_STACK +#define MAP_STACK 0 +#endif +#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS) +#define MAP_ANONYMOUS MAP_ANON +#endif + sigstk.ss_sp = mmap(nullptr, sigstk.ss_size, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS | MAP_STACK, -1, 0); + if (sigstk.ss_sp == MAP_FAILED) { + ABSL_RAW_LOG(FATAL, "mmap() for alternate signal stack failed"); + } +#else + sigstk.ss_sp = malloc(sigstk.ss_size); + if (sigstk.ss_sp == nullptr) { + ABSL_RAW_LOG(FATAL, "malloc() for alternate signal stack failed"); + } +#endif + + if (sigaltstack(&sigstk, nullptr) != 0) { + ABSL_RAW_LOG(FATAL, "sigaltstack() failed with errno=%d", errno); + } + return true; +} + +#endif + +#ifdef ABSL_HAVE_SIGACTION + +// Sets up an alternate stack for signal handlers once. +// Returns the appropriate flag for sig_action.sa_flags +// if the system supports using an alternate stack. +static int MaybeSetupAlternateStack() { +#ifdef ABSL_HAVE_SIGALTSTACK + ABSL_ATTRIBUTE_UNUSED static const bool kOnce = SetupAlternateStackOnce(); + return SA_ONSTACK; +#else + return 0; +#endif +} + +static void InstallOneFailureHandler(FailureSignalData* data, + void (*handler)(int, siginfo_t*, void*)) { + struct sigaction act; + memset(&act, 0, sizeof(act)); + sigemptyset(&act.sa_mask); + act.sa_flags |= SA_SIGINFO; + // SA_NODEFER is required to handle SIGABRT from + // ImmediateAbortSignalHandler(). + act.sa_flags |= SA_NODEFER; + if (fsh_options.use_alternate_stack) { + act.sa_flags |= MaybeSetupAlternateStack(); + } + act.sa_sigaction = handler; + ABSL_RAW_CHECK(sigaction(data->signo, &act, &data->previous_action) == 0, + "sigaction() failed"); +} + +#else + +static void InstallOneFailureHandler(FailureSignalData* data, + void (*handler)(int)) { + data->previous_handler = signal(data->signo, handler); + ABSL_RAW_CHECK(data->previous_handler != SIG_ERR, "signal() failed"); +} + +#endif + +static void WriteToStderr(const char* data) { + absl::raw_logging_internal::AsyncSignalSafeWriteToStderr(data, strlen(data)); +} + +static void WriteSignalMessage(int signo, int cpu, + void (*writerfn)(const char*)) { + char buf[96]; + char on_cpu[32] = {0}; + if (cpu != -1) { + snprintf(on_cpu, sizeof(on_cpu), " on cpu %d", cpu); + } + const char* const signal_string = + debugging_internal::FailureSignalToString(signo); + if (signal_string != nullptr && signal_string[0] != '\0') { + snprintf(buf, sizeof(buf), "*** %s received at time=%ld%s ***\n", + signal_string, + static_cast(time(nullptr)), // NOLINT(runtime/int) + on_cpu); + } else { + snprintf(buf, sizeof(buf), "*** Signal %d received at time=%ld%s ***\n", + signo, static_cast(time(nullptr)), // NOLINT(runtime/int) + on_cpu); + } + writerfn(buf); +} + +// `void*` might not be big enough to store `void(*)(const char*)`. +struct WriterFnStruct { + void (*writerfn)(const char*); +}; + +// Many of the absl::debugging_internal::Dump* functions in +// examine_stack.h take a writer function pointer that has a void* arg +// for historical reasons. failure_signal_handler_writer only takes a +// data pointer. This function converts between these types. +static void WriterFnWrapper(const char* data, void* arg) { + static_cast(arg)->writerfn(data); +} + +// Convenient wrapper around DumpPCAndFrameSizesAndStackTrace() for signal +// handlers. "noinline" so that GetStackFrames() skips the top-most stack +// frame for this function. +ABSL_ATTRIBUTE_NOINLINE static void WriteStackTrace( + void* ucontext, bool symbolize_stacktrace, + void (*writerfn)(const char*, void*), void* writerfn_arg) { + constexpr int kNumStackFrames = 32; + void* stack[kNumStackFrames]; + int frame_sizes[kNumStackFrames]; + int min_dropped_frames; + int depth = absl::GetStackFramesWithContext( + stack, frame_sizes, kNumStackFrames, + 1, // Do not include this function in stack trace. + ucontext, &min_dropped_frames); + absl::debugging_internal::DumpPCAndFrameSizesAndStackTrace( + absl::debugging_internal::GetProgramCounter(ucontext), stack, frame_sizes, + depth, min_dropped_frames, symbolize_stacktrace, writerfn, writerfn_arg); +} + +// Called by AbslFailureSignalHandler() to write the failure info. It is +// called once with writerfn set to WriteToStderr() and then possibly +// with writerfn set to the user provided function. +static void WriteFailureInfo(int signo, void* ucontext, int cpu, + void (*writerfn)(const char*)) { + WriterFnStruct writerfn_struct{writerfn}; + WriteSignalMessage(signo, cpu, writerfn); + WriteStackTrace(ucontext, fsh_options.symbolize_stacktrace, WriterFnWrapper, + &writerfn_struct); +} + +// absl::SleepFor() can't be used here since AbslInternalSleepFor() +// may be overridden to do something that isn't async-signal-safe on +// some platforms. +static void PortableSleepForSeconds(int seconds) { +#ifdef _WIN32 + Sleep(seconds * 1000); +#else + struct timespec sleep_time; + sleep_time.tv_sec = seconds; + sleep_time.tv_nsec = 0; + while (nanosleep(&sleep_time, &sleep_time) != 0 && errno == EINTR) {} +#endif +} + +#ifdef ABSL_HAVE_ALARM +// AbslFailureSignalHandler() installs this as a signal handler for +// SIGALRM, then sets an alarm to be delivered to the program after a +// set amount of time. If AbslFailureSignalHandler() hangs for more than +// the alarm timeout, ImmediateAbortSignalHandler() will abort the +// program. +static void ImmediateAbortSignalHandler(int) { + RaiseToDefaultHandler(SIGABRT); +} +#endif + +// absl::base_internal::GetTID() returns pid_t on most platforms, but +// returns absl::base_internal::pid_t on Windows. +using GetTidType = decltype(absl::base_internal::GetTID()); +ABSL_CONST_INIT static std::atomic failed_tid(0); + +#ifndef ABSL_HAVE_SIGACTION +static void AbslFailureSignalHandler(int signo) { + void* ucontext = nullptr; +#else +static void AbslFailureSignalHandler(int signo, siginfo_t*, void* ucontext) { +#endif + + const GetTidType this_tid = absl::base_internal::GetTID(); + GetTidType previous_failed_tid = 0; + if (!failed_tid.compare_exchange_strong( + previous_failed_tid, static_cast(this_tid), + std::memory_order_acq_rel, std::memory_order_relaxed)) { + ABSL_RAW_LOG( + ERROR, + "Signal %d raised at PC=%p while already in AbslFailureSignalHandler()", + signo, absl::debugging_internal::GetProgramCounter(ucontext)); + if (this_tid != previous_failed_tid) { + // Another thread is already in AbslFailureSignalHandler(), so wait + // a bit for it to finish. If the other thread doesn't kill us, + // we do so after sleeping. + PortableSleepForSeconds(3); + RaiseToDefaultHandler(signo); + // The recursively raised signal may be blocked until we return. + return; + } + } + + // Increase the chance that the CPU we report was the same CPU on which the + // signal was received by doing this as early as possible, i.e. after + // verifying that this is not a recursive signal handler invocation. + int my_cpu = -1; +#ifdef ABSL_HAVE_SCHED_GETCPU + my_cpu = sched_getcpu(); +#endif + +#ifdef ABSL_HAVE_ALARM + // Set an alarm to abort the program in case this code hangs or deadlocks. + if (fsh_options.alarm_on_failure_secs > 0) { + alarm(0); // Cancel any existing alarms. + signal(SIGALRM, ImmediateAbortSignalHandler); + alarm(fsh_options.alarm_on_failure_secs); + } +#endif + + // First write to stderr. + WriteFailureInfo(signo, ucontext, my_cpu, WriteToStderr); + + // Riskier code (because it is less likely to be async-signal-safe) + // goes after this point. + if (fsh_options.writerfn != nullptr) { + WriteFailureInfo(signo, ucontext, my_cpu, fsh_options.writerfn); + fsh_options.writerfn(nullptr); + } + + if (fsh_options.call_previous_handler) { + RaiseToPreviousHandler(signo); + } else { + RaiseToDefaultHandler(signo); + } +} + +void InstallFailureSignalHandler(const FailureSignalHandlerOptions& options) { + fsh_options = options; + for (auto& it : failure_signal_data) { + InstallOneFailureHandler(&it, AbslFailureSignalHandler); + } +} + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/failure_signal_handler.h b/src/absl/debugging/failure_signal_handler.h new file mode 100644 index 0000000..500115c --- /dev/null +++ b/src/absl/debugging/failure_signal_handler.h @@ -0,0 +1,121 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: failure_signal_handler.h +// ----------------------------------------------------------------------------- +// +// This file configures the Abseil *failure signal handler* to capture and dump +// useful debugging information (such as a stacktrace) upon program failure. +// +// To use the failure signal handler, call `absl::InstallFailureSignalHandler()` +// very early in your program, usually in the first few lines of main(): +// +// int main(int argc, char** argv) { +// // Initialize the symbolizer to get a human-readable stack trace +// absl::InitializeSymbolizer(argv[0]); +// +// absl::FailureSignalHandlerOptions options; +// absl::InstallFailureSignalHandler(options); +// DoSomethingInteresting(); +// return 0; +// } +// +// Any program that raises a fatal signal (such as `SIGSEGV`, `SIGILL`, +// `SIGFPE`, `SIGABRT`, `SIGTERM`, `SIGBUG`, and `SIGTRAP`) will call the +// installed failure signal handler and provide debugging information to stderr. +// +// Note that you should *not* install the Abseil failure signal handler more +// than once. You may, of course, have another (non-Abseil) failure signal +// handler installed (which would be triggered if Abseil's failure signal +// handler sets `call_previous_handler` to `true`). + +#ifndef ABSL_DEBUGGING_FAILURE_SIGNAL_HANDLER_H_ +#define ABSL_DEBUGGING_FAILURE_SIGNAL_HANDLER_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// FailureSignalHandlerOptions +// +// Struct for holding `absl::InstallFailureSignalHandler()` configuration +// options. +struct FailureSignalHandlerOptions { + // If true, try to symbolize the stacktrace emitted on failure, provided that + // you have initialized a symbolizer for that purpose. (See symbolize.h for + // more information.) + bool symbolize_stacktrace = true; + + // If true, try to run signal handlers on an alternate stack (if supported on + // the given platform). An alternate stack is useful for program crashes due + // to a stack overflow; by running on a alternate stack, the signal handler + // may run even when normal stack space has been exausted. The downside of + // using an alternate stack is that extra memory for the alternate stack needs + // to be pre-allocated. + bool use_alternate_stack = true; + + // If positive, indicates the number of seconds after which the failure signal + // handler is invoked to abort the program. Setting such an alarm is useful in + // cases where the failure signal handler itself may become hung or + // deadlocked. + int alarm_on_failure_secs = 3; + + // If true, call the previously registered signal handler for the signal that + // was received (if one was registered) after the existing signal handler + // runs. This mechanism can be used to chain signal handlers together. + // + // If false, the signal is raised to the default handler for that signal + // (which normally terminates the program). + // + // IMPORTANT: If true, the chained fatal signal handlers must not try to + // recover from the fatal signal. Instead, they should terminate the program + // via some mechanism, like raising the default handler for the signal, or by + // calling `_exit()`. Note that the failure signal handler may put parts of + // the Abseil library into a state from which they cannot recover. + bool call_previous_handler = false; + + // If non-null, indicates a pointer to a callback function that will be called + // upon failure, with a string argument containing failure data. This function + // may be used as a hook to write failure data to a secondary location, such + // as a log file. This function will also be called with null data, as a hint + // to flush any buffered data before the program may be terminated. Consider + // flushing any buffered data in all calls to this function. + // + // Since this function runs within a signal handler, it should be + // async-signal-safe if possible. + // See http://man7.org/linux/man-pages/man7/signal-safety.7.html + void (*writerfn)(const char*) = nullptr; +}; + +// InstallFailureSignalHandler() +// +// Installs a signal handler for the common failure signals `SIGSEGV`, `SIGILL`, +// `SIGFPE`, `SIGABRT`, `SIGTERM`, `SIGBUG`, and `SIGTRAP` (provided they exist +// on the given platform). The failure signal handler dumps program failure data +// useful for debugging in an unspecified format to stderr. This data may +// include the program counter, a stacktrace, and register information on some +// systems; do not rely on an exact format for the output, as it is subject to +// change. +void InstallFailureSignalHandler(const FailureSignalHandlerOptions& options); + +namespace debugging_internal { +const char* FailureSignalToString(int signo); +} // namespace debugging_internal + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_FAILURE_SIGNAL_HANDLER_H_ diff --git a/src/absl/debugging/internal/address_is_readable.cc b/src/absl/debugging/internal/address_is_readable.cc new file mode 100644 index 0000000..4be6256 --- /dev/null +++ b/src/absl/debugging/internal/address_is_readable.cc @@ -0,0 +1,96 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// base::AddressIsReadable() probes an address to see whether it is readable, +// without faulting. + +#include "absl/debugging/internal/address_is_readable.h" + +#if !defined(__linux__) || defined(__ANDROID__) + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// On platforms other than Linux, just return true. +bool AddressIsReadable(const void* /* addr */) { return true; } + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#else // __linux__ && !__ANDROID__ + +#include +#include +#include + +#include "absl/base/internal/errno_saver.h" +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// NOTE: be extra careful about adding any interposable function calls here +// (such as open(), read(), etc.). These symbols may be interposed and will get +// invoked in contexts they don't expect. +// +// NOTE: any new system calls here may also require sandbox reconfiguration. +// +bool AddressIsReadable(const void *addr) { + // Align address on 8-byte boundary. On aarch64, checking last + // byte before inaccessible page returned unexpected EFAULT. + const uintptr_t u_addr = reinterpret_cast(addr) & ~7; + addr = reinterpret_cast(u_addr); + + // rt_sigprocmask below will succeed for this input. + if (addr == nullptr) return false; + + absl::base_internal::ErrnoSaver errno_saver; + + // Here we probe with some syscall which + // - accepts an 8-byte region of user memory as input + // - tests for EFAULT before other validation + // - has no problematic side-effects + // + // rt_sigprocmask(2) works for this. It copies sizeof(kernel_sigset_t)==8 + // bytes from the address into the kernel memory before any validation. + // + // The call can never succeed, since the `how` parameter is not one of + // SIG_BLOCK, SIG_UNBLOCK, SIG_SETMASK. + // + // This strategy depends on Linux implementation details, + // so we rely on the test to alert us if it stops working. + // + // Some discarded past approaches: + // - msync() doesn't reject PROT_NONE regions + // - write() on /dev/null doesn't return EFAULT + // - write() on a pipe requires creating it and draining the writes + // - connect() works but is problematic for sandboxes and needs a valid + // file descriptor + // + // This can never succeed (invalid first argument to sigprocmask). + ABSL_RAW_CHECK(syscall(SYS_rt_sigprocmask, ~0, addr, nullptr, + /*sizeof(kernel_sigset_t)*/ 8) == -1, + "unexpected success"); + ABSL_RAW_CHECK(errno == EFAULT || errno == EINVAL, "unexpected errno"); + return errno != EFAULT; +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // __linux__ && !__ANDROID__ diff --git a/src/absl/debugging/internal/address_is_readable.h b/src/absl/debugging/internal/address_is_readable.h new file mode 100644 index 0000000..4bbaf4d --- /dev/null +++ b/src/absl/debugging/internal/address_is_readable.h @@ -0,0 +1,32 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_DEBUGGING_INTERNAL_ADDRESS_IS_READABLE_H_ +#define ABSL_DEBUGGING_INTERNAL_ADDRESS_IS_READABLE_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// Return whether the byte at *addr is readable, without faulting. +// Save and restores errno. +bool AddressIsReadable(const void *addr); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_ADDRESS_IS_READABLE_H_ diff --git a/src/absl/debugging/internal/demangle.cc b/src/absl/debugging/internal/demangle.cc new file mode 100644 index 0000000..93ae327 --- /dev/null +++ b/src/absl/debugging/internal/demangle.cc @@ -0,0 +1,1959 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// For reference check out: +// https://itanium-cxx-abi.github.io/cxx-abi/abi.html#mangling +// +// Note that we only have partial C++11 support yet. + +#include "absl/debugging/internal/demangle.h" + +#include +#include +#include + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +typedef struct { + const char *abbrev; + const char *real_name; + // Number of arguments in context, or 0 if disallowed. + int arity; +} AbbrevPair; + +// List of operators from Itanium C++ ABI. +static const AbbrevPair kOperatorList[] = { + // New has special syntax (not currently supported). + {"nw", "new", 0}, + {"na", "new[]", 0}, + + // Works except that the 'gs' prefix is not supported. + {"dl", "delete", 1}, + {"da", "delete[]", 1}, + + {"ps", "+", 1}, // "positive" + {"ng", "-", 1}, // "negative" + {"ad", "&", 1}, // "address-of" + {"de", "*", 1}, // "dereference" + {"co", "~", 1}, + + {"pl", "+", 2}, + {"mi", "-", 2}, + {"ml", "*", 2}, + {"dv", "/", 2}, + {"rm", "%", 2}, + {"an", "&", 2}, + {"or", "|", 2}, + {"eo", "^", 2}, + {"aS", "=", 2}, + {"pL", "+=", 2}, + {"mI", "-=", 2}, + {"mL", "*=", 2}, + {"dV", "/=", 2}, + {"rM", "%=", 2}, + {"aN", "&=", 2}, + {"oR", "|=", 2}, + {"eO", "^=", 2}, + {"ls", "<<", 2}, + {"rs", ">>", 2}, + {"lS", "<<=", 2}, + {"rS", ">>=", 2}, + {"eq", "==", 2}, + {"ne", "!=", 2}, + {"lt", "<", 2}, + {"gt", ">", 2}, + {"le", "<=", 2}, + {"ge", ">=", 2}, + {"nt", "!", 1}, + {"aa", "&&", 2}, + {"oo", "||", 2}, + {"pp", "++", 1}, + {"mm", "--", 1}, + {"cm", ",", 2}, + {"pm", "->*", 2}, + {"pt", "->", 0}, // Special syntax + {"cl", "()", 0}, // Special syntax + {"ix", "[]", 2}, + {"qu", "?", 3}, + {"st", "sizeof", 0}, // Special syntax + {"sz", "sizeof", 1}, // Not a real operator name, but used in expressions. + {nullptr, nullptr, 0}, +}; + +// List of builtin types from Itanium C++ ABI. +// +// Invariant: only one- or two-character type abbreviations here. +static const AbbrevPair kBuiltinTypeList[] = { + {"v", "void", 0}, + {"w", "wchar_t", 0}, + {"b", "bool", 0}, + {"c", "char", 0}, + {"a", "signed char", 0}, + {"h", "unsigned char", 0}, + {"s", "short", 0}, + {"t", "unsigned short", 0}, + {"i", "int", 0}, + {"j", "unsigned int", 0}, + {"l", "long", 0}, + {"m", "unsigned long", 0}, + {"x", "long long", 0}, + {"y", "unsigned long long", 0}, + {"n", "__int128", 0}, + {"o", "unsigned __int128", 0}, + {"f", "float", 0}, + {"d", "double", 0}, + {"e", "long double", 0}, + {"g", "__float128", 0}, + {"z", "ellipsis", 0}, + + {"De", "decimal128", 0}, // IEEE 754r decimal floating point (128 bits) + {"Dd", "decimal64", 0}, // IEEE 754r decimal floating point (64 bits) + {"Dc", "decltype(auto)", 0}, + {"Da", "auto", 0}, + {"Dn", "std::nullptr_t", 0}, // i.e., decltype(nullptr) + {"Df", "decimal32", 0}, // IEEE 754r decimal floating point (32 bits) + {"Di", "char32_t", 0}, + {"Du", "char8_t", 0}, + {"Ds", "char16_t", 0}, + {"Dh", "float16", 0}, // IEEE 754r half-precision float (16 bits) + {nullptr, nullptr, 0}, +}; + +// List of substitutions Itanium C++ ABI. +static const AbbrevPair kSubstitutionList[] = { + {"St", "", 0}, + {"Sa", "allocator", 0}, + {"Sb", "basic_string", 0}, + // std::basic_string,std::allocator > + {"Ss", "string", 0}, + // std::basic_istream > + {"Si", "istream", 0}, + // std::basic_ostream > + {"So", "ostream", 0}, + // std::basic_iostream > + {"Sd", "iostream", 0}, + {nullptr, nullptr, 0}, +}; + +// State needed for demangling. This struct is copied in almost every stack +// frame, so every byte counts. +typedef struct { + int mangled_idx; // Cursor of mangled name. + int out_cur_idx; // Cursor of output string. + int prev_name_idx; // For constructors/destructors. + signed int prev_name_length : 16; // For constructors/destructors. + signed int nest_level : 15; // For nested names. + unsigned int append : 1; // Append flag. + // Note: for some reason MSVC can't pack "bool append : 1" into the same int + // with the above two fields, so we use an int instead. Amusingly it can pack + // "signed bool" as expected, but relying on that to continue to be a legal + // type seems ill-advised (as it's illegal in at least clang). +} ParseState; + +static_assert(sizeof(ParseState) == 4 * sizeof(int), + "unexpected size of ParseState"); + +// One-off state for demangling that's not subject to backtracking -- either +// constant data, data that's intentionally immune to backtracking (steps), or +// data that would never be changed by backtracking anyway (recursion_depth). +// +// Only one copy of this exists for each call to Demangle, so the size of this +// struct is nearly inconsequential. +typedef struct { + const char *mangled_begin; // Beginning of input string. + char *out; // Beginning of output string. + int out_end_idx; // One past last allowed output character. + int recursion_depth; // For stack exhaustion prevention. + int steps; // Cap how much work we'll do, regardless of depth. + ParseState parse_state; // Backtrackable state copied for most frames. +} State; + +namespace { +// Prevent deep recursion / stack exhaustion. +// Also prevent unbounded handling of complex inputs. +class ComplexityGuard { + public: + explicit ComplexityGuard(State *state) : state_(state) { + ++state->recursion_depth; + ++state->steps; + } + ~ComplexityGuard() { --state_->recursion_depth; } + + // 256 levels of recursion seems like a reasonable upper limit on depth. + // 128 is not enough to demagle synthetic tests from demangle_unittest.txt: + // "_ZaaZZZZ..." and "_ZaaZcvZcvZ..." + static constexpr int kRecursionDepthLimit = 256; + + // We're trying to pick a charitable upper-limit on how many parse steps are + // necessary to handle something that a human could actually make use of. + // This is mostly in place as a bound on how much work we'll do if we are + // asked to demangle an mangled name from an untrusted source, so it should be + // much larger than the largest expected symbol, but much smaller than the + // amount of work we can do in, e.g., a second. + // + // Some real-world symbols from an arbitrary binary started failing between + // 2^12 and 2^13, so we multiply the latter by an extra factor of 16 to set + // the limit. + // + // Spending one second on 2^17 parse steps would require each step to take + // 7.6us, or ~30000 clock cycles, so it's safe to say this can be done in + // under a second. + static constexpr int kParseStepsLimit = 1 << 17; + + bool IsTooComplex() const { + return state_->recursion_depth > kRecursionDepthLimit || + state_->steps > kParseStepsLimit; + } + + private: + State *state_; +}; +} // namespace + +// We don't use strlen() in libc since it's not guaranteed to be async +// signal safe. +static size_t StrLen(const char *str) { + size_t len = 0; + while (*str != '\0') { + ++str; + ++len; + } + return len; +} + +// Returns true if "str" has at least "n" characters remaining. +static bool AtLeastNumCharsRemaining(const char *str, int n) { + for (int i = 0; i < n; ++i) { + if (str[i] == '\0') { + return false; + } + } + return true; +} + +// Returns true if "str" has "prefix" as a prefix. +static bool StrPrefix(const char *str, const char *prefix) { + size_t i = 0; + while (str[i] != '\0' && prefix[i] != '\0' && str[i] == prefix[i]) { + ++i; + } + return prefix[i] == '\0'; // Consumed everything in "prefix". +} + +static void InitState(State *state, const char *mangled, char *out, + int out_size) { + state->mangled_begin = mangled; + state->out = out; + state->out_end_idx = out_size; + state->recursion_depth = 0; + state->steps = 0; + + state->parse_state.mangled_idx = 0; + state->parse_state.out_cur_idx = 0; + state->parse_state.prev_name_idx = 0; + state->parse_state.prev_name_length = -1; + state->parse_state.nest_level = -1; + state->parse_state.append = true; +} + +static inline const char *RemainingInput(State *state) { + return &state->mangled_begin[state->parse_state.mangled_idx]; +} + +// Returns true and advances "mangled_idx" if we find "one_char_token" +// at "mangled_idx" position. It is assumed that "one_char_token" does +// not contain '\0'. +static bool ParseOneCharToken(State *state, const char one_char_token) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (RemainingInput(state)[0] == one_char_token) { + ++state->parse_state.mangled_idx; + return true; + } + return false; +} + +// Returns true and advances "mangled_cur" if we find "two_char_token" +// at "mangled_cur" position. It is assumed that "two_char_token" does +// not contain '\0'. +static bool ParseTwoCharToken(State *state, const char *two_char_token) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (RemainingInput(state)[0] == two_char_token[0] && + RemainingInput(state)[1] == two_char_token[1]) { + state->parse_state.mangled_idx += 2; + return true; + } + return false; +} + +// Returns true and advances "mangled_cur" if we find any character in +// "char_class" at "mangled_cur" position. +static bool ParseCharClass(State *state, const char *char_class) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (RemainingInput(state)[0] == '\0') { + return false; + } + const char *p = char_class; + for (; *p != '\0'; ++p) { + if (RemainingInput(state)[0] == *p) { + ++state->parse_state.mangled_idx; + return true; + } + } + return false; +} + +static bool ParseDigit(State *state, int *digit) { + char c = RemainingInput(state)[0]; + if (ParseCharClass(state, "0123456789")) { + if (digit != nullptr) { + *digit = c - '0'; + } + return true; + } + return false; +} + +// This function is used for handling an optional non-terminal. +static bool Optional(bool /*status*/) { return true; } + +// This function is used for handling + syntax. +typedef bool (*ParseFunc)(State *); +static bool OneOrMore(ParseFunc parse_func, State *state) { + if (parse_func(state)) { + while (parse_func(state)) { + } + return true; + } + return false; +} + +// This function is used for handling * syntax. The function +// always returns true and must be followed by a termination token or a +// terminating sequence not handled by parse_func (e.g. +// ParseOneCharToken(state, 'E')). +static bool ZeroOrMore(ParseFunc parse_func, State *state) { + while (parse_func(state)) { + } + return true; +} + +// Append "str" at "out_cur_idx". If there is an overflow, out_cur_idx is +// set to out_end_idx+1. The output string is ensured to +// always terminate with '\0' as long as there is no overflow. +static void Append(State *state, const char *const str, const int length) { + for (int i = 0; i < length; ++i) { + if (state->parse_state.out_cur_idx + 1 < + state->out_end_idx) { // +1 for '\0' + state->out[state->parse_state.out_cur_idx++] = str[i]; + } else { + // signal overflow + state->parse_state.out_cur_idx = state->out_end_idx + 1; + break; + } + } + if (state->parse_state.out_cur_idx < state->out_end_idx) { + state->out[state->parse_state.out_cur_idx] = + '\0'; // Terminate it with '\0' + } +} + +// We don't use equivalents in libc to avoid locale issues. +static bool IsLower(char c) { return c >= 'a' && c <= 'z'; } + +static bool IsAlpha(char c) { + return (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z'); +} + +static bool IsDigit(char c) { return c >= '0' && c <= '9'; } + +// Returns true if "str" is a function clone suffix. These suffixes are used +// by GCC 4.5.x and later versions (and our locally-modified version of GCC +// 4.4.x) to indicate functions which have been cloned during optimization. +// We treat any sequence (.+.+)+ as a function clone suffix. +// Additionally, '_' is allowed along with the alphanumeric sequence. +static bool IsFunctionCloneSuffix(const char *str) { + size_t i = 0; + while (str[i] != '\0') { + bool parsed = false; + // Consume a single [. | _]*[.]* sequence. + if (str[i] == '.' && (IsAlpha(str[i + 1]) || str[i + 1] == '_')) { + parsed = true; + i += 2; + while (IsAlpha(str[i]) || str[i] == '_') { + ++i; + } + } + if (str[i] == '.' && IsDigit(str[i + 1])) { + parsed = true; + i += 2; + while (IsDigit(str[i])) { + ++i; + } + } + if (!parsed) + return false; + } + return true; // Consumed everything in "str". +} + +static bool EndsWith(State *state, const char chr) { + return state->parse_state.out_cur_idx > 0 && + state->parse_state.out_cur_idx < state->out_end_idx && + chr == state->out[state->parse_state.out_cur_idx - 1]; +} + +// Append "str" with some tweaks, iff "append" state is true. +static void MaybeAppendWithLength(State *state, const char *const str, + const int length) { + if (state->parse_state.append && length > 0) { + // Append a space if the output buffer ends with '<' and "str" + // starts with '<' to avoid <<<. + if (str[0] == '<' && EndsWith(state, '<')) { + Append(state, " ", 1); + } + // Remember the last identifier name for ctors/dtors, + // but only if we haven't yet overflown the buffer. + if (state->parse_state.out_cur_idx < state->out_end_idx && + (IsAlpha(str[0]) || str[0] == '_')) { + state->parse_state.prev_name_idx = state->parse_state.out_cur_idx; + state->parse_state.prev_name_length = length; + } + Append(state, str, length); + } +} + +// Appends a positive decimal number to the output if appending is enabled. +static bool MaybeAppendDecimal(State *state, unsigned int val) { + // Max {32-64}-bit unsigned int is 20 digits. + constexpr size_t kMaxLength = 20; + char buf[kMaxLength]; + + // We can't use itoa or sprintf as neither is specified to be + // async-signal-safe. + if (state->parse_state.append) { + // We can't have a one-before-the-beginning pointer, so instead start with + // one-past-the-end and manipulate one character before the pointer. + char *p = &buf[kMaxLength]; + do { // val=0 is the only input that should write a leading zero digit. + *--p = (val % 10) + '0'; + val /= 10; + } while (p > buf && val != 0); + + // 'p' landed on the last character we set. How convenient. + Append(state, p, kMaxLength - (p - buf)); + } + + return true; +} + +// A convenient wrapper around MaybeAppendWithLength(). +// Returns true so that it can be placed in "if" conditions. +static bool MaybeAppend(State *state, const char *const str) { + if (state->parse_state.append) { + int length = StrLen(str); + MaybeAppendWithLength(state, str, length); + } + return true; +} + +// This function is used for handling nested names. +static bool EnterNestedName(State *state) { + state->parse_state.nest_level = 0; + return true; +} + +// This function is used for handling nested names. +static bool LeaveNestedName(State *state, int16_t prev_value) { + state->parse_state.nest_level = prev_value; + return true; +} + +// Disable the append mode not to print function parameters, etc. +static bool DisableAppend(State *state) { + state->parse_state.append = false; + return true; +} + +// Restore the append mode to the previous state. +static bool RestoreAppend(State *state, bool prev_value) { + state->parse_state.append = prev_value; + return true; +} + +// Increase the nest level for nested names. +static void MaybeIncreaseNestLevel(State *state) { + if (state->parse_state.nest_level > -1) { + ++state->parse_state.nest_level; + } +} + +// Appends :: for nested names if necessary. +static void MaybeAppendSeparator(State *state) { + if (state->parse_state.nest_level >= 1) { + MaybeAppend(state, "::"); + } +} + +// Cancel the last separator if necessary. +static void MaybeCancelLastSeparator(State *state) { + if (state->parse_state.nest_level >= 1 && state->parse_state.append && + state->parse_state.out_cur_idx >= 2) { + state->parse_state.out_cur_idx -= 2; + state->out[state->parse_state.out_cur_idx] = '\0'; + } +} + +// Returns true if the identifier of the given length pointed to by +// "mangled_cur" is anonymous namespace. +static bool IdentifierIsAnonymousNamespace(State *state, int length) { + // Returns true if "anon_prefix" is a proper prefix of "mangled_cur". + static const char anon_prefix[] = "_GLOBAL__N_"; + return (length > static_cast(sizeof(anon_prefix) - 1) && + StrPrefix(RemainingInput(state), anon_prefix)); +} + +// Forward declarations of our parsing functions. +static bool ParseMangledName(State *state); +static bool ParseEncoding(State *state); +static bool ParseName(State *state); +static bool ParseUnscopedName(State *state); +static bool ParseNestedName(State *state); +static bool ParsePrefix(State *state); +static bool ParseUnqualifiedName(State *state); +static bool ParseSourceName(State *state); +static bool ParseLocalSourceName(State *state); +static bool ParseUnnamedTypeName(State *state); +static bool ParseNumber(State *state, int *number_out); +static bool ParseFloatNumber(State *state); +static bool ParseSeqId(State *state); +static bool ParseIdentifier(State *state, int length); +static bool ParseOperatorName(State *state, int *arity); +static bool ParseSpecialName(State *state); +static bool ParseCallOffset(State *state); +static bool ParseNVOffset(State *state); +static bool ParseVOffset(State *state); +static bool ParseCtorDtorName(State *state); +static bool ParseDecltype(State *state); +static bool ParseType(State *state); +static bool ParseCVQualifiers(State *state); +static bool ParseBuiltinType(State *state); +static bool ParseFunctionType(State *state); +static bool ParseBareFunctionType(State *state); +static bool ParseClassEnumType(State *state); +static bool ParseArrayType(State *state); +static bool ParsePointerToMemberType(State *state); +static bool ParseTemplateParam(State *state); +static bool ParseTemplateTemplateParam(State *state); +static bool ParseTemplateArgs(State *state); +static bool ParseTemplateArg(State *state); +static bool ParseBaseUnresolvedName(State *state); +static bool ParseUnresolvedName(State *state); +static bool ParseExpression(State *state); +static bool ParseExprPrimary(State *state); +static bool ParseExprCastValue(State *state); +static bool ParseLocalName(State *state); +static bool ParseLocalNameSuffix(State *state); +static bool ParseDiscriminator(State *state); +static bool ParseSubstitution(State *state, bool accept_std); + +// Implementation note: the following code is a straightforward +// translation of the Itanium C++ ABI defined in BNF with a couple of +// exceptions. +// +// - Support GNU extensions not defined in the Itanium C++ ABI +// - and are combined to avoid infinite loop +// - Reorder patterns to shorten the code +// - Reorder patterns to give greedier functions precedence +// We'll mark "Less greedy than" for these cases in the code +// +// Each parsing function changes the parse state and returns true on +// success, or returns false and doesn't change the parse state (note: +// the parse-steps counter increases regardless of success or failure). +// To ensure that the parse state isn't changed in the latter case, we +// save the original state before we call multiple parsing functions +// consecutively with &&, and restore it if unsuccessful. See +// ParseEncoding() as an example of this convention. We follow the +// convention throughout the code. +// +// Originally we tried to do demangling without following the full ABI +// syntax but it turned out we needed to follow the full syntax to +// parse complicated cases like nested template arguments. Note that +// implementing a full-fledged demangler isn't trivial (libiberty's +// cp-demangle.c has +4300 lines). +// +// Note that (foo) in <(foo) ...> is a modifier to be ignored. +// +// Reference: +// - Itanium C++ ABI +// + +// ::= _Z +static bool ParseMangledName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + return ParseTwoCharToken(state, "_Z") && ParseEncoding(state); +} + +// ::= <(function) name> +// ::= <(data) name> +// ::= +static bool ParseEncoding(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + // Implementing the first two productions together as + // [] avoids exponential blowup of backtracking. + // + // Since Optional(...) can't fail, there's no need to copy the state for + // backtracking. + if (ParseName(state) && Optional(ParseBareFunctionType(state))) { + return true; + } + + if (ParseSpecialName(state)) { + return true; + } + return false; +} + +// ::= +// ::= +// ::= +// ::= +static bool ParseName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseNestedName(state) || ParseLocalName(state)) { + return true; + } + + // We reorganize the productions to avoid re-parsing unscoped names. + // - Inline productions: + // ::= + // ::= + // ::= + // - Merge the two productions that start with unscoped-name: + // ::= [] + + ParseState copy = state->parse_state; + // "std<...>" isn't a valid name. + if (ParseSubstitution(state, /*accept_std=*/false) && + ParseTemplateArgs(state)) { + return true; + } + state->parse_state = copy; + + // Note there's no need to restore state after this since only the first + // subparser can fail. + return ParseUnscopedName(state) && Optional(ParseTemplateArgs(state)); +} + +// ::= +// ::= St +static bool ParseUnscopedName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseUnqualifiedName(state)) { + return true; + } + + ParseState copy = state->parse_state; + if (ParseTwoCharToken(state, "St") && MaybeAppend(state, "std::") && + ParseUnqualifiedName(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= R // lvalue method reference qualifier +// ::= O // rvalue method reference qualifier +static inline bool ParseRefQualifier(State *state) { + return ParseCharClass(state, "OR"); +} + +// ::= N [] [] +// E +// ::= N [] [] +// E +static bool ParseNestedName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'N') && EnterNestedName(state) && + Optional(ParseCVQualifiers(state)) && + Optional(ParseRefQualifier(state)) && ParsePrefix(state) && + LeaveNestedName(state, copy.nest_level) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + return false; +} + +// This part is tricky. If we literally translate them to code, we'll +// end up infinite loop. Hence we merge them to avoid the case. +// +// ::= +// ::= +// ::= +// ::= +// ::= # empty +// ::= <(template) unqualified-name> +// ::= +// ::= +static bool ParsePrefix(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + bool has_something = false; + while (true) { + MaybeAppendSeparator(state); + if (ParseTemplateParam(state) || + ParseSubstitution(state, /*accept_std=*/true) || + ParseUnscopedName(state) || + (ParseOneCharToken(state, 'M') && ParseUnnamedTypeName(state))) { + has_something = true; + MaybeIncreaseNestLevel(state); + continue; + } + MaybeCancelLastSeparator(state); + if (has_something && ParseTemplateArgs(state)) { + return ParsePrefix(state); + } else { + break; + } + } + return true; +} + +// ::= +// ::= +// ::= +// ::= // GCC extension; see below. +// ::= +static bool ParseUnqualifiedName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + return (ParseOperatorName(state, nullptr) || ParseCtorDtorName(state) || + ParseSourceName(state) || ParseLocalSourceName(state) || + ParseUnnamedTypeName(state)); +} + +// ::= +static bool ParseSourceName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + int length = -1; + if (ParseNumber(state, &length) && ParseIdentifier(state, length)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= L [] +// +// References: +// https://gcc.gnu.org/bugzilla/show_bug.cgi?id=31775 +// https://gcc.gnu.org/viewcvs?view=rev&revision=124467 +static bool ParseLocalSourceName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'L') && ParseSourceName(state) && + Optional(ParseDiscriminator(state))) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= Ut [<(nonnegative) number>] _ +// ::= +// ::= Ul E [<(nonnegative) number>] _ +// ::= <(parameter) type>+ +static bool ParseUnnamedTypeName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + // Type's 1-based index n is encoded as { "", n == 1; itoa(n-2), otherwise }. + // Optionally parse the encoded value into 'which' and add 2 to get the index. + int which = -1; + + // Unnamed type local to function or class. + if (ParseTwoCharToken(state, "Ut") && Optional(ParseNumber(state, &which)) && + which <= std::numeric_limits::max() - 2 && // Don't overflow. + ParseOneCharToken(state, '_')) { + MaybeAppend(state, "{unnamed type#"); + MaybeAppendDecimal(state, 2 + which); + MaybeAppend(state, "}"); + return true; + } + state->parse_state = copy; + + // Closure type. + which = -1; + if (ParseTwoCharToken(state, "Ul") && DisableAppend(state) && + OneOrMore(ParseType, state) && RestoreAppend(state, copy.append) && + ParseOneCharToken(state, 'E') && Optional(ParseNumber(state, &which)) && + which <= std::numeric_limits::max() - 2 && // Don't overflow. + ParseOneCharToken(state, '_')) { + MaybeAppend(state, "{lambda()#"); + MaybeAppendDecimal(state, 2 + which); + MaybeAppend(state, "}"); + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= [n] +// If "number_out" is non-null, then *number_out is set to the value of the +// parsed number on success. +static bool ParseNumber(State *state, int *number_out) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + bool negative = false; + if (ParseOneCharToken(state, 'n')) { + negative = true; + } + const char *p = RemainingInput(state); + uint64_t number = 0; + for (; *p != '\0'; ++p) { + if (IsDigit(*p)) { + number = number * 10 + (*p - '0'); + } else { + break; + } + } + // Apply the sign with uint64_t arithmetic so overflows aren't UB. Gives + // "incorrect" results for out-of-range inputs, but negative values only + // appear for literals, which aren't printed. + if (negative) { + number = ~number + 1; + } + if (p != RemainingInput(state)) { // Conversion succeeded. + state->parse_state.mangled_idx += p - RemainingInput(state); + if (number_out != nullptr) { + // Note: possibly truncate "number". + *number_out = number; + } + return true; + } + return false; +} + +// Floating-point literals are encoded using a fixed-length lowercase +// hexadecimal string. +static bool ParseFloatNumber(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + const char *p = RemainingInput(state); + for (; *p != '\0'; ++p) { + if (!IsDigit(*p) && !(*p >= 'a' && *p <= 'f')) { + break; + } + } + if (p != RemainingInput(state)) { // Conversion succeeded. + state->parse_state.mangled_idx += p - RemainingInput(state); + return true; + } + return false; +} + +// The is a sequence number in base 36, +// using digits and upper case letters +static bool ParseSeqId(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + const char *p = RemainingInput(state); + for (; *p != '\0'; ++p) { + if (!IsDigit(*p) && !(*p >= 'A' && *p <= 'Z')) { + break; + } + } + if (p != RemainingInput(state)) { // Conversion succeeded. + state->parse_state.mangled_idx += p - RemainingInput(state); + return true; + } + return false; +} + +// ::= (of given length) +static bool ParseIdentifier(State *state, int length) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (length < 0 || !AtLeastNumCharsRemaining(RemainingInput(state), length)) { + return false; + } + if (IdentifierIsAnonymousNamespace(state, length)) { + MaybeAppend(state, "(anonymous namespace)"); + } else { + MaybeAppendWithLength(state, RemainingInput(state), length); + } + state->parse_state.mangled_idx += length; + return true; +} + +// ::= nw, and other two letters cases +// ::= cv # (cast) +// ::= v # vendor extended operator +static bool ParseOperatorName(State *state, int *arity) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (!AtLeastNumCharsRemaining(RemainingInput(state), 2)) { + return false; + } + // First check with "cv" (cast) case. + ParseState copy = state->parse_state; + if (ParseTwoCharToken(state, "cv") && MaybeAppend(state, "operator ") && + EnterNestedName(state) && ParseType(state) && + LeaveNestedName(state, copy.nest_level)) { + if (arity != nullptr) { + *arity = 1; + } + return true; + } + state->parse_state = copy; + + // Then vendor extended operators. + if (ParseOneCharToken(state, 'v') && ParseDigit(state, arity) && + ParseSourceName(state)) { + return true; + } + state->parse_state = copy; + + // Other operator names should start with a lower alphabet followed + // by a lower/upper alphabet. + if (!(IsLower(RemainingInput(state)[0]) && + IsAlpha(RemainingInput(state)[1]))) { + return false; + } + // We may want to perform a binary search if we really need speed. + const AbbrevPair *p; + for (p = kOperatorList; p->abbrev != nullptr; ++p) { + if (RemainingInput(state)[0] == p->abbrev[0] && + RemainingInput(state)[1] == p->abbrev[1]) { + if (arity != nullptr) { + *arity = p->arity; + } + MaybeAppend(state, "operator"); + if (IsLower(*p->real_name)) { // new, delete, etc. + MaybeAppend(state, " "); + } + MaybeAppend(state, p->real_name); + state->parse_state.mangled_idx += 2; + return true; + } + } + return false; +} + +// ::= TV +// ::= TT +// ::= TI +// ::= TS +// ::= TH # thread-local +// ::= Tc <(base) encoding> +// ::= GV <(object) name> +// ::= T <(base) encoding> +// G++ extensions: +// ::= TC <(offset) number> _ <(base) type> +// ::= TF +// ::= TJ +// ::= GR +// ::= GA +// ::= Th <(base) encoding> +// ::= Tv <(base) encoding> +// +// Note: we don't care much about them since they don't appear in +// stack traces. The are special data. +static bool ParseSpecialName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'T') && ParseCharClass(state, "VTISH") && + ParseType(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "Tc") && ParseCallOffset(state) && + ParseCallOffset(state) && ParseEncoding(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "GV") && ParseName(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'T') && ParseCallOffset(state) && + ParseEncoding(state)) { + return true; + } + state->parse_state = copy; + + // G++ extensions + if (ParseTwoCharToken(state, "TC") && ParseType(state) && + ParseNumber(state, nullptr) && ParseOneCharToken(state, '_') && + DisableAppend(state) && ParseType(state)) { + RestoreAppend(state, copy.append); + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'T') && ParseCharClass(state, "FJ") && + ParseType(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "GR") && ParseName(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "GA") && ParseEncoding(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'T') && ParseCharClass(state, "hv") && + ParseCallOffset(state) && ParseEncoding(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= h _ +// ::= v _ +static bool ParseCallOffset(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'h') && ParseNVOffset(state) && + ParseOneCharToken(state, '_')) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'v') && ParseVOffset(state) && + ParseOneCharToken(state, '_')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= <(offset) number> +static bool ParseNVOffset(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + return ParseNumber(state, nullptr); +} + +// ::= <(offset) number> _ <(virtual offset) number> +static bool ParseVOffset(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseNumber(state, nullptr) && ParseOneCharToken(state, '_') && + ParseNumber(state, nullptr)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= C1 | C2 | C3 | CI1 | CI2 +// +// ::= D0 | D1 | D2 +// # GCC extensions: "unified" constructor/destructor. See +// # +// https://github.com/gcc-mirror/gcc/blob/7ad17b583c3643bd4557f29b8391ca7ef08391f5/gcc/cp/mangle.c#L1847 +// ::= C4 | D4 +static bool ParseCtorDtorName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'C')) { + if (ParseCharClass(state, "1234")) { + const char *const prev_name = + state->out + state->parse_state.prev_name_idx; + MaybeAppendWithLength(state, prev_name, + state->parse_state.prev_name_length); + return true; + } else if (ParseOneCharToken(state, 'I') && ParseCharClass(state, "12") && + ParseClassEnumType(state)) { + return true; + } + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'D') && ParseCharClass(state, "0124")) { + const char *const prev_name = state->out + state->parse_state.prev_name_idx; + MaybeAppend(state, "~"); + MaybeAppendWithLength(state, prev_name, + state->parse_state.prev_name_length); + return true; + } + state->parse_state = copy; + return false; +} + +// ::= Dt E # decltype of an id-expression or class +// # member access (C++0x) +// ::= DT E # decltype of an expression (C++0x) +static bool ParseDecltype(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'D') && ParseCharClass(state, "tT") && + ParseExpression(state) && ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= +// ::= P # pointer-to +// ::= R # reference-to +// ::= O # rvalue reference-to (C++0x) +// ::= C # complex pair (C 2000) +// ::= G # imaginary (C 2000) +// ::= U # vendor extended type qualifier +// ::= +// ::= +// ::= # note: just an alias for +// ::= +// ::= +// ::= +// ::= +// ::= +// ::= +// ::= Dp # pack expansion of (C++0x) +// ::= Dv _ # GNU vector extension +// +static bool ParseType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + + // We should check CV-qualifers, and PRGC things first. + // + // CV-qualifiers overlap with some operator names, but an operator name is not + // valid as a type. To avoid an ambiguity that can lead to exponential time + // complexity, refuse to backtrack the CV-qualifiers. + // + // _Z4aoeuIrMvvE + // => _Z 4aoeuI rM v v E + // aoeu + // => _Z 4aoeuI r Mv v E + // aoeu + // + // By consuming the CV-qualifiers first, the former parse is disabled. + if (ParseCVQualifiers(state)) { + const bool result = ParseType(state); + if (!result) state->parse_state = copy; + return result; + } + state->parse_state = copy; + + // Similarly, these tag characters can overlap with other s resulting in + // two different parse prefixes that land on in the same + // place, such as "C3r1xI...". So, disable the "ctor-name = C3" parse by + // refusing to backtrack the tag characters. + if (ParseCharClass(state, "OPRCG")) { + const bool result = ParseType(state); + if (!result) state->parse_state = copy; + return result; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "Dp") && ParseType(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'U') && ParseSourceName(state) && + ParseType(state)) { + return true; + } + state->parse_state = copy; + + if (ParseBuiltinType(state) || ParseFunctionType(state) || + ParseClassEnumType(state) || ParseArrayType(state) || + ParsePointerToMemberType(state) || ParseDecltype(state) || + // "std" on its own isn't a type. + ParseSubstitution(state, /*accept_std=*/false)) { + return true; + } + + if (ParseTemplateTemplateParam(state) && ParseTemplateArgs(state)) { + return true; + } + state->parse_state = copy; + + // Less greedy than . + if (ParseTemplateParam(state)) { + return true; + } + + if (ParseTwoCharToken(state, "Dv") && ParseNumber(state, nullptr) && + ParseOneCharToken(state, '_')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= [r] [V] [K] +// We don't allow empty to avoid infinite loop in +// ParseType(). +static bool ParseCVQualifiers(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + int num_cv_qualifiers = 0; + num_cv_qualifiers += ParseOneCharToken(state, 'r'); + num_cv_qualifiers += ParseOneCharToken(state, 'V'); + num_cv_qualifiers += ParseOneCharToken(state, 'K'); + return num_cv_qualifiers > 0; +} + +// ::= v, etc. # single-character builtin types +// ::= u +// ::= Dd, etc. # two-character builtin types +// +// Not supported: +// ::= DF _ # _FloatN (N bits) +// +static bool ParseBuiltinType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + const AbbrevPair *p; + for (p = kBuiltinTypeList; p->abbrev != nullptr; ++p) { + // Guaranteed only 1- or 2-character strings in kBuiltinTypeList. + if (p->abbrev[1] == '\0') { + if (ParseOneCharToken(state, p->abbrev[0])) { + MaybeAppend(state, p->real_name); + return true; + } + } else if (p->abbrev[2] == '\0' && ParseTwoCharToken(state, p->abbrev)) { + MaybeAppend(state, p->real_name); + return true; + } + } + + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'u') && ParseSourceName(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= Do # non-throwing +// exception-specification (e.g., +// noexcept, throw()) +// ::= DO E # computed (instantiation-dependent) +// noexcept +// ::= Dw + E # dynamic exception specification +// with instantiation-dependent types +static bool ParseExceptionSpec(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + + if (ParseTwoCharToken(state, "Do")) return true; + + ParseState copy = state->parse_state; + if (ParseTwoCharToken(state, "DO") && ParseExpression(state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + if (ParseTwoCharToken(state, "Dw") && OneOrMore(ParseType, state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= [exception-spec] F [Y] [O] E +static bool ParseFunctionType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (Optional(ParseExceptionSpec(state)) && ParseOneCharToken(state, 'F') && + Optional(ParseOneCharToken(state, 'Y')) && ParseBareFunctionType(state) && + Optional(ParseOneCharToken(state, 'O')) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= <(signature) type>+ +static bool ParseBareFunctionType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + DisableAppend(state); + if (OneOrMore(ParseType, state)) { + RestoreAppend(state, copy.append); + MaybeAppend(state, "()"); + return true; + } + state->parse_state = copy; + return false; +} + +// ::= +static bool ParseClassEnumType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + return ParseName(state); +} + +// ::= A <(positive dimension) number> _ <(element) type> +// ::= A [<(dimension) expression>] _ <(element) type> +static bool ParseArrayType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'A') && ParseNumber(state, nullptr) && + ParseOneCharToken(state, '_') && ParseType(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'A') && Optional(ParseExpression(state)) && + ParseOneCharToken(state, '_') && ParseType(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= M <(class) type> <(member) type> +static bool ParsePointerToMemberType(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'M') && ParseType(state) && ParseType(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= T_ +// ::= T _ +static bool ParseTemplateParam(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseTwoCharToken(state, "T_")) { + MaybeAppend(state, "?"); // We don't support template substitutions. + return true; + } + + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'T') && ParseNumber(state, nullptr) && + ParseOneCharToken(state, '_')) { + MaybeAppend(state, "?"); // We don't support template substitutions. + return true; + } + state->parse_state = copy; + return false; +} + +// ::= +// ::= +static bool ParseTemplateTemplateParam(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + return (ParseTemplateParam(state) || + // "std" on its own isn't a template. + ParseSubstitution(state, /*accept_std=*/false)); +} + +// ::= I + E +static bool ParseTemplateArgs(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + DisableAppend(state); + if (ParseOneCharToken(state, 'I') && OneOrMore(ParseTemplateArg, state) && + ParseOneCharToken(state, 'E')) { + RestoreAppend(state, copy.append); + MaybeAppend(state, "<>"); + return true; + } + state->parse_state = copy; + return false; +} + +// ::= +// ::= +// ::= J * E # argument pack +// ::= X E +static bool ParseTemplateArg(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'J') && ZeroOrMore(ParseTemplateArg, state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + // There can be significant overlap between the following leading to + // exponential backtracking: + // + // ::= L E + // e.g. L 2xxIvE 1 E + // ==> + // e.g. L 2xx IvE + // + // This means parsing an entire twice, and can contain + // , so this can generate exponential backtracking. There is + // only overlap when the remaining input starts with "L ", so + // parse all cases that can start this way jointly to share the common prefix. + // + // We have: + // + // ::= + // ::= + // + // First, drop all the productions of that must start with something + // other than 'L'. All that's left is ; inline it. + // + // ::= # starts with 'N' + // ::= + // ::= + // ::= # starts with 'Z' + // + // Drop and inline again: + // + // ::= + // ::= + // ::= # starts with 'S' + // + // Merge the first two, inline , drop last: + // + // ::= [] + // ::= St [] # starts with 'S' + // + // Drop and inline: + // + // ::= [] # starts with lowercase + // ::= [] # starts with 'C' or 'D' + // ::= [] # starts with digit + // ::= [] + // ::= [] # starts with 'U' + // + // One more time: + // + // ::= L [] + // + // Likewise with : + // + // ::= L E + // ::= LZ E # cannot overlap; drop + // ::= L E # cannot overlap; drop + // + // By similar reasoning as shown above, the only s starting with + // are " []". Inline this. + // + // ::= L [] E + // + // Now inline both of these into : + // + // ::= L [] + // ::= L [] E + // + // Merge them and we're done: + // + // ::= L [] [ E] + if (ParseLocalSourceName(state) && Optional(ParseTemplateArgs(state))) { + copy = state->parse_state; + if (ParseExprCastValue(state) && ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + return true; + } + + // Now that the overlapping cases can't reach this code, we can safely call + // both of these. + if (ParseType(state) || ParseExprPrimary(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'X') && ParseExpression(state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= [] +// ::= +// ::= +static inline bool ParseUnresolvedType(State *state) { + // No ComplexityGuard because we don't copy the state in this stack frame. + return (ParseTemplateParam(state) && Optional(ParseTemplateArgs(state))) || + ParseDecltype(state) || ParseSubstitution(state, /*accept_std=*/false); +} + +// ::= [] +static inline bool ParseSimpleId(State *state) { + // No ComplexityGuard because we don't copy the state in this stack frame. + + // Note: cannot be followed by a parameter pack; see comment in + // ParseUnresolvedType. + return ParseSourceName(state) && Optional(ParseTemplateArgs(state)); +} + +// ::= [] +// ::= on [] +// ::= dn +static bool ParseBaseUnresolvedName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + + if (ParseSimpleId(state)) { + return true; + } + + ParseState copy = state->parse_state; + if (ParseTwoCharToken(state, "on") && ParseOperatorName(state, nullptr) && + Optional(ParseTemplateArgs(state))) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "dn") && + (ParseUnresolvedType(state) || ParseSimpleId(state))) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= [gs] +// ::= sr +// ::= srN + E +// +// ::= [gs] sr + E +// +static bool ParseUnresolvedName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + + ParseState copy = state->parse_state; + if (Optional(ParseTwoCharToken(state, "gs")) && + ParseBaseUnresolvedName(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "sr") && ParseUnresolvedType(state) && + ParseBaseUnresolvedName(state)) { + return true; + } + state->parse_state = copy; + + if (ParseTwoCharToken(state, "sr") && ParseOneCharToken(state, 'N') && + ParseUnresolvedType(state) && + OneOrMore(/* ::= */ ParseSimpleId, state) && + ParseOneCharToken(state, 'E') && ParseBaseUnresolvedName(state)) { + return true; + } + state->parse_state = copy; + + if (Optional(ParseTwoCharToken(state, "gs")) && + ParseTwoCharToken(state, "sr") && + OneOrMore(/* ::= */ ParseSimpleId, state) && + ParseOneCharToken(state, 'E') && ParseBaseUnresolvedName(state)) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= <1-ary operator-name> +// ::= <2-ary operator-name> +// ::= <3-ary operator-name> +// ::= cl + E +// ::= cp * E # Clang-specific. +// ::= cv # type (expression) +// ::= cv _ * E # type (expr-list) +// ::= st +// ::= +// ::= +// ::= +// ::= dt # expr.name +// ::= pt # expr->name +// ::= sp # argument pack expansion +// ::= sr +// ::= sr +// ::= fp <(top-level) CV-qualifiers> _ +// ::= fp <(top-level) CV-qualifiers> _ +// ::= fL p <(top-level) CV-qualifiers> _ +// ::= fL p <(top-level) CV-qualifiers> _ +static bool ParseExpression(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseTemplateParam(state) || ParseExprPrimary(state)) { + return true; + } + + ParseState copy = state->parse_state; + + // Object/function call expression. + if (ParseTwoCharToken(state, "cl") && OneOrMore(ParseExpression, state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + // Clang-specific "cp * E" + // https://clang.llvm.org/doxygen/ItaniumMangle_8cpp_source.html#l04338 + if (ParseTwoCharToken(state, "cp") && ParseSimpleId(state) && + ZeroOrMore(ParseExpression, state) && ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + // Function-param expression (level 0). + if (ParseTwoCharToken(state, "fp") && Optional(ParseCVQualifiers(state)) && + Optional(ParseNumber(state, nullptr)) && ParseOneCharToken(state, '_')) { + return true; + } + state->parse_state = copy; + + // Function-param expression (level 1+). + if (ParseTwoCharToken(state, "fL") && Optional(ParseNumber(state, nullptr)) && + ParseOneCharToken(state, 'p') && Optional(ParseCVQualifiers(state)) && + Optional(ParseNumber(state, nullptr)) && ParseOneCharToken(state, '_')) { + return true; + } + state->parse_state = copy; + + // Parse the conversion expressions jointly to avoid re-parsing the in + // their common prefix. Parsed as: + // ::= cv + // ::= _ * E + // ::= + // + // Also don't try ParseOperatorName after seeing "cv", since ParseOperatorName + // also needs to accept "cv " in other contexts. + if (ParseTwoCharToken(state, "cv")) { + if (ParseType(state)) { + ParseState copy2 = state->parse_state; + if (ParseOneCharToken(state, '_') && ZeroOrMore(ParseExpression, state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy2; + if (ParseExpression(state)) { + return true; + } + } + } else { + // Parse unary, binary, and ternary operator expressions jointly, taking + // care not to re-parse subexpressions repeatedly. Parse like: + // ::= + // [] + // ::= [] + int arity = -1; + if (ParseOperatorName(state, &arity) && + arity > 0 && // 0 arity => disabled. + (arity < 3 || ParseExpression(state)) && + (arity < 2 || ParseExpression(state)) && + (arity < 1 || ParseExpression(state))) { + return true; + } + } + state->parse_state = copy; + + // sizeof type + if (ParseTwoCharToken(state, "st") && ParseType(state)) { + return true; + } + state->parse_state = copy; + + // Object and pointer member access expressions. + if ((ParseTwoCharToken(state, "dt") || ParseTwoCharToken(state, "pt")) && + ParseExpression(state) && ParseType(state)) { + return true; + } + state->parse_state = copy; + + // Pointer-to-member access expressions. This parses the same as a binary + // operator, but it's implemented separately because "ds" shouldn't be + // accepted in other contexts that parse an operator name. + if (ParseTwoCharToken(state, "ds") && ParseExpression(state) && + ParseExpression(state)) { + return true; + } + state->parse_state = copy; + + // Parameter pack expansion + if (ParseTwoCharToken(state, "sp") && ParseExpression(state)) { + return true; + } + state->parse_state = copy; + + return ParseUnresolvedName(state); +} + +// ::= L <(value) number> E +// ::= L <(value) float> E +// ::= L E +// // A bug in g++'s C++ ABI version 2 (-fabi-version=2). +// ::= LZ E +// +// Warning, subtle: the "bug" LZ production above is ambiguous with the first +// production where starts with , which can lead to +// exponential backtracking in two scenarios: +// +// - When whatever follows the E in the in the first production is +// not a name, we backtrack the whole and re-parse the whole thing. +// +// - When whatever follows the in the first production is not a +// number and this may be followed by a name, we backtrack the +// and re-parse it. +// +// Moreover this ambiguity isn't always resolved -- for example, the following +// has two different parses: +// +// _ZaaILZ4aoeuE1x1EvE +// => operator&& +// => operator&&<(aoeu::x)(1), void> +// +// To resolve this, we just do what GCC's demangler does, and refuse to parse +// casts to types. +static bool ParseExprPrimary(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + + // The "LZ" special case: if we see LZ, we commit to accept "LZ E" + // or fail, no backtracking. + if (ParseTwoCharToken(state, "LZ")) { + if (ParseEncoding(state) && ParseOneCharToken(state, 'E')) { + return true; + } + + state->parse_state = copy; + return false; + } + + // The merged cast production. + if (ParseOneCharToken(state, 'L') && ParseType(state) && + ParseExprCastValue(state)) { + return true; + } + state->parse_state = copy; + + if (ParseOneCharToken(state, 'L') && ParseMangledName(state) && + ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// or , followed by 'E', as described above ParseExprPrimary. +static bool ParseExprCastValue(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + // We have to be able to backtrack after accepting a number because we could + // have e.g. "7fffE", which will accept "7" as a number but then fail to find + // the 'E'. + ParseState copy = state->parse_state; + if (ParseNumber(state, nullptr) && ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + if (ParseFloatNumber(state) && ParseOneCharToken(state, 'E')) { + return true; + } + state->parse_state = copy; + + return false; +} + +// ::= Z <(function) encoding> E <(entity) name> [] +// ::= Z <(function) encoding> E s [] +// +// Parsing a common prefix of these two productions together avoids an +// exponential blowup of backtracking. Parse like: +// := Z E +// ::= s [] +// ::= [] + +static bool ParseLocalNameSuffix(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + + if (MaybeAppend(state, "::") && ParseName(state) && + Optional(ParseDiscriminator(state))) { + return true; + } + + // Since we're not going to overwrite the above "::" by re-parsing the + // (whose trailing '\0' byte was in the byte now holding the + // first ':'), we have to rollback the "::" if the parse failed. + if (state->parse_state.append) { + state->out[state->parse_state.out_cur_idx - 2] = '\0'; + } + + return ParseOneCharToken(state, 's') && Optional(ParseDiscriminator(state)); +} + +static bool ParseLocalName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'Z') && ParseEncoding(state) && + ParseOneCharToken(state, 'E') && ParseLocalNameSuffix(state)) { + return true; + } + state->parse_state = copy; + return false; +} + +// := _ <(non-negative) number> +static bool ParseDiscriminator(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, '_') && ParseNumber(state, nullptr)) { + return true; + } + state->parse_state = copy; + return false; +} + +// ::= S_ +// ::= S _ +// ::= St, etc. +// +// "St" is special in that it's not valid as a standalone name, and it *is* +// allowed to precede a name without being wrapped in "N...E". This means that +// if we accept it on its own, we can accept "St1a" and try to parse +// template-args, then fail and backtrack, accept "St" on its own, then "1a" as +// an unqualified name and re-parse the same template-args. To block this +// exponential backtracking, we disable it with 'accept_std=false' in +// problematic contexts. +static bool ParseSubstitution(State *state, bool accept_std) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseTwoCharToken(state, "S_")) { + MaybeAppend(state, "?"); // We don't support substitutions. + return true; + } + + ParseState copy = state->parse_state; + if (ParseOneCharToken(state, 'S') && ParseSeqId(state) && + ParseOneCharToken(state, '_')) { + MaybeAppend(state, "?"); // We don't support substitutions. + return true; + } + state->parse_state = copy; + + // Expand abbreviations like "St" => "std". + if (ParseOneCharToken(state, 'S')) { + const AbbrevPair *p; + for (p = kSubstitutionList; p->abbrev != nullptr; ++p) { + if (RemainingInput(state)[0] == p->abbrev[1] && + (accept_std || p->abbrev[1] != 't')) { + MaybeAppend(state, "std"); + if (p->real_name[0] != '\0') { + MaybeAppend(state, "::"); + MaybeAppend(state, p->real_name); + } + ++state->parse_state.mangled_idx; + return true; + } + } + } + state->parse_state = copy; + return false; +} + +// Parse , optionally followed by either a function-clone suffix +// or version suffix. Returns true only if all of "mangled_cur" was consumed. +static bool ParseTopLevelMangledName(State *state) { + ComplexityGuard guard(state); + if (guard.IsTooComplex()) return false; + if (ParseMangledName(state)) { + if (RemainingInput(state)[0] != '\0') { + // Drop trailing function clone suffix, if any. + if (IsFunctionCloneSuffix(RemainingInput(state))) { + return true; + } + // Append trailing version suffix if any. + // ex. _Z3foo@@GLIBCXX_3.4 + if (RemainingInput(state)[0] == '@') { + MaybeAppend(state, RemainingInput(state)); + return true; + } + return false; // Unconsumed suffix. + } + return true; + } + return false; +} + +static bool Overflowed(const State *state) { + return state->parse_state.out_cur_idx >= state->out_end_idx; +} + +// The demangler entry point. +bool Demangle(const char *mangled, char *out, int out_size) { + State state; + InitState(&state, mangled, out, out_size); + return ParseTopLevelMangledName(&state) && !Overflowed(&state) && + state.parse_state.out_cur_idx > 0; +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/internal/demangle.h b/src/absl/debugging/internal/demangle.h new file mode 100644 index 0000000..c314d9b --- /dev/null +++ b/src/absl/debugging/internal/demangle.h @@ -0,0 +1,71 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// An async-signal-safe and thread-safe demangler for Itanium C++ ABI +// (aka G++ V3 ABI). +// +// The demangler is implemented to be used in async signal handlers to +// symbolize stack traces. We cannot use libstdc++'s +// abi::__cxa_demangle() in such signal handlers since it's not async +// signal safe (it uses malloc() internally). +// +// Note that this demangler doesn't support full demangling. More +// specifically, it doesn't print types of function parameters and +// types of template arguments. It just skips them. However, it's +// still very useful to extract basic information such as class, +// function, constructor, destructor, and operator names. +// +// See the implementation note in demangle.cc if you are interested. +// +// Example: +// +// | Mangled Name | The Demangler | abi::__cxa_demangle() +// |---------------|---------------|----------------------- +// | _Z1fv | f() | f() +// | _Z1fi | f() | f(int) +// | _Z3foo3bar | foo() | foo(bar) +// | _Z1fIiEvi | f<>() | void f(int) +// | _ZN1N1fE | N::f | N::f +// | _ZN3Foo3BarEv | Foo::Bar() | Foo::Bar() +// | _Zrm1XS_" | operator%() | operator%(X, X) +// | _ZN3FooC1Ev | Foo::Foo() | Foo::Foo() +// | _Z1fSs | f() | f(std::basic_string, +// | | | std::allocator >) +// +// See the unit test for more examples. +// +// Note: we might want to write demanglers for ABIs other than Itanium +// C++ ABI in the future. +// + +#ifndef ABSL_DEBUGGING_INTERNAL_DEMANGLE_H_ +#define ABSL_DEBUGGING_INTERNAL_DEMANGLE_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// Demangle `mangled`. On success, return true and write the +// demangled symbol name to `out`. Otherwise, return false. +// `out` is modified even if demangling is unsuccessful. +bool Demangle(const char *mangled, char *out, int out_size); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_DEMANGLE_H_ diff --git a/src/absl/debugging/internal/elf_mem_image.cc b/src/absl/debugging/internal/elf_mem_image.cc new file mode 100644 index 0000000..a9d6671 --- /dev/null +++ b/src/absl/debugging/internal/elf_mem_image.cc @@ -0,0 +1,387 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Allow dynamic symbol lookup in an in-memory Elf image. +// + +#include "absl/debugging/internal/elf_mem_image.h" + +#ifdef ABSL_HAVE_ELF_MEM_IMAGE // defined in elf_mem_image.h + +#include +#include +#include +#include "absl/base/config.h" +#include "absl/base/internal/raw_logging.h" + +// From binutils/include/elf/common.h (this doesn't appear to be documented +// anywhere else). +// +// /* This flag appears in a Versym structure. It means that the symbol +// is hidden, and is only visible with an explicit version number. +// This is a GNU extension. */ +// #define VERSYM_HIDDEN 0x8000 +// +// /* This is the mask for the rest of the Versym information. */ +// #define VERSYM_VERSION 0x7fff + +#define VERSYM_VERSION 0x7fff + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +namespace { + +#if __SIZEOF_POINTER__ == 4 +const int kElfClass = ELFCLASS32; +int ElfBind(const ElfW(Sym) *symbol) { return ELF32_ST_BIND(symbol->st_info); } +int ElfType(const ElfW(Sym) *symbol) { return ELF32_ST_TYPE(symbol->st_info); } +#elif __SIZEOF_POINTER__ == 8 +const int kElfClass = ELFCLASS64; +int ElfBind(const ElfW(Sym) *symbol) { return ELF64_ST_BIND(symbol->st_info); } +int ElfType(const ElfW(Sym) *symbol) { return ELF64_ST_TYPE(symbol->st_info); } +#else +const int kElfClass = -1; +int ElfBind(const ElfW(Sym) *) { + ABSL_RAW_LOG(FATAL, "Unexpected word size"); + return 0; +} +int ElfType(const ElfW(Sym) *) { + ABSL_RAW_LOG(FATAL, "Unexpected word size"); + return 0; +} +#endif + +// Extract an element from one of the ELF tables, cast it to desired type. +// This is just a simple arithmetic and a glorified cast. +// Callers are responsible for bounds checking. +template +const T *GetTableElement(const ElfW(Ehdr) * ehdr, ElfW(Off) table_offset, + ElfW(Word) element_size, size_t index) { + return reinterpret_cast(reinterpret_cast(ehdr) + + table_offset + + index * element_size); +} + +} // namespace + +// The value of this variable doesn't matter; it's used only for its +// unique address. +const int ElfMemImage::kInvalidBaseSentinel = 0; + +ElfMemImage::ElfMemImage(const void *base) { + ABSL_RAW_CHECK(base != kInvalidBase, "bad pointer"); + Init(base); +} + +int ElfMemImage::GetNumSymbols() const { + if (!hash_) { + return 0; + } + // See http://www.caldera.com/developers/gabi/latest/ch5.dynamic.html#hash + return hash_[1]; +} + +const ElfW(Sym) *ElfMemImage::GetDynsym(int index) const { + ABSL_RAW_CHECK(index < GetNumSymbols(), "index out of range"); + return dynsym_ + index; +} + +const ElfW(Versym) *ElfMemImage::GetVersym(int index) const { + ABSL_RAW_CHECK(index < GetNumSymbols(), "index out of range"); + return versym_ + index; +} + +const ElfW(Phdr) *ElfMemImage::GetPhdr(int index) const { + ABSL_RAW_CHECK(index < ehdr_->e_phnum, "index out of range"); + return GetTableElement(ehdr_, + ehdr_->e_phoff, + ehdr_->e_phentsize, + index); +} + +const char *ElfMemImage::GetDynstr(ElfW(Word) offset) const { + ABSL_RAW_CHECK(offset < strsize_, "offset out of range"); + return dynstr_ + offset; +} + +const void *ElfMemImage::GetSymAddr(const ElfW(Sym) *sym) const { + if (sym->st_shndx == SHN_UNDEF || sym->st_shndx >= SHN_LORESERVE) { + // Symbol corresponds to "special" (e.g. SHN_ABS) section. + return reinterpret_cast(sym->st_value); + } + ABSL_RAW_CHECK(link_base_ < sym->st_value, "symbol out of range"); + return GetTableElement(ehdr_, 0, 1, sym->st_value - link_base_); +} + +const ElfW(Verdef) *ElfMemImage::GetVerdef(int index) const { + ABSL_RAW_CHECK(0 <= index && static_cast(index) <= verdefnum_, + "index out of range"); + const ElfW(Verdef) *version_definition = verdef_; + while (version_definition->vd_ndx < index && version_definition->vd_next) { + const char *const version_definition_as_char = + reinterpret_cast(version_definition); + version_definition = + reinterpret_cast(version_definition_as_char + + version_definition->vd_next); + } + return version_definition->vd_ndx == index ? version_definition : nullptr; +} + +const ElfW(Verdaux) *ElfMemImage::GetVerdefAux( + const ElfW(Verdef) *verdef) const { + return reinterpret_cast(verdef+1); +} + +const char *ElfMemImage::GetVerstr(ElfW(Word) offset) const { + ABSL_RAW_CHECK(offset < strsize_, "offset out of range"); + return dynstr_ + offset; +} + +void ElfMemImage::Init(const void *base) { + ehdr_ = nullptr; + dynsym_ = nullptr; + dynstr_ = nullptr; + versym_ = nullptr; + verdef_ = nullptr; + hash_ = nullptr; + strsize_ = 0; + verdefnum_ = 0; + link_base_ = ~0L; // Sentinel: PT_LOAD .p_vaddr can't possibly be this. + if (!base) { + return; + } + const char *const base_as_char = reinterpret_cast(base); + if (base_as_char[EI_MAG0] != ELFMAG0 || base_as_char[EI_MAG1] != ELFMAG1 || + base_as_char[EI_MAG2] != ELFMAG2 || base_as_char[EI_MAG3] != ELFMAG3) { + assert(false); + return; + } + int elf_class = base_as_char[EI_CLASS]; + if (elf_class != kElfClass) { + assert(false); + return; + } + switch (base_as_char[EI_DATA]) { + case ELFDATA2LSB: { +#ifndef ABSL_IS_LITTLE_ENDIAN + assert(false); + return; +#endif + break; + } + case ELFDATA2MSB: { +#ifndef ABSL_IS_BIG_ENDIAN + assert(false); + return; +#endif + break; + } + default: { + assert(false); + return; + } + } + + ehdr_ = reinterpret_cast(base); + const ElfW(Phdr) *dynamic_program_header = nullptr; + for (int i = 0; i < ehdr_->e_phnum; ++i) { + const ElfW(Phdr) *const program_header = GetPhdr(i); + switch (program_header->p_type) { + case PT_LOAD: + if (!~link_base_) { + link_base_ = program_header->p_vaddr; + } + break; + case PT_DYNAMIC: + dynamic_program_header = program_header; + break; + } + } + if (!~link_base_ || !dynamic_program_header) { + assert(false); + // Mark this image as not present. Can not recur infinitely. + Init(nullptr); + return; + } + ptrdiff_t relocation = + base_as_char - reinterpret_cast(link_base_); + ElfW(Dyn) *dynamic_entry = + reinterpret_cast(dynamic_program_header->p_vaddr + + relocation); + for (; dynamic_entry->d_tag != DT_NULL; ++dynamic_entry) { + const auto value = dynamic_entry->d_un.d_val + relocation; + switch (dynamic_entry->d_tag) { + case DT_HASH: + hash_ = reinterpret_cast(value); + break; + case DT_SYMTAB: + dynsym_ = reinterpret_cast(value); + break; + case DT_STRTAB: + dynstr_ = reinterpret_cast(value); + break; + case DT_VERSYM: + versym_ = reinterpret_cast(value); + break; + case DT_VERDEF: + verdef_ = reinterpret_cast(value); + break; + case DT_VERDEFNUM: + verdefnum_ = dynamic_entry->d_un.d_val; + break; + case DT_STRSZ: + strsize_ = dynamic_entry->d_un.d_val; + break; + default: + // Unrecognized entries explicitly ignored. + break; + } + } + if (!hash_ || !dynsym_ || !dynstr_ || !versym_ || + !verdef_ || !verdefnum_ || !strsize_) { + assert(false); // invalid VDSO + // Mark this image as not present. Can not recur infinitely. + Init(nullptr); + return; + } +} + +bool ElfMemImage::LookupSymbol(const char *name, + const char *version, + int type, + SymbolInfo *info_out) const { + for (const SymbolInfo& info : *this) { + if (strcmp(info.name, name) == 0 && strcmp(info.version, version) == 0 && + ElfType(info.symbol) == type) { + if (info_out) { + *info_out = info; + } + return true; + } + } + return false; +} + +bool ElfMemImage::LookupSymbolByAddress(const void *address, + SymbolInfo *info_out) const { + for (const SymbolInfo& info : *this) { + const char *const symbol_start = + reinterpret_cast(info.address); + const char *const symbol_end = symbol_start + info.symbol->st_size; + if (symbol_start <= address && address < symbol_end) { + if (info_out) { + // Client wants to know details for that symbol (the usual case). + if (ElfBind(info.symbol) == STB_GLOBAL) { + // Strong symbol; just return it. + *info_out = info; + return true; + } else { + // Weak or local. Record it, but keep looking for a strong one. + *info_out = info; + } + } else { + // Client only cares if there is an overlapping symbol. + return true; + } + } + } + return false; +} + +ElfMemImage::SymbolIterator::SymbolIterator(const void *const image, int index) + : index_(index), image_(image) { +} + +const ElfMemImage::SymbolInfo *ElfMemImage::SymbolIterator::operator->() const { + return &info_; +} + +const ElfMemImage::SymbolInfo& ElfMemImage::SymbolIterator::operator*() const { + return info_; +} + +bool ElfMemImage::SymbolIterator::operator==(const SymbolIterator &rhs) const { + return this->image_ == rhs.image_ && this->index_ == rhs.index_; +} + +bool ElfMemImage::SymbolIterator::operator!=(const SymbolIterator &rhs) const { + return !(*this == rhs); +} + +ElfMemImage::SymbolIterator &ElfMemImage::SymbolIterator::operator++() { + this->Update(1); + return *this; +} + +ElfMemImage::SymbolIterator ElfMemImage::begin() const { + SymbolIterator it(this, 0); + it.Update(0); + return it; +} + +ElfMemImage::SymbolIterator ElfMemImage::end() const { + return SymbolIterator(this, GetNumSymbols()); +} + +void ElfMemImage::SymbolIterator::Update(int increment) { + const ElfMemImage *image = reinterpret_cast(image_); + ABSL_RAW_CHECK(image->IsPresent() || increment == 0, ""); + if (!image->IsPresent()) { + return; + } + index_ += increment; + if (index_ >= image->GetNumSymbols()) { + index_ = image->GetNumSymbols(); + return; + } + const ElfW(Sym) *symbol = image->GetDynsym(index_); + const ElfW(Versym) *version_symbol = image->GetVersym(index_); + ABSL_RAW_CHECK(symbol && version_symbol, ""); + const char *const symbol_name = image->GetDynstr(symbol->st_name); +#if defined(__NetBSD__) + const int version_index = version_symbol->vs_vers & VERSYM_VERSION; +#else + const ElfW(Versym) version_index = version_symbol[0] & VERSYM_VERSION; +#endif + const ElfW(Verdef) *version_definition = nullptr; + const char *version_name = ""; + if (symbol->st_shndx == SHN_UNDEF) { + // Undefined symbols reference DT_VERNEED, not DT_VERDEF, and + // version_index could well be greater than verdefnum_, so calling + // GetVerdef(version_index) may trigger assertion. + } else { + version_definition = image->GetVerdef(version_index); + } + if (version_definition) { + // I am expecting 1 or 2 auxiliary entries: 1 for the version itself, + // optional 2nd if the version has a parent. + ABSL_RAW_CHECK( + version_definition->vd_cnt == 1 || version_definition->vd_cnt == 2, + "wrong number of entries"); + const ElfW(Verdaux) *version_aux = image->GetVerdefAux(version_definition); + version_name = image->GetVerstr(version_aux->vda_name); + } + info_.name = symbol_name; + info_.version = version_name; + info_.address = image->GetSymAddr(symbol); + info_.symbol = symbol; +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_HAVE_ELF_MEM_IMAGE diff --git a/src/absl/debugging/internal/elf_mem_image.h b/src/absl/debugging/internal/elf_mem_image.h new file mode 100644 index 0000000..113071a --- /dev/null +++ b/src/absl/debugging/internal/elf_mem_image.h @@ -0,0 +1,139 @@ +/* + * Copyright 2017 The Abseil Authors. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * https://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +// Allow dynamic symbol lookup for in-memory Elf images. + +#ifndef ABSL_DEBUGGING_INTERNAL_ELF_MEM_IMAGE_H_ +#define ABSL_DEBUGGING_INTERNAL_ELF_MEM_IMAGE_H_ + +// Including this will define the __GLIBC__ macro if glibc is being +// used. +#include + +#include "absl/base/config.h" + +// Maybe one day we can rewrite this file not to require the elf +// symbol extensions in glibc, but for right now we need them. +#ifdef ABSL_HAVE_ELF_MEM_IMAGE +#error ABSL_HAVE_ELF_MEM_IMAGE cannot be directly set +#endif + +#if defined(__ELF__) && !defined(__OpenBSD__) && !defined(__QNX__) && \ + !defined(__native_client__) && !defined(__asmjs__) && \ + !defined(__wasm__) && !defined(__HAIKU__) +#define ABSL_HAVE_ELF_MEM_IMAGE 1 +#endif + +#ifdef ABSL_HAVE_ELF_MEM_IMAGE + +#include // for ElfW + +#if defined(__FreeBSD__) && !defined(ElfW) +#define ElfW(x) __ElfN(x) +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// An in-memory ELF image (may not exist on disk). +class ElfMemImage { + private: + // Sentinel: there could never be an elf image at &kInvalidBaseSentinel. + static const int kInvalidBaseSentinel; + + public: + // Sentinel: there could never be an elf image at this address. + static constexpr const void *const kInvalidBase = + static_cast(&kInvalidBaseSentinel); + + // Information about a single vdso symbol. + // All pointers are into .dynsym, .dynstr, or .text of the VDSO. + // Do not free() them or modify through them. + struct SymbolInfo { + const char *name; // E.g. "__vdso_getcpu" + const char *version; // E.g. "LINUX_2.6", could be "" + // for unversioned symbol. + const void *address; // Relocated symbol address. + const ElfW(Sym) *symbol; // Symbol in the dynamic symbol table. + }; + + // Supports iteration over all dynamic symbols. + class SymbolIterator { + public: + friend class ElfMemImage; + const SymbolInfo *operator->() const; + const SymbolInfo &operator*() const; + SymbolIterator& operator++(); + bool operator!=(const SymbolIterator &rhs) const; + bool operator==(const SymbolIterator &rhs) const; + private: + SymbolIterator(const void *const image, int index); + void Update(int incr); + SymbolInfo info_; + int index_; + const void *const image_; + }; + + + explicit ElfMemImage(const void *base); + void Init(const void *base); + bool IsPresent() const { return ehdr_ != nullptr; } + const ElfW(Phdr)* GetPhdr(int index) const; + const ElfW(Sym)* GetDynsym(int index) const; + const ElfW(Versym)* GetVersym(int index) const; + const ElfW(Verdef)* GetVerdef(int index) const; + const ElfW(Verdaux)* GetVerdefAux(const ElfW(Verdef) *verdef) const; + const char* GetDynstr(ElfW(Word) offset) const; + const void* GetSymAddr(const ElfW(Sym) *sym) const; + const char* GetVerstr(ElfW(Word) offset) const; + int GetNumSymbols() const; + + SymbolIterator begin() const; + SymbolIterator end() const; + + // Look up versioned dynamic symbol in the image. + // Returns false if image is not present, or doesn't contain given + // symbol/version/type combination. + // If info_out is non-null, additional details are filled in. + bool LookupSymbol(const char *name, const char *version, + int symbol_type, SymbolInfo *info_out) const; + + // Find info about symbol (if any) which overlaps given address. + // Returns true if symbol was found; false if image isn't present + // or doesn't have a symbol overlapping given address. + // If info_out is non-null, additional details are filled in. + bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const; + + private: + const ElfW(Ehdr) *ehdr_; + const ElfW(Sym) *dynsym_; + const ElfW(Versym) *versym_; + const ElfW(Verdef) *verdef_; + const ElfW(Word) *hash_; + const char *dynstr_; + size_t strsize_; + size_t verdefnum_; + ElfW(Addr) link_base_; // Link-time base (p_vaddr of first PT_LOAD). +}; + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_HAVE_ELF_MEM_IMAGE + +#endif // ABSL_DEBUGGING_INTERNAL_ELF_MEM_IMAGE_H_ diff --git a/src/absl/debugging/internal/examine_stack.cc b/src/absl/debugging/internal/examine_stack.cc new file mode 100644 index 0000000..5bdd341 --- /dev/null +++ b/src/absl/debugging/internal/examine_stack.cc @@ -0,0 +1,315 @@ +// +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#include "absl/debugging/internal/examine_stack.h" + +#ifndef _WIN32 +#include +#endif + +#include "absl/base/config.h" + +#ifdef ABSL_HAVE_MMAP +#include +#endif + +#if defined(__linux__) || defined(__APPLE__) +#include +#endif + +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/macros.h" +#include "absl/debugging/stacktrace.h" +#include "absl/debugging/symbolize.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +namespace { +constexpr int kDefaultDumpStackFramesLimit = 64; +// The %p field width for printf() functions is two characters per byte, +// and two extra for the leading "0x". +constexpr int kPrintfPointerFieldWidth = 2 + 2 * sizeof(void*); + +ABSL_CONST_INIT SymbolizeUrlEmitter debug_stack_trace_hook = nullptr; + +// Async-signal safe mmap allocator. +void* Allocate(size_t num_bytes) { +#ifdef ABSL_HAVE_MMAP + void* p = ::mmap(nullptr, num_bytes, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + return p == MAP_FAILED ? nullptr : p; +#else + (void)num_bytes; + return nullptr; +#endif // ABSL_HAVE_MMAP +} + +void Deallocate(void* p, size_t size) { +#ifdef ABSL_HAVE_MMAP + ::munmap(p, size); +#else + (void)p; + (void)size; +#endif // ABSL_HAVE_MMAP +} + +// Print a program counter only. +void DumpPC(OutputWriter* writer, void* writer_arg, void* const pc, + const char* const prefix) { + char buf[100]; + snprintf(buf, sizeof(buf), "%s@ %*p\n", prefix, kPrintfPointerFieldWidth, pc); + writer(buf, writer_arg); +} + +// Print a program counter and the corresponding stack frame size. +void DumpPCAndFrameSize(OutputWriter* writer, void* writer_arg, void* const pc, + int framesize, const char* const prefix) { + char buf[100]; + if (framesize <= 0) { + snprintf(buf, sizeof(buf), "%s@ %*p (unknown)\n", prefix, + kPrintfPointerFieldWidth, pc); + } else { + snprintf(buf, sizeof(buf), "%s@ %*p %9d\n", prefix, + kPrintfPointerFieldWidth, pc, framesize); + } + writer(buf, writer_arg); +} + +// Print a program counter and the corresponding symbol. +void DumpPCAndSymbol(OutputWriter* writer, void* writer_arg, void* const pc, + const char* const prefix) { + char tmp[1024]; + const char* symbol = "(unknown)"; + // Symbolizes the previous address of pc because pc may be in the + // next function. The overrun happens when the function ends with + // a call to a function annotated noreturn (e.g. CHECK). + // If symbolization of pc-1 fails, also try pc on the off-chance + // that we crashed on the first instruction of a function (that + // actually happens very often for e.g. __restore_rt). + const uintptr_t prev_pc = reinterpret_cast(pc) - 1; + if (absl::Symbolize(reinterpret_cast(prev_pc), tmp, + sizeof(tmp)) || + absl::Symbolize(pc, tmp, sizeof(tmp))) { + symbol = tmp; + } + char buf[1024]; + snprintf(buf, sizeof(buf), "%s@ %*p %s\n", prefix, kPrintfPointerFieldWidth, + pc, symbol); + writer(buf, writer_arg); +} + +// Print a program counter, its stack frame size, and its symbol name. +// Note that there is a separate symbolize_pc argument. Return addresses may be +// at the end of the function, and this allows the caller to back up from pc if +// appropriate. +void DumpPCAndFrameSizeAndSymbol(OutputWriter* writer, void* writer_arg, + void* const pc, void* const symbolize_pc, + int framesize, const char* const prefix) { + char tmp[1024]; + const char* symbol = "(unknown)"; + if (absl::Symbolize(symbolize_pc, tmp, sizeof(tmp))) { + symbol = tmp; + } + char buf[1024]; + if (framesize <= 0) { + snprintf(buf, sizeof(buf), "%s@ %*p (unknown) %s\n", prefix, + kPrintfPointerFieldWidth, pc, symbol); + } else { + snprintf(buf, sizeof(buf), "%s@ %*p %9d %s\n", prefix, + kPrintfPointerFieldWidth, pc, framesize, symbol); + } + writer(buf, writer_arg); +} + +} // namespace + +void RegisterDebugStackTraceHook(SymbolizeUrlEmitter hook) { + debug_stack_trace_hook = hook; +} + +SymbolizeUrlEmitter GetDebugStackTraceHook() { return debug_stack_trace_hook; } + +// Returns the program counter from signal context, nullptr if +// unknown. vuc is a ucontext_t*. We use void* to avoid the use of +// ucontext_t on non-POSIX systems. +void* GetProgramCounter(void* const vuc) { +#ifdef __linux__ + if (vuc != nullptr) { + ucontext_t* context = reinterpret_cast(vuc); +#if defined(__aarch64__) + return reinterpret_cast(context->uc_mcontext.pc); +#elif defined(__alpha__) + return reinterpret_cast(context->uc_mcontext.sc_pc); +#elif defined(__arm__) + return reinterpret_cast(context->uc_mcontext.arm_pc); +#elif defined(__hppa__) + return reinterpret_cast(context->uc_mcontext.sc_iaoq[0]); +#elif defined(__i386__) + if (14 < ABSL_ARRAYSIZE(context->uc_mcontext.gregs)) + return reinterpret_cast(context->uc_mcontext.gregs[14]); +#elif defined(__ia64__) + return reinterpret_cast(context->uc_mcontext.sc_ip); +#elif defined(__m68k__) + return reinterpret_cast(context->uc_mcontext.gregs[16]); +#elif defined(__mips__) + return reinterpret_cast(context->uc_mcontext.pc); +#elif defined(__powerpc64__) + return reinterpret_cast(context->uc_mcontext.gp_regs[32]); +#elif defined(__powerpc__) + return reinterpret_cast(context->uc_mcontext.uc_regs->gregs[32]); +#elif defined(__riscv) + return reinterpret_cast(context->uc_mcontext.__gregs[REG_PC]); +#elif defined(__s390__) && !defined(__s390x__) + return reinterpret_cast(context->uc_mcontext.psw.addr & 0x7fffffff); +#elif defined(__s390__) && defined(__s390x__) + return reinterpret_cast(context->uc_mcontext.psw.addr); +#elif defined(__sh__) + return reinterpret_cast(context->uc_mcontext.pc); +#elif defined(__sparc__) && !defined(__arch64__) + return reinterpret_cast(context->uc_mcontext.gregs[19]); +#elif defined(__sparc__) && defined(__arch64__) + return reinterpret_cast(context->uc_mcontext.mc_gregs[19]); +#elif defined(__x86_64__) + if (16 < ABSL_ARRAYSIZE(context->uc_mcontext.gregs)) + return reinterpret_cast(context->uc_mcontext.gregs[16]); +#elif defined(__e2k__) + return reinterpret_cast(context->uc_mcontext.cr0_hi); +#elif defined(__loongarch__) + return reinterpret_cast(context->uc_mcontext.__pc); +#else +#error "Undefined Architecture." +#endif + } +#elif defined(__APPLE__) + if (vuc != nullptr) { + ucontext_t* signal_ucontext = reinterpret_cast(vuc); +#if defined(__aarch64__) + return reinterpret_cast( + __darwin_arm_thread_state64_get_pc(signal_ucontext->uc_mcontext->__ss)); +#elif defined(__arm__) +#if __DARWIN_UNIX03 + return reinterpret_cast(signal_ucontext->uc_mcontext->__ss.__pc); +#else + return reinterpret_cast(signal_ucontext->uc_mcontext->ss.pc); +#endif +#elif defined(__i386__) +#if __DARWIN_UNIX03 + return reinterpret_cast(signal_ucontext->uc_mcontext->__ss.__eip); +#else + return reinterpret_cast(signal_ucontext->uc_mcontext->ss.eip); +#endif +#elif defined(__x86_64__) +#if __DARWIN_UNIX03 + return reinterpret_cast(signal_ucontext->uc_mcontext->__ss.__rip); +#else + return reinterpret_cast(signal_ucontext->uc_mcontext->ss.rip); +#endif +#endif + } +#elif defined(__akaros__) + auto* ctx = reinterpret_cast(vuc); + return reinterpret_cast(get_user_ctx_pc(ctx)); +#endif + static_cast(vuc); + return nullptr; +} + +void DumpPCAndFrameSizesAndStackTrace(void* const pc, void* const stack[], + int frame_sizes[], int depth, + int min_dropped_frames, + bool symbolize_stacktrace, + OutputWriter* writer, void* writer_arg) { + if (pc != nullptr) { + // We don't know the stack frame size for PC, use 0. + if (symbolize_stacktrace) { + DumpPCAndFrameSizeAndSymbol(writer, writer_arg, pc, pc, 0, "PC: "); + } else { + DumpPCAndFrameSize(writer, writer_arg, pc, 0, "PC: "); + } + } + for (int i = 0; i < depth; i++) { + if (symbolize_stacktrace) { + // Pass the previous address of pc as the symbol address because pc is a + // return address, and an overrun may occur when the function ends with a + // call to a function annotated noreturn (e.g. CHECK). Note that we don't + // do this for pc above, as the adjustment is only correct for return + // addresses. + DumpPCAndFrameSizeAndSymbol(writer, writer_arg, stack[i], + reinterpret_cast(stack[i]) - 1, + frame_sizes[i], " "); + } else { + DumpPCAndFrameSize(writer, writer_arg, stack[i], frame_sizes[i], " "); + } + } + if (min_dropped_frames > 0) { + char buf[100]; + snprintf(buf, sizeof(buf), " @ ... and at least %d more frames\n", + min_dropped_frames); + writer(buf, writer_arg); + } +} + +// Dump current stack trace as directed by writer. +// Make sure this function is not inlined to avoid skipping too many top frames. +ABSL_ATTRIBUTE_NOINLINE +void DumpStackTrace(int min_dropped_frames, int max_num_frames, + bool symbolize_stacktrace, OutputWriter* writer, + void* writer_arg) { + // Print stack trace + void* stack_buf[kDefaultDumpStackFramesLimit]; + void** stack = stack_buf; + int num_stack = kDefaultDumpStackFramesLimit; + int allocated_bytes = 0; + + if (num_stack >= max_num_frames) { + // User requested fewer frames than we already have space for. + num_stack = max_num_frames; + } else { + const size_t needed_bytes = max_num_frames * sizeof(stack[0]); + void* p = Allocate(needed_bytes); + if (p != nullptr) { // We got the space. + num_stack = max_num_frames; + stack = reinterpret_cast(p); + allocated_bytes = needed_bytes; + } + } + + size_t depth = absl::GetStackTrace(stack, num_stack, min_dropped_frames + 1); + for (size_t i = 0; i < depth; i++) { + if (symbolize_stacktrace) { + DumpPCAndSymbol(writer, writer_arg, stack[i], " "); + } else { + DumpPC(writer, writer_arg, stack[i], " "); + } + } + + auto hook = GetDebugStackTraceHook(); + if (hook != nullptr) { + (*hook)(stack, depth, writer, writer_arg); + } + + if (allocated_bytes != 0) Deallocate(stack, allocated_bytes); +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/internal/examine_stack.h b/src/absl/debugging/internal/examine_stack.h new file mode 100644 index 0000000..190af87 --- /dev/null +++ b/src/absl/debugging/internal/examine_stack.h @@ -0,0 +1,64 @@ +// +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +#ifndef ABSL_DEBUGGING_INTERNAL_EXAMINE_STACK_H_ +#define ABSL_DEBUGGING_INTERNAL_EXAMINE_STACK_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// Type of function used for printing in stack trace dumping, etc. +// We avoid closures to keep things simple. +typedef void OutputWriter(const char*, void*); + +// RegisterDebugStackTraceHook() allows to register a single routine +// `hook` that is called each time DumpStackTrace() is called. +// `hook` may be called from a signal handler. +typedef void (*SymbolizeUrlEmitter)(void* const stack[], int depth, + OutputWriter* writer, void* writer_arg); + +// Registration of SymbolizeUrlEmitter for use inside of a signal handler. +// This is inherently unsafe and must be signal safe code. +void RegisterDebugStackTraceHook(SymbolizeUrlEmitter hook); +SymbolizeUrlEmitter GetDebugStackTraceHook(); + +// Returns the program counter from signal context, or nullptr if +// unknown. `vuc` is a ucontext_t*. We use void* to avoid the use of +// ucontext_t on non-POSIX systems. +void* GetProgramCounter(void* const vuc); + +// Uses `writer` to dump the program counter, stack trace, and stack +// frame sizes. +void DumpPCAndFrameSizesAndStackTrace(void* const pc, void* const stack[], + int frame_sizes[], int depth, + int min_dropped_frames, + bool symbolize_stacktrace, + OutputWriter* writer, void* writer_arg); + +// Dump current stack trace omitting the topmost `min_dropped_frames` stack +// frames. +void DumpStackTrace(int min_dropped_frames, int max_num_frames, + bool symbolize_stacktrace, OutputWriter* writer, + void* writer_arg); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_EXAMINE_STACK_H_ diff --git a/src/absl/debugging/internal/stack_consumption.cc b/src/absl/debugging/internal/stack_consumption.cc new file mode 100644 index 0000000..5134864 --- /dev/null +++ b/src/absl/debugging/internal/stack_consumption.cc @@ -0,0 +1,185 @@ +// +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/debugging/internal/stack_consumption.h" + +#ifdef ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION + +#include +#include +#include + +#include + +#include "absl/base/attributes.h" +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +namespace { + +// This code requires that we know the direction in which the stack +// grows. It is commonly believed that this can be detected by putting +// a variable on the stack and then passing its address to a function +// that compares the address of this variable to the address of a +// variable on the function's own stack. However, this is unspecified +// behavior in C++: If two pointers p and q of the same type point to +// different objects that are not members of the same object or +// elements of the same array or to different functions, or if only +// one of them is null, the results of pq, p<=q, and p>=q are +// unspecified. Therefore, instead we hardcode the direction of the +// stack on platforms we know about. +#if defined(__i386__) || defined(__x86_64__) || defined(__ppc__) || \ + defined(__aarch64__) || defined(__riscv) +constexpr bool kStackGrowsDown = true; +#else +#error Need to define kStackGrowsDown +#endif + +// To measure the stack footprint of some code, we create a signal handler +// (for SIGUSR2 say) that exercises this code on an alternate stack. This +// alternate stack is initialized to some known pattern (0x55, 0x55, 0x55, +// ...). We then self-send this signal, and after the signal handler returns, +// look at the alternate stack buffer to see what portion has been touched. +// +// This trick gives us the the stack footprint of the signal handler. But the +// signal handler, even before the code for it is exercised, consumes some +// stack already. We however only want the stack usage of the code inside the +// signal handler. To measure this accurately, we install two signal handlers: +// one that does nothing and just returns, and the user-provided signal +// handler. The difference between the stack consumption of these two signals +// handlers should give us the stack foorprint of interest. + +void EmptySignalHandler(int) {} + +// This is arbitrary value, and could be increase further, at the cost of +// memset()ting it all to known sentinel value. +constexpr int kAlternateStackSize = 64 << 10; // 64KiB + +constexpr int kSafetyMargin = 32; +constexpr char kAlternateStackFillValue = 0x55; + +// These helper functions look at the alternate stack buffer, and figure +// out what portion of this buffer has been touched - this is the stack +// consumption of the signal handler running on this alternate stack. +// This function will return -1 if the alternate stack buffer has not been +// touched. It will abort the program if the buffer has overflowed or is about +// to overflow. +int GetStackConsumption(const void* const altstack) { + const char* begin; + int increment; + if (kStackGrowsDown) { + begin = reinterpret_cast(altstack); + increment = 1; + } else { + begin = reinterpret_cast(altstack) + kAlternateStackSize - 1; + increment = -1; + } + + for (int usage_count = kAlternateStackSize; usage_count > 0; --usage_count) { + if (*begin != kAlternateStackFillValue) { + ABSL_RAW_CHECK(usage_count <= kAlternateStackSize - kSafetyMargin, + "Buffer has overflowed or is about to overflow"); + return usage_count; + } + begin += increment; + } + + ABSL_RAW_LOG(FATAL, "Unreachable code"); + return -1; +} + +} // namespace + +int GetSignalHandlerStackConsumption(void (*signal_handler)(int)) { + // The alt-signal-stack cannot be heap allocated because there is a + // bug in glibc-2.2 where some signal handler setup code looks at the + // current stack pointer to figure out what thread is currently running. + // Therefore, the alternate stack must be allocated from the main stack + // itself. + void* altstack = mmap(nullptr, kAlternateStackSize, PROT_READ | PROT_WRITE, + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); + ABSL_RAW_CHECK(altstack != MAP_FAILED, "mmap() failed"); + + // Set up the alt-signal-stack (and save the older one). + stack_t sigstk; + memset(&sigstk, 0, sizeof(sigstk)); + sigstk.ss_sp = altstack; + sigstk.ss_size = kAlternateStackSize; + sigstk.ss_flags = 0; + stack_t old_sigstk; + memset(&old_sigstk, 0, sizeof(old_sigstk)); + ABSL_RAW_CHECK(sigaltstack(&sigstk, &old_sigstk) == 0, + "sigaltstack() failed"); + + // Set up SIGUSR1 and SIGUSR2 signal handlers (and save the older ones). + struct sigaction sa; + memset(&sa, 0, sizeof(sa)); + struct sigaction old_sa1, old_sa2; + sigemptyset(&sa.sa_mask); + sa.sa_flags = SA_ONSTACK; + + // SIGUSR1 maps to EmptySignalHandler. + sa.sa_handler = EmptySignalHandler; + ABSL_RAW_CHECK(sigaction(SIGUSR1, &sa, &old_sa1) == 0, "sigaction() failed"); + + // SIGUSR2 maps to signal_handler. + sa.sa_handler = signal_handler; + ABSL_RAW_CHECK(sigaction(SIGUSR2, &sa, &old_sa2) == 0, "sigaction() failed"); + + // Send SIGUSR1 signal and measure the stack consumption of the empty + // signal handler. + // The first signal might use more stack space. Run once and ignore the + // results to get that out of the way. + ABSL_RAW_CHECK(kill(getpid(), SIGUSR1) == 0, "kill() failed"); + + memset(altstack, kAlternateStackFillValue, kAlternateStackSize); + ABSL_RAW_CHECK(kill(getpid(), SIGUSR1) == 0, "kill() failed"); + int base_stack_consumption = GetStackConsumption(altstack); + + // Send SIGUSR2 signal and measure the stack consumption of signal_handler. + ABSL_RAW_CHECK(kill(getpid(), SIGUSR2) == 0, "kill() failed"); + int signal_handler_stack_consumption = GetStackConsumption(altstack); + + // Now restore the old alt-signal-stack and signal handlers. + if (old_sigstk.ss_sp == nullptr && old_sigstk.ss_size == 0 && + (old_sigstk.ss_flags & SS_DISABLE)) { + // https://git.musl-libc.org/cgit/musl/commit/src/signal/sigaltstack.c?id=7829f42a2c8944555439380498ab8b924d0f2070 + // The original stack has ss_size==0 and ss_flags==SS_DISABLE, but some + // versions of musl have a bug that rejects ss_size==0. Work around this by + // setting ss_size to MINSIGSTKSZ, which should be ignored by the kernel + // when SS_DISABLE is set. + old_sigstk.ss_size = MINSIGSTKSZ; + } + ABSL_RAW_CHECK(sigaltstack(&old_sigstk, nullptr) == 0, + "sigaltstack() failed"); + ABSL_RAW_CHECK(sigaction(SIGUSR1, &old_sa1, nullptr) == 0, + "sigaction() failed"); + ABSL_RAW_CHECK(sigaction(SIGUSR2, &old_sa2, nullptr) == 0, + "sigaction() failed"); + + ABSL_RAW_CHECK(munmap(altstack, kAlternateStackSize) == 0, "munmap() failed"); + if (signal_handler_stack_consumption != -1 && base_stack_consumption != -1) { + return signal_handler_stack_consumption - base_stack_consumption; + } + return -1; +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION diff --git a/src/absl/debugging/internal/stack_consumption.h b/src/absl/debugging/internal/stack_consumption.h new file mode 100644 index 0000000..f41b64c --- /dev/null +++ b/src/absl/debugging/internal/stack_consumption.h @@ -0,0 +1,50 @@ +// +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Helper function for measuring stack consumption of signal handlers. + +#ifndef ABSL_DEBUGGING_INTERNAL_STACK_CONSUMPTION_H_ +#define ABSL_DEBUGGING_INTERNAL_STACK_CONSUMPTION_H_ + +#include "absl/base/config.h" + +// The code in this module is not portable. +// Use this feature test macro to detect its availability. +#ifdef ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION +#error ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION cannot be set directly +#elif !defined(__APPLE__) && !defined(_WIN32) && \ + (defined(__i386__) || defined(__x86_64__) || defined(__ppc__) || \ + defined(__aarch64__) || defined(__riscv)) +#define ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION 1 + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// Returns the stack consumption in bytes for the code exercised by +// signal_handler. To measure stack consumption, signal_handler is registered +// as a signal handler, so the code that it exercises must be async-signal +// safe. The argument of signal_handler is an implementation detail of signal +// handlers and should ignored by the code for signal_handler. Use global +// variables to pass information between your test code and signal_handler. +int GetSignalHandlerStackConsumption(void (*signal_handler)(int)); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_INTERNAL_HAVE_DEBUGGING_STACK_CONSUMPTION + +#endif // ABSL_DEBUGGING_INTERNAL_STACK_CONSUMPTION_H_ diff --git a/src/absl/debugging/internal/stacktrace_aarch64-inl.inc b/src/absl/debugging/internal/stacktrace_aarch64-inl.inc new file mode 100644 index 0000000..4f9db9d --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_aarch64-inl.inc @@ -0,0 +1,204 @@ +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_AARCH64_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_AARCH64_INL_H_ + +// Generate stack tracer for aarch64 + +#if defined(__linux__) +#include +#include +#include +#endif + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/debugging/internal/address_is_readable.h" +#include "absl/debugging/internal/vdso_support.h" // a no-op on non-elf or non-glibc systems +#include "absl/debugging/stacktrace.h" + +static const uintptr_t kUnknownFrameSize = 0; + +#if defined(__linux__) +// Returns the address of the VDSO __kernel_rt_sigreturn function, if present. +static const unsigned char* GetKernelRtSigreturnAddress() { + constexpr uintptr_t kImpossibleAddress = 1; + ABSL_CONST_INIT static std::atomic memoized{kImpossibleAddress}; + uintptr_t address = memoized.load(std::memory_order_relaxed); + if (address != kImpossibleAddress) { + return reinterpret_cast(address); + } + + address = reinterpret_cast(nullptr); + +#ifdef ABSL_HAVE_VDSO_SUPPORT + absl::debugging_internal::VDSOSupport vdso; + if (vdso.IsPresent()) { + absl::debugging_internal::VDSOSupport::SymbolInfo symbol_info; + auto lookup = [&](int type) { + return vdso.LookupSymbol("__kernel_rt_sigreturn", "LINUX_2.6.39", type, + &symbol_info); + }; + if ((!lookup(STT_FUNC) && !lookup(STT_NOTYPE)) || + symbol_info.address == nullptr) { + // Unexpected: VDSO is present, yet the expected symbol is missing + // or null. + assert(false && "VDSO is present, but doesn't have expected symbol"); + } else { + if (reinterpret_cast(symbol_info.address) != + kImpossibleAddress) { + address = reinterpret_cast(symbol_info.address); + } else { + assert(false && "VDSO returned invalid address"); + } + } + } +#endif + + memoized.store(address, std::memory_order_relaxed); + return reinterpret_cast(address); +} +#endif // __linux__ + +// Compute the size of a stack frame in [low..high). We assume that +// low < high. Return size of kUnknownFrameSize. +template +static inline uintptr_t ComputeStackFrameSize(const T* low, + const T* high) { + const char* low_char_ptr = reinterpret_cast(low); + const char* high_char_ptr = reinterpret_cast(high); + return low < high ? high_char_ptr - low_char_ptr : kUnknownFrameSize; +} + +// Given a pointer to a stack frame, locate and return the calling +// stackframe, or return null if no stackframe can be found. Perform sanity +// checks (the strictness of which is controlled by the boolean parameter +// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static void **NextStackFrame(void **old_frame_pointer, const void *uc) { + void **new_frame_pointer = reinterpret_cast(*old_frame_pointer); + bool check_frame_size = true; + +#if defined(__linux__) + if (WITH_CONTEXT && uc != nullptr) { + // Check to see if next frame's return address is __kernel_rt_sigreturn. + if (old_frame_pointer[1] == GetKernelRtSigreturnAddress()) { + const ucontext_t *ucv = static_cast(uc); + // old_frame_pointer[0] is not suitable for unwinding, look at + // ucontext to discover frame pointer before signal. + void **const pre_signal_frame_pointer = + reinterpret_cast(ucv->uc_mcontext.regs[29]); + + // Check that alleged frame pointer is actually readable. This is to + // prevent "double fault" in case we hit the first fault due to e.g. + // stack corruption. + if (!absl::debugging_internal::AddressIsReadable( + pre_signal_frame_pointer)) + return nullptr; + + // Alleged frame pointer is readable, use it for further unwinding. + new_frame_pointer = pre_signal_frame_pointer; + + // Skip frame size check if we return from a signal. We may be using a + // an alternate stack for signals. + check_frame_size = false; + } + } +#endif + + // aarch64 ABI requires stack pointer to be 16-byte-aligned. + if ((reinterpret_cast(new_frame_pointer) & 15) != 0) + return nullptr; + + // Check frame size. In strict mode, we assume frames to be under + // 100,000 bytes. In non-strict mode, we relax the limit to 1MB. + if (check_frame_size) { + const uintptr_t max_size = STRICT_UNWINDING ? 100000 : 1000000; + const uintptr_t frame_size = + ComputeStackFrameSize(old_frame_pointer, new_frame_pointer); + if (frame_size == kUnknownFrameSize || frame_size > max_size) + return nullptr; + } + + return new_frame_pointer; +} + +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static int UnwindImpl(void** result, int* sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { +#ifdef __GNUC__ + void **frame_pointer = reinterpret_cast(__builtin_frame_address(0)); +#else +# error reading stack point not yet supported on this platform. +#endif + + skip_count++; // Skip the frame for this function. + int n = 0; + + // The frame pointer points to low address of a frame. The first 64-bit + // word of a frame points to the next frame up the call chain, which normally + // is just after the high address of the current frame. The second word of + // a frame contains return adress of to the caller. To find a pc value + // associated with the current frame, we need to go down a level in the call + // chain. So we remember return the address of the last frame seen. This + // does not work for the first stack frame, which belongs to UnwindImp() but + // we skip the frame for UnwindImp() anyway. + void* prev_return_address = nullptr; + + while (frame_pointer && n < max_depth) { + // The absl::GetStackFrames routine is called when we are in some + // informational context (the failure signal handler for example). + // Use the non-strict unwinding rules to produce a stack trace + // that is as complete as possible (even if it contains a few bogus + // entries in some rare cases). + void **next_frame_pointer = + NextStackFrame(frame_pointer, ucp); + + if (skip_count > 0) { + skip_count--; + } else { + result[n] = prev_return_address; + if (IS_STACK_FRAMES) { + sizes[n] = ComputeStackFrameSize(frame_pointer, next_frame_pointer); + } + n++; + } + prev_return_address = frame_pointer[1]; + frame_pointer = next_frame_pointer; + } + if (min_dropped_frames != nullptr) { + // Implementation detail: we clamp the max of frames we are willing to + // count, so as not to spend too much time in the loop below. + const int kMaxUnwind = 200; + int num_dropped_frames = 0; + for (int j = 0; frame_pointer != nullptr && j < kMaxUnwind; j++) { + if (skip_count > 0) { + skip_count--; + } else { + num_dropped_frames++; + } + frame_pointer = + NextStackFrame(frame_pointer, ucp); + } + *min_dropped_frames = num_dropped_frames; + } + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return true; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_AARCH64_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_arm-inl.inc b/src/absl/debugging/internal/stacktrace_arm-inl.inc new file mode 100644 index 0000000..102a2a1 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_arm-inl.inc @@ -0,0 +1,139 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// This is inspired by Craig Silverstein's PowerPC stacktrace code. + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_ARM_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_ARM_INL_H_ + +#include + +#include "absl/debugging/stacktrace.h" + +// WARNING: +// This only works if all your code is in either ARM or THUMB mode. With +// interworking, the frame pointer of the caller can either be in r11 (ARM +// mode) or r7 (THUMB mode). A callee only saves the frame pointer of its +// mode in a fixed location on its stack frame. If the caller is a different +// mode, there is no easy way to find the frame pointer. It can either be +// still in the designated register or saved on stack along with other callee +// saved registers. + +// Given a pointer to a stack frame, locate and return the calling +// stackframe, or return nullptr if no stackframe can be found. Perform sanity +// checks (the strictness of which is controlled by the boolean parameter +// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. +template +static void **NextStackFrame(void **old_sp) { + void **new_sp = (void**) old_sp[-1]; + + // Check that the transition from frame pointer old_sp to frame + // pointer new_sp isn't clearly bogus + if (STRICT_UNWINDING) { + // With the stack growing downwards, older stack frame must be + // at a greater address that the current one. + if (new_sp <= old_sp) return nullptr; + // Assume stack frames larger than 100,000 bytes are bogus. + if ((uintptr_t)new_sp - (uintptr_t)old_sp > 100000) return nullptr; + } else { + // In the non-strict mode, allow discontiguous stack frames. + // (alternate-signal-stacks for example). + if (new_sp == old_sp) return nullptr; + // And allow frames upto about 1MB. + if ((new_sp > old_sp) + && ((uintptr_t)new_sp - (uintptr_t)old_sp > 1000000)) return nullptr; + } + if ((uintptr_t)new_sp & (sizeof(void *) - 1)) return nullptr; + return new_sp; +} + +// This ensures that absl::GetStackTrace sets up the Link Register properly. +#ifdef __GNUC__ +void StacktraceArmDummyFunction() __attribute__((noinline)); +void StacktraceArmDummyFunction() { __asm__ volatile(""); } +#else +# error StacktraceArmDummyFunction() needs to be ported to this platform. +#endif + +template +static int UnwindImpl(void** result, int* sizes, int max_depth, int skip_count, + const void * /* ucp */, int *min_dropped_frames) { +#ifdef __GNUC__ + void **sp = reinterpret_cast(__builtin_frame_address(0)); +#else +# error reading stack point not yet supported on this platform. +#endif + + // On ARM, the return address is stored in the link register (r14). + // This is not saved on the stack frame of a leaf function. To + // simplify code that reads return addresses, we call a dummy + // function so that the return address of this function is also + // stored in the stack frame. This works at least for gcc. + StacktraceArmDummyFunction(); + + int n = 0; + while (sp && n < max_depth) { + // The absl::GetStackFrames routine is called when we are in some + // informational context (the failure signal handler for example). + // Use the non-strict unwinding rules to produce a stack trace + // that is as complete as possible (even if it contains a few bogus + // entries in some rare cases). + void **next_sp = NextStackFrame(sp); + + if (skip_count > 0) { + skip_count--; + } else { + result[n] = *sp; + + if (IS_STACK_FRAMES) { + if (next_sp > sp) { + sizes[n] = (uintptr_t)next_sp - (uintptr_t)sp; + } else { + // A frame-size of 0 is used to indicate unknown frame size. + sizes[n] = 0; + } + } + n++; + } + sp = next_sp; + } + if (min_dropped_frames != nullptr) { + // Implementation detail: we clamp the max of frames we are willing to + // count, so as not to spend too much time in the loop below. + const int kMaxUnwind = 200; + int num_dropped_frames = 0; + for (int j = 0; sp != nullptr && j < kMaxUnwind; j++) { + if (skip_count > 0) { + skip_count--; + } else { + num_dropped_frames++; + } + sp = NextStackFrame(sp); + } + *min_dropped_frames = num_dropped_frames; + } + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return false; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_ARM_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_config.h b/src/absl/debugging/internal/stacktrace_config.h new file mode 100644 index 0000000..3929b1b --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_config.h @@ -0,0 +1,88 @@ +/* + * Copyright 2017 The Abseil Authors. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * https://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + + * Defines ABSL_STACKTRACE_INL_HEADER to the *-inl.h containing + * actual unwinder implementation. + * This header is "private" to stacktrace.cc. + * DO NOT include it into any other files. +*/ +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_CONFIG_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_CONFIG_H_ + +#include "absl/base/config.h" + +#if defined(ABSL_STACKTRACE_INL_HEADER) +#error ABSL_STACKTRACE_INL_HEADER cannot be directly set + +#elif defined(_WIN32) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_win32-inl.inc" + +#elif defined(__APPLE__) +#ifdef ABSL_HAVE_THREAD_LOCAL +// Thread local support required for UnwindImpl. +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_generic-inl.inc" +#endif // defined(ABSL_HAVE_THREAD_LOCAL) + +// Emscripten stacktraces rely on JS. Do not use them in standalone mode. +#elif defined(__EMSCRIPTEN__) && !defined(STANDALONE_WASM) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_emscripten-inl.inc" + +#elif defined(__linux__) && !defined(__ANDROID__) + +#if defined(NO_FRAME_POINTER) && \ + (defined(__i386__) || defined(__x86_64__) || defined(__aarch64__)) +// Note: The libunwind-based implementation is not available to open-source +// users. +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_libunwind-inl.inc" +#define STACKTRACE_USES_LIBUNWIND 1 +#elif defined(NO_FRAME_POINTER) && defined(__has_include) +#if __has_include() +// Note: When using glibc this may require -funwind-tables to function properly. +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_generic-inl.inc" +#endif // __has_include() +#elif defined(__i386__) || defined(__x86_64__) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_x86-inl.inc" +#elif defined(__ppc__) || defined(__PPC__) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_powerpc-inl.inc" +#elif defined(__aarch64__) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_aarch64-inl.inc" +#elif defined(__riscv) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_riscv-inl.inc" +#elif defined(__has_include) +#if __has_include() +// Note: When using glibc this may require -funwind-tables to function properly. +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_generic-inl.inc" +#endif // __has_include() +#endif // defined(__has_include) + +#endif // defined(__linux__) && !defined(__ANDROID__) + +// Fallback to the empty implementation. +#if !defined(ABSL_STACKTRACE_INL_HEADER) +#define ABSL_STACKTRACE_INL_HEADER \ + "absl/debugging/internal/stacktrace_unimplemented-inl.inc" +#endif + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_CONFIG_H_ diff --git a/src/absl/debugging/internal/stacktrace_emscripten-inl.inc b/src/absl/debugging/internal/stacktrace_emscripten-inl.inc new file mode 100644 index 0000000..0f44451 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_emscripten-inl.inc @@ -0,0 +1,110 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Portable implementation - just use glibc +// +// Note: The glibc implementation may cause a call to malloc. +// This can cause a deadlock in HeapProfiler. + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_EMSCRIPTEN_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_EMSCRIPTEN_INL_H_ + +#include + +#include +#include + +#include "absl/base/attributes.h" +#include "absl/debugging/stacktrace.h" + +extern "C" { +uintptr_t emscripten_stack_snapshot(); +uint32_t emscripten_stack_unwind_buffer(uintptr_t pc, void *buffer, + uint32_t depth); +} + +// Sometimes, we can try to get a stack trace from within a stack +// trace, which can cause a self-deadlock. +// Protect against such reentrant call by failing to get a stack trace. +// +// We use __thread here because the code here is extremely low level -- it is +// called while collecting stack traces from within malloc and mmap, and thus +// can not call anything which might call malloc or mmap itself. +static __thread int recursive = 0; + +// The stack trace function might be invoked very early in the program's +// execution (e.g. from the very first malloc). +// As such, we suppress usage of backtrace during this early stage of execution. +static std::atomic disable_stacktraces(true); // Disabled until healthy. +// Waiting until static initializers run seems to be late enough. +// This file is included into stacktrace.cc so this will only run once. +ABSL_ATTRIBUTE_UNUSED static int stacktraces_enabler = []() { + // Check if we can even create stacktraces. If not, bail early and leave + // disable_stacktraces set as-is. + // clang-format off + if (!EM_ASM_INT({ return (typeof wasmOffsetConverter !== 'undefined'); })) { + return 0; + } + // clang-format on + disable_stacktraces.store(false, std::memory_order_relaxed); + return 0; +}(); + +template +static int UnwindImpl(void **result, int *sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { + if (recursive || disable_stacktraces.load(std::memory_order_relaxed)) { + return 0; + } + ++recursive; + + static_cast(ucp); // Unused. + constexpr int kStackLength = 64; + void *stack[kStackLength]; + + int size; + uintptr_t pc = emscripten_stack_snapshot(); + size = emscripten_stack_unwind_buffer(pc, stack, kStackLength); + + int result_count = size - skip_count; + if (result_count < 0) result_count = 0; + if (result_count > max_depth) result_count = max_depth; + for (int i = 0; i < result_count; i++) result[i] = stack[i + skip_count]; + + if (IS_STACK_FRAMES) { + // No implementation for finding out the stack frame sizes yet. + memset(sizes, 0, sizeof(*sizes) * result_count); + } + if (min_dropped_frames != nullptr) { + if (size - skip_count - max_depth > 0) { + *min_dropped_frames = size - skip_count - max_depth; + } else { + *min_dropped_frames = 0; + } + } + + --recursive; + + return result_count; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { return true; } +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_EMSCRIPTEN_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_generic-inl.inc b/src/absl/debugging/internal/stacktrace_generic-inl.inc new file mode 100644 index 0000000..b2792a1 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_generic-inl.inc @@ -0,0 +1,108 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Portable implementation - just use glibc +// +// Note: The glibc implementation may cause a call to malloc. +// This can cause a deadlock in HeapProfiler. + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_GENERIC_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_GENERIC_INL_H_ + +#include +#include +#include + +#include "absl/debugging/stacktrace.h" +#include "absl/base/attributes.h" + +// Sometimes, we can try to get a stack trace from within a stack +// trace, because we don't block signals inside this code (which would be too +// expensive: the two extra system calls per stack trace do matter here). +// That can cause a self-deadlock. +// Protect against such reentrant call by failing to get a stack trace. +// +// We use __thread here because the code here is extremely low level -- it is +// called while collecting stack traces from within malloc and mmap, and thus +// can not call anything which might call malloc or mmap itself. +static __thread int recursive = 0; + +// The stack trace function might be invoked very early in the program's +// execution (e.g. from the very first malloc if using tcmalloc). Also, the +// glibc implementation itself will trigger malloc the first time it is called. +// As such, we suppress usage of backtrace during this early stage of execution. +static std::atomic disable_stacktraces(true); // Disabled until healthy. +// Waiting until static initializers run seems to be late enough. +// This file is included into stacktrace.cc so this will only run once. +ABSL_ATTRIBUTE_UNUSED static int stacktraces_enabler = []() { + void* unused_stack[1]; + // Force the first backtrace to happen early to get the one-time shared lib + // loading (allocation) out of the way. After the first call it is much safer + // to use backtrace from a signal handler if we crash somewhere later. + backtrace(unused_stack, 1); + disable_stacktraces.store(false, std::memory_order_relaxed); + return 0; +}(); + +template +static int UnwindImpl(void** result, int* sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { + if (recursive || disable_stacktraces.load(std::memory_order_relaxed)) { + return 0; + } + ++recursive; + + static_cast(ucp); // Unused. + static const int kStackLength = 64; + void * stack[kStackLength]; + int size; + + size = backtrace(stack, kStackLength); + skip_count++; // we want to skip the current frame as well + int result_count = size - skip_count; + if (result_count < 0) + result_count = 0; + if (result_count > max_depth) + result_count = max_depth; + for (int i = 0; i < result_count; i++) + result[i] = stack[i + skip_count]; + + if (IS_STACK_FRAMES) { + // No implementation for finding out the stack frame sizes yet. + memset(sizes, 0, sizeof(*sizes) * result_count); + } + if (min_dropped_frames != nullptr) { + if (size - skip_count - max_depth > 0) { + *min_dropped_frames = size - skip_count - max_depth; + } else { + *min_dropped_frames = 0; + } + } + + --recursive; + + return result_count; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return true; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_GENERIC_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_powerpc-inl.inc b/src/absl/debugging/internal/stacktrace_powerpc-inl.inc new file mode 100644 index 0000000..085cef6 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_powerpc-inl.inc @@ -0,0 +1,258 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Produce stack trace. I'm guessing (hoping!) the code is much like +// for x86. For apple machines, at least, it seems to be; see +// https://developer.apple.com/documentation/mac/runtimehtml/RTArch-59.html +// https://www.linux-foundation.org/spec/ELF/ppc64/PPC-elf64abi-1.9.html#STACK +// Linux has similar code: http://patchwork.ozlabs.org/linuxppc/patch?id=8882 + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_POWERPC_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_POWERPC_INL_H_ + +#if defined(__linux__) +#include // for PT_NIP. +#include // for ucontext_t +#endif + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/base/optimization.h" +#include "absl/base/port.h" +#include "absl/debugging/stacktrace.h" +#include "absl/debugging/internal/address_is_readable.h" +#include "absl/debugging/internal/vdso_support.h" // a no-op on non-elf or non-glibc systems + +// Given a stack pointer, return the saved link register value. +// Note that this is the link register for a callee. +static inline void *StacktracePowerPCGetLR(void **sp) { + // PowerPC has 3 main ABIs, which say where in the stack the + // Link Register is. For DARWIN and AIX (used by apple and + // linux ppc64), it's in sp[2]. For SYSV (used by linux ppc), + // it's in sp[1]. +#if defined(_CALL_AIX) || defined(_CALL_DARWIN) + return *(sp+2); +#elif defined(_CALL_SYSV) + return *(sp+1); +#elif defined(__APPLE__) || defined(__FreeBSD__) || \ + (defined(__linux__) && defined(__PPC64__)) + // This check is in case the compiler doesn't define _CALL_AIX/etc. + return *(sp+2); +#elif defined(__linux) + // This check is in case the compiler doesn't define _CALL_SYSV. + return *(sp+1); +#else +#error Need to specify the PPC ABI for your archiecture. +#endif +} + +// Given a pointer to a stack frame, locate and return the calling +// stackframe, or return null if no stackframe can be found. Perform sanity +// checks (the strictness of which is controlled by the boolean parameter +// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static void **NextStackFrame(void **old_sp, const void *uc) { + void **new_sp = (void **) *old_sp; + enum { kStackAlignment = 16 }; + + // Check that the transition from frame pointer old_sp to frame + // pointer new_sp isn't clearly bogus + if (STRICT_UNWINDING) { + // With the stack growing downwards, older stack frame must be + // at a greater address that the current one. + if (new_sp <= old_sp) return nullptr; + // Assume stack frames larger than 100,000 bytes are bogus. + if ((uintptr_t)new_sp - (uintptr_t)old_sp > 100000) return nullptr; + } else { + // In the non-strict mode, allow discontiguous stack frames. + // (alternate-signal-stacks for example). + if (new_sp == old_sp) return nullptr; + // And allow frames upto about 1MB. + if ((new_sp > old_sp) + && ((uintptr_t)new_sp - (uintptr_t)old_sp > 1000000)) return nullptr; + } + if ((uintptr_t)new_sp % kStackAlignment != 0) return nullptr; + +#if defined(__linux__) + enum StackTraceKernelSymbolStatus { + kNotInitialized = 0, kAddressValid, kAddressInvalid }; + + if (IS_WITH_CONTEXT && uc != nullptr) { + static StackTraceKernelSymbolStatus kernel_symbol_status = + kNotInitialized; // Sentinel: not computed yet. + // Initialize with sentinel value: __kernel_rt_sigtramp_rt64 can not + // possibly be there. + static const unsigned char *kernel_sigtramp_rt64_address = nullptr; + if (kernel_symbol_status == kNotInitialized) { + absl::debugging_internal::VDSOSupport vdso; + if (vdso.IsPresent()) { + absl::debugging_internal::VDSOSupport::SymbolInfo + sigtramp_rt64_symbol_info; + if (!vdso.LookupSymbol( + "__kernel_sigtramp_rt64", "LINUX_2.6.15", + absl::debugging_internal::VDSOSupport::kVDSOSymbolType, + &sigtramp_rt64_symbol_info) || + sigtramp_rt64_symbol_info.address == nullptr) { + // Unexpected: VDSO is present, yet the expected symbol is missing + // or null. + assert(false && "VDSO is present, but doesn't have expected symbol"); + kernel_symbol_status = kAddressInvalid; + } else { + kernel_sigtramp_rt64_address = + reinterpret_cast( + sigtramp_rt64_symbol_info.address); + kernel_symbol_status = kAddressValid; + } + } else { + kernel_symbol_status = kAddressInvalid; + } + } + + if (new_sp != nullptr && + kernel_symbol_status == kAddressValid && + StacktracePowerPCGetLR(new_sp) == kernel_sigtramp_rt64_address) { + const ucontext_t* signal_context = + reinterpret_cast(uc); + void **const sp_before_signal = +#if defined(__PPC64__) + reinterpret_cast(signal_context->uc_mcontext.gp_regs[PT_R1]); +#else + reinterpret_cast( + signal_context->uc_mcontext.uc_regs->gregs[PT_R1]); +#endif + // Check that alleged sp before signal is nonnull and is reasonably + // aligned. + if (sp_before_signal != nullptr && + ((uintptr_t)sp_before_signal % kStackAlignment) == 0) { + // Check that alleged stack pointer is actually readable. This is to + // prevent a "double fault" in case we hit the first fault due to e.g. + // a stack corruption. + if (absl::debugging_internal::AddressIsReadable(sp_before_signal)) { + // Alleged stack pointer is readable, use it for further unwinding. + new_sp = sp_before_signal; + } + } + } + } +#endif + + return new_sp; +} + +// This ensures that absl::GetStackTrace sets up the Link Register properly. +ABSL_ATTRIBUTE_NOINLINE static void AbslStacktracePowerPCDummyFunction() { + ABSL_BLOCK_TAIL_CALL_OPTIMIZATION(); +} + +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static int UnwindImpl(void** result, int* sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { + void **sp; + // Apple macOS uses an old version of gnu as -- both Darwin 7.9.0 (Panther) + // and Darwin 8.8.1 (Tiger) use as 1.38. This means we have to use a + // different asm syntax. I don't know quite the best way to discriminate + // systems using the old as from the new one; I've gone with __APPLE__. +#ifdef __APPLE__ + __asm__ volatile ("mr %0,r1" : "=r" (sp)); +#else + __asm__ volatile ("mr %0,1" : "=r" (sp)); +#endif + + // On PowerPC, the "Link Register" or "Link Record" (LR), is a stack + // entry that holds the return address of the subroutine call (what + // instruction we run after our function finishes). This is the + // same as the stack-pointer of our parent routine, which is what we + // want here. While the compiler will always(?) set up LR for + // subroutine calls, it may not for leaf functions (such as this one). + // This routine forces the compiler (at least gcc) to push it anyway. + AbslStacktracePowerPCDummyFunction(); + + // The LR save area is used by the callee, so the top entry is bogus. + skip_count++; + + int n = 0; + + // Unlike ABIs of X86 and ARM, PowerPC ABIs say that return address (in + // the link register) of a function call is stored in the caller's stack + // frame instead of the callee's. When we look for the return address + // associated with a stack frame, we need to make sure that there is a + // caller frame before it. So we call NextStackFrame before entering the + // loop below and check next_sp instead of sp for loop termination. + // The outermost frame is set up by runtimes and it does not have a + // caller frame, so it is skipped. + + // The absl::GetStackFrames routine is called when we are in some + // informational context (the failure signal handler for example). + // Use the non-strict unwinding rules to produce a stack trace + // that is as complete as possible (even if it contains a few + // bogus entries in some rare cases). + void **next_sp = NextStackFrame(sp, ucp); + + while (next_sp && n < max_depth) { + if (skip_count > 0) { + skip_count--; + } else { + result[n] = StacktracePowerPCGetLR(sp); + if (IS_STACK_FRAMES) { + if (next_sp > sp) { + sizes[n] = (uintptr_t)next_sp - (uintptr_t)sp; + } else { + // A frame-size of 0 is used to indicate unknown frame size. + sizes[n] = 0; + } + } + n++; + } + + sp = next_sp; + next_sp = NextStackFrame(sp, ucp); + } + + if (min_dropped_frames != nullptr) { + // Implementation detail: we clamp the max of frames we are willing to + // count, so as not to spend too much time in the loop below. + const int kMaxUnwind = 1000; + int num_dropped_frames = 0; + for (int j = 0; next_sp != nullptr && j < kMaxUnwind; j++) { + if (skip_count > 0) { + skip_count--; + } else { + num_dropped_frames++; + } + next_sp = NextStackFrame(next_sp, ucp); + } + *min_dropped_frames = num_dropped_frames; + } + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return true; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_POWERPC_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_riscv-inl.inc b/src/absl/debugging/internal/stacktrace_riscv-inl.inc new file mode 100644 index 0000000..7123b71 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_riscv-inl.inc @@ -0,0 +1,236 @@ +// Copyright 2021 The Abseil Authors +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_RISCV_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_RISCV_INL_H_ + +// Generate stack trace for riscv + +#include + +#include "absl/base/config.h" +#if defined(__linux__) +#include +#include +#include +#endif + +#include +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/debugging/internal/address_is_readable.h" +#include "absl/debugging/internal/vdso_support.h" +#include "absl/debugging/stacktrace.h" + +static const uintptr_t kUnknownFrameSize = 0; + +#if defined(__linux__) +// Returns the address of the VDSO __kernel_rt_sigreturn function, if present. +static const unsigned char *GetKernelRtSigreturnAddress() { + constexpr uintptr_t kImpossibleAddress = 0; + ABSL_CONST_INIT static std::atomic memoized(kImpossibleAddress); + uintptr_t address = memoized.load(std::memory_order_relaxed); + if (address != kImpossibleAddress) { + return reinterpret_cast(address); + } + + address = reinterpret_cast(nullptr); + +#if ABSL_HAVE_VDSO_SUPPORT + absl::debugging_internal::VDSOSupport vdso; + if (vdso.IsPresent()) { + absl::debugging_internal::VDSOSupport::SymbolInfo symbol_info; + // Symbol versioning pulled from arch/riscv/kernel/vdso/vdso.lds at v5.10. + auto lookup = [&](int type) { + return vdso.LookupSymbol("__vdso_rt_sigreturn", "LINUX_4.15", type, + &symbol_info); + }; + if ((!lookup(STT_FUNC) && !lookup(STT_NOTYPE)) || + symbol_info.address == nullptr) { + // Unexpected: VDSO is present, yet the expected symbol is missing or + // null. + assert(false && "VDSO is present, but doesn't have expected symbol"); + } else { + if (reinterpret_cast(symbol_info.address) != + kImpossibleAddress) { + address = reinterpret_cast(symbol_info.address); + } else { + assert(false && "VDSO returned invalid address"); + } + } + } +#endif + + memoized.store(address, std::memory_order_relaxed); + return reinterpret_cast(address); +} +#endif // __linux__ + +// Compute the size of a stack frame in [low..high). We assume that low < high. +// Return size of kUnknownFrameSize. +template +static inline uintptr_t ComputeStackFrameSize(const T *low, const T *high) { + const char *low_char_ptr = reinterpret_cast(low); + const char *high_char_ptr = reinterpret_cast(high); + return low < high ? high_char_ptr - low_char_ptr : kUnknownFrameSize; +} + +// Given a pointer to a stack frame, locate and return the calling stackframe, +// or return null if no stackframe can be found. Perform sanity checks (the +// strictness of which is controlled by the boolean parameter +// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static void ** NextStackFrame(void **old_frame_pointer, const void *uc) { + // . + // . + // . + // +-> +----------------+ + // | | return address | + // | | previous fp | + // | | ... | + // | +----------------+ <-+ + // | | return address | | + // +---|- previous fp | | + // | ... | | + // $fp ->|----------------+ | + // | return address | | + // | previous fp -|---+ + // $sp ->| ... | + // +----------------+ + void **new_frame_pointer = reinterpret_cast(old_frame_pointer[-2]); + bool check_frame_size = true; + +#if defined(__linux__) + if (WITH_CONTEXT && uc != nullptr) { + // Check to see if next frame's return address is __kernel_rt_sigreturn. + if (old_frame_pointer[-1] == GetKernelRtSigreturnAddress()) { + const ucontext_t *ucv = static_cast(uc); + // old_frame_pointer is not suitable for unwinding, look at ucontext to + // discover frame pointer before signal. + // + // RISCV ELF psABI has the frame pointer at x8/fp/s0. + // -- RISCV psABI Table 18.2 + void **const pre_signal_frame_pointer = + reinterpret_cast(ucv->uc_mcontext.__gregs[8]); + + // Check the alleged frame pointer is actually readable. This is to + // prevent "double fault" in case we hit the first fault due to stack + // corruption. + if (!absl::debugging_internal::AddressIsReadable( + pre_signal_frame_pointer)) + return nullptr; + + // Alleged frame pointer is readable, use it for further unwinding. + new_frame_pointer = pre_signal_frame_pointer; + + // Skip frame size check if we return from a signal. We may be using an + // alterate stack for signals. + check_frame_size = false; + } + } +#endif + + // The RISCV ELF psABI mandates that the stack pointer is always 16-byte + // aligned. + // FIXME(abdulras) this doesn't hold for ILP32E which only mandates a 4-byte + // alignment. + if ((reinterpret_cast(new_frame_pointer) & 15) != 0) + return nullptr; + + // Check frame size. In strict mode, we assume frames to be under 100,000 + // bytes. In non-strict mode, we relax the limit to 1MB. + if (check_frame_size) { + const uintptr_t max_size = STRICT_UNWINDING ? 100000 : 1000000; + const uintptr_t frame_size = + ComputeStackFrameSize(old_frame_pointer, new_frame_pointer); + if (frame_size == kUnknownFrameSize || frame_size > max_size) + return nullptr; + } + + return new_frame_pointer; +} + +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static int UnwindImpl(void **result, int *sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { + // The `frame_pointer` that is computed here points to the top of the frame. + // The two words preceding the address are the return address and the previous + // frame pointer. +#if defined(__GNUC__) + void **frame_pointer = reinterpret_cast(__builtin_frame_address(0)); +#else +#error reading stack pointer not yet supported on this platform +#endif + + int n = 0; + void *return_address = nullptr; + while (frame_pointer && n < max_depth) { + return_address = frame_pointer[-1]; + + // The absl::GetStackFrames routine is called when we are in some + // informational context (the failure signal handler for example). Use the + // non-strict unwinding rules to produce a stack trace that is as complete + // as possible (even if it contains a few bogus entries in some rare cases). + void **next_frame_pointer = + NextStackFrame(frame_pointer, ucp); + + if (skip_count > 0) { + skip_count--; + } else { + result[n] = return_address; + if (IS_STACK_FRAMES) { + sizes[n] = ComputeStackFrameSize(frame_pointer, next_frame_pointer); + } + n++; + } + + frame_pointer = next_frame_pointer; + } + + if (min_dropped_frames != nullptr) { + // Implementation detail: we clamp the max of frames we are willing to + // count, so as not to spend too much time in the loop below. + const int kMaxUnwind = 200; + int num_dropped_frames = 0; + for (int j = 0; frame_pointer != nullptr && j < kMaxUnwind; j++) { + if (skip_count > 0) { + skip_count--; + } else { + num_dropped_frames++; + } + frame_pointer = + NextStackFrame(frame_pointer, ucp); + } + *min_dropped_frames = num_dropped_frames; + } + + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { return true; } +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif diff --git a/src/absl/debugging/internal/stacktrace_unimplemented-inl.inc b/src/absl/debugging/internal/stacktrace_unimplemented-inl.inc new file mode 100644 index 0000000..5b8fb19 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_unimplemented-inl.inc @@ -0,0 +1,24 @@ +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_UNIMPLEMENTED_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_UNIMPLEMENTED_INL_H_ + +template +static int UnwindImpl(void** /* result */, int* /* sizes */, + int /* max_depth */, int /* skip_count */, + const void* /* ucp */, int *min_dropped_frames) { + if (min_dropped_frames != nullptr) { + *min_dropped_frames = 0; + } + return 0; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return false; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_UNIMPLEMENTED_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_win32-inl.inc b/src/absl/debugging/internal/stacktrace_win32-inl.inc new file mode 100644 index 0000000..1c666c8 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_win32-inl.inc @@ -0,0 +1,93 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Produces a stack trace for Windows. Normally, one could use +// stacktrace_x86-inl.h or stacktrace_x86_64-inl.h -- and indeed, that +// should work for binaries compiled using MSVC in "debug" mode. +// However, in "release" mode, Windows uses frame-pointer +// optimization, which makes getting a stack trace very difficult. +// +// There are several approaches one can take. One is to use Windows +// intrinsics like StackWalk64. These can work, but have restrictions +// on how successful they can be. Another attempt is to write a +// version of stacktrace_x86-inl.h that has heuristic support for +// dealing with FPO, similar to what WinDbg does (see +// http://www.nynaeve.net/?p=97). There are (non-working) examples of +// these approaches, complete with TODOs, in stacktrace_win32-inl.h#1 +// +// The solution we've ended up doing is to call the undocumented +// windows function RtlCaptureStackBackTrace, which probably doesn't +// work with FPO but at least is fast, and doesn't require a symbol +// server. +// +// This code is inspired by a patch from David Vitek: +// https://code.google.com/p/google-perftools/issues/detail?id=83 + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_WIN32_INL_H_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_WIN32_INL_H_ + +#include // for GetProcAddress and GetModuleHandle +#include + +typedef USHORT NTAPI RtlCaptureStackBackTrace_Function( + IN ULONG frames_to_skip, + IN ULONG frames_to_capture, + OUT PVOID *backtrace, + OUT PULONG backtrace_hash); + +// It is not possible to load RtlCaptureStackBackTrace at static init time in +// UWP. CaptureStackBackTrace is the public version of RtlCaptureStackBackTrace +#if WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_APP) && \ + !WINAPI_FAMILY_PARTITION(WINAPI_PARTITION_DESKTOP) +static RtlCaptureStackBackTrace_Function* const RtlCaptureStackBackTrace_fn = + &::CaptureStackBackTrace; +#else +// Load the function we need at static init time, where we don't have +// to worry about someone else holding the loader's lock. +static RtlCaptureStackBackTrace_Function* const RtlCaptureStackBackTrace_fn = + (RtlCaptureStackBackTrace_Function*)GetProcAddress( + GetModuleHandleA("ntdll.dll"), "RtlCaptureStackBackTrace"); +#endif // WINAPI_PARTITION_APP && !WINAPI_PARTITION_DESKTOP + +template +static int UnwindImpl(void** result, int* sizes, int max_depth, int skip_count, + const void*, int* min_dropped_frames) { + int n = 0; + if (!RtlCaptureStackBackTrace_fn) { + // can't find a stacktrace with no function to call + } else { + n = (int)RtlCaptureStackBackTrace_fn(skip_count + 2, max_depth, result, 0); + } + if (IS_STACK_FRAMES) { + // No implementation for finding out the stack frame sizes yet. + memset(sizes, 0, sizeof(*sizes) * n); + } + if (min_dropped_frames != nullptr) { + // Not implemented. + *min_dropped_frames = 0; + } + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return false; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_WIN32_INL_H_ diff --git a/src/absl/debugging/internal/stacktrace_x86-inl.inc b/src/absl/debugging/internal/stacktrace_x86-inl.inc new file mode 100644 index 0000000..1b5d823 --- /dev/null +++ b/src/absl/debugging/internal/stacktrace_x86-inl.inc @@ -0,0 +1,369 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Produce stack trace + +#ifndef ABSL_DEBUGGING_INTERNAL_STACKTRACE_X86_INL_INC_ +#define ABSL_DEBUGGING_INTERNAL_STACKTRACE_X86_INL_INC_ + +#if defined(__linux__) && (defined(__i386__) || defined(__x86_64__)) +#include // for ucontext_t +#endif + +#if !defined(_WIN32) +#include +#endif + +#include +#include +#include + +#include "absl/base/macros.h" +#include "absl/base/port.h" +#include "absl/debugging/internal/address_is_readable.h" +#include "absl/debugging/internal/vdso_support.h" // a no-op on non-elf or non-glibc systems +#include "absl/debugging/stacktrace.h" + +#include "absl/base/internal/raw_logging.h" + +using absl::debugging_internal::AddressIsReadable; + +#if defined(__linux__) && defined(__i386__) +// Count "push %reg" instructions in VDSO __kernel_vsyscall(), +// preceeding "syscall" or "sysenter". +// If __kernel_vsyscall uses frame pointer, answer 0. +// +// kMaxBytes tells how many instruction bytes of __kernel_vsyscall +// to analyze before giving up. Up to kMaxBytes+1 bytes of +// instructions could be accessed. +// +// Here are known __kernel_vsyscall instruction sequences: +// +// SYSENTER (linux-2.6.26/arch/x86/vdso/vdso32/sysenter.S). +// Used on Intel. +// 0xffffe400 <__kernel_vsyscall+0>: push %ecx +// 0xffffe401 <__kernel_vsyscall+1>: push %edx +// 0xffffe402 <__kernel_vsyscall+2>: push %ebp +// 0xffffe403 <__kernel_vsyscall+3>: mov %esp,%ebp +// 0xffffe405 <__kernel_vsyscall+5>: sysenter +// +// SYSCALL (see linux-2.6.26/arch/x86/vdso/vdso32/syscall.S). +// Used on AMD. +// 0xffffe400 <__kernel_vsyscall+0>: push %ebp +// 0xffffe401 <__kernel_vsyscall+1>: mov %ecx,%ebp +// 0xffffe403 <__kernel_vsyscall+3>: syscall +// + +// The sequence below isn't actually expected in Google fleet, +// here only for completeness. Remove this comment from OSS release. + +// i386 (see linux-2.6.26/arch/x86/vdso/vdso32/int80.S) +// 0xffffe400 <__kernel_vsyscall+0>: int $0x80 +// 0xffffe401 <__kernel_vsyscall+1>: ret +// +static const int kMaxBytes = 10; + +// We use assert()s instead of DCHECK()s -- this is too low level +// for DCHECK(). + +static int CountPushInstructions(const unsigned char *const addr) { + int result = 0; + for (int i = 0; i < kMaxBytes; ++i) { + if (addr[i] == 0x89) { + // "mov reg,reg" + if (addr[i + 1] == 0xE5) { + // Found "mov %esp,%ebp". + return 0; + } + ++i; // Skip register encoding byte. + } else if (addr[i] == 0x0F && + (addr[i + 1] == 0x34 || addr[i + 1] == 0x05)) { + // Found "sysenter" or "syscall". + return result; + } else if ((addr[i] & 0xF0) == 0x50) { + // Found "push %reg". + ++result; + } else if (addr[i] == 0xCD && addr[i + 1] == 0x80) { + // Found "int $0x80" + assert(result == 0); + return 0; + } else { + // Unexpected instruction. + assert(false && "unexpected instruction in __kernel_vsyscall"); + return 0; + } + } + // Unexpected: didn't find SYSENTER or SYSCALL in + // [__kernel_vsyscall, __kernel_vsyscall + kMaxBytes) interval. + assert(false && "did not find SYSENTER or SYSCALL in __kernel_vsyscall"); + return 0; +} +#endif + +// Assume stack frames larger than 100,000 bytes are bogus. +static const int kMaxFrameBytes = 100000; + +// Returns the stack frame pointer from signal context, 0 if unknown. +// vuc is a ucontext_t *. We use void* to avoid the use +// of ucontext_t on non-POSIX systems. +static uintptr_t GetFP(const void *vuc) { +#if !defined(__linux__) + static_cast(vuc); // Avoid an unused argument compiler warning. +#else + if (vuc != nullptr) { + auto *uc = reinterpret_cast(vuc); +#if defined(__i386__) + const auto bp = uc->uc_mcontext.gregs[REG_EBP]; + const auto sp = uc->uc_mcontext.gregs[REG_ESP]; +#elif defined(__x86_64__) + const auto bp = uc->uc_mcontext.gregs[REG_RBP]; + const auto sp = uc->uc_mcontext.gregs[REG_RSP]; +#else + const uintptr_t bp = 0; + const uintptr_t sp = 0; +#endif + // Sanity-check that the base pointer is valid. It's possible that some + // code in the process is compiled with --copt=-fomit-frame-pointer or + // --copt=-momit-leaf-frame-pointer. + // + // TODO(bcmills): -momit-leaf-frame-pointer is currently the default + // behavior when building with clang. Talk to the C++ toolchain team about + // fixing that. + if (bp >= sp && bp - sp <= kMaxFrameBytes) return bp; + + // If bp isn't a plausible frame pointer, return the stack pointer instead. + // If we're lucky, it points to the start of a stack frame; otherwise, we'll + // get one frame of garbage in the stack trace and fail the sanity check on + // the next iteration. + return sp; + } +#endif + return 0; +} + +// Given a pointer to a stack frame, locate and return the calling +// stackframe, or return null if no stackframe can be found. Perform sanity +// checks (the strictness of which is controlled by the boolean parameter +// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +static void **NextStackFrame(void **old_fp, const void *uc, + size_t stack_low, size_t stack_high) { + void **new_fp = (void **)*old_fp; + +#if defined(__linux__) && defined(__i386__) + if (WITH_CONTEXT && uc != nullptr) { + // How many "push %reg" instructions are there at __kernel_vsyscall? + // This is constant for a given kernel and processor, so compute + // it only once. + static int num_push_instructions = -1; // Sentinel: not computed yet. + // Initialize with sentinel value: __kernel_rt_sigreturn can not possibly + // be there. + static const unsigned char *kernel_rt_sigreturn_address = nullptr; + static const unsigned char *kernel_vsyscall_address = nullptr; + if (num_push_instructions == -1) { +#ifdef ABSL_HAVE_VDSO_SUPPORT + absl::debugging_internal::VDSOSupport vdso; + if (vdso.IsPresent()) { + absl::debugging_internal::VDSOSupport::SymbolInfo + rt_sigreturn_symbol_info; + absl::debugging_internal::VDSOSupport::SymbolInfo vsyscall_symbol_info; + if (!vdso.LookupSymbol("__kernel_rt_sigreturn", "LINUX_2.5", STT_FUNC, + &rt_sigreturn_symbol_info) || + !vdso.LookupSymbol("__kernel_vsyscall", "LINUX_2.5", STT_FUNC, + &vsyscall_symbol_info) || + rt_sigreturn_symbol_info.address == nullptr || + vsyscall_symbol_info.address == nullptr) { + // Unexpected: 32-bit VDSO is present, yet one of the expected + // symbols is missing or null. + assert(false && "VDSO is present, but doesn't have expected symbols"); + num_push_instructions = 0; + } else { + kernel_rt_sigreturn_address = + reinterpret_cast( + rt_sigreturn_symbol_info.address); + kernel_vsyscall_address = + reinterpret_cast( + vsyscall_symbol_info.address); + num_push_instructions = + CountPushInstructions(kernel_vsyscall_address); + } + } else { + num_push_instructions = 0; + } +#else // ABSL_HAVE_VDSO_SUPPORT + num_push_instructions = 0; +#endif // ABSL_HAVE_VDSO_SUPPORT + } + if (num_push_instructions != 0 && kernel_rt_sigreturn_address != nullptr && + old_fp[1] == kernel_rt_sigreturn_address) { + const ucontext_t *ucv = static_cast(uc); + // This kernel does not use frame pointer in its VDSO code, + // and so %ebp is not suitable for unwinding. + void **const reg_ebp = + reinterpret_cast(ucv->uc_mcontext.gregs[REG_EBP]); + const unsigned char *const reg_eip = + reinterpret_cast(ucv->uc_mcontext.gregs[REG_EIP]); + if (new_fp == reg_ebp && kernel_vsyscall_address <= reg_eip && + reg_eip - kernel_vsyscall_address < kMaxBytes) { + // We "stepped up" to __kernel_vsyscall, but %ebp is not usable. + // Restore from 'ucv' instead. + void **const reg_esp = + reinterpret_cast(ucv->uc_mcontext.gregs[REG_ESP]); + // Check that alleged %esp is not null and is reasonably aligned. + if (reg_esp && + ((uintptr_t)reg_esp & (sizeof(reg_esp) - 1)) == 0) { + // Check that alleged %esp is actually readable. This is to prevent + // "double fault" in case we hit the first fault due to e.g. stack + // corruption. + void *const reg_esp2 = reg_esp[num_push_instructions - 1]; + if (AddressIsReadable(reg_esp2)) { + // Alleged %esp is readable, use it for further unwinding. + new_fp = reinterpret_cast(reg_esp2); + } + } + } + } + } +#endif + + const uintptr_t old_fp_u = reinterpret_cast(old_fp); + const uintptr_t new_fp_u = reinterpret_cast(new_fp); + + // Check that the transition from frame pointer old_fp to frame + // pointer new_fp isn't clearly bogus. Skip the checks if new_fp + // matches the signal context, so that we don't skip out early when + // using an alternate signal stack. + // + // TODO(bcmills): The GetFP call should be completely unnecessary when + // ENABLE_COMBINED_UNWINDER is set (because we should be back in the thread's + // stack by this point), but it is empirically still needed (e.g. when the + // stack includes a call to abort). unw_get_reg returns UNW_EBADREG for some + // frames. Figure out why GetValidFrameAddr and/or libunwind isn't doing what + // it's supposed to. + if (STRICT_UNWINDING && + (!WITH_CONTEXT || uc == nullptr || new_fp_u != GetFP(uc))) { + // With the stack growing downwards, older stack frame must be + // at a greater address that the current one. + if (new_fp_u <= old_fp_u) return nullptr; + if (new_fp_u - old_fp_u > kMaxFrameBytes) return nullptr; + + if (stack_low < old_fp_u && old_fp_u <= stack_high) { + // Old BP was in the expected stack region... + if (!(stack_low < new_fp_u && new_fp_u <= stack_high)) { + // ... but new BP is outside of expected stack region. + // It is most likely bogus. + return nullptr; + } + } else { + // We may be here if we are executing in a co-routine with a + // separate stack. We can't do safety checks in this case. + } + } else { + if (new_fp == nullptr) return nullptr; // skip AddressIsReadable() below + // In the non-strict mode, allow discontiguous stack frames. + // (alternate-signal-stacks for example). + if (new_fp == old_fp) return nullptr; + } + + if (new_fp_u & (sizeof(void *) - 1)) return nullptr; +#ifdef __i386__ + // On 32-bit machines, the stack pointer can be very close to + // 0xffffffff, so we explicitly check for a pointer into the + // last two pages in the address space + if (new_fp_u >= 0xffffe000) return nullptr; +#endif +#if !defined(_WIN32) + if (!STRICT_UNWINDING) { + // Lax sanity checks cause a crash in 32-bit tcmalloc/crash_reason_test + // on AMD-based machines with VDSO-enabled kernels. + // Make an extra sanity check to insure new_fp is readable. + // Note: NextStackFrame() is only called while the program + // is already on its last leg, so it's ok to be slow here. + + if (!AddressIsReadable(new_fp)) { + return nullptr; + } + } +#endif + return new_fp; +} + +template +ABSL_ATTRIBUTE_NO_SANITIZE_ADDRESS // May read random elements from stack. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY // May read random elements from stack. +ABSL_ATTRIBUTE_NOINLINE +static int UnwindImpl(void **result, int *sizes, int max_depth, int skip_count, + const void *ucp, int *min_dropped_frames) { + int n = 0; + void **fp = reinterpret_cast(__builtin_frame_address(0)); + + size_t stack_low = getpagesize(); // Assume that the first page is not stack. + size_t stack_high = std::numeric_limits::max() - sizeof(void *); + + while (fp && n < max_depth) { + if (*(fp + 1) == reinterpret_cast(0)) { + // In 64-bit code, we often see a frame that + // points to itself and has a return address of 0. + break; + } + void **next_fp = NextStackFrame( + fp, ucp, stack_low, stack_high); + if (skip_count > 0) { + skip_count--; + } else { + result[n] = *(fp + 1); + if (IS_STACK_FRAMES) { + if (next_fp > fp) { + sizes[n] = (uintptr_t)next_fp - (uintptr_t)fp; + } else { + // A frame-size of 0 is used to indicate unknown frame size. + sizes[n] = 0; + } + } + n++; + } + fp = next_fp; + } + if (min_dropped_frames != nullptr) { + // Implementation detail: we clamp the max of frames we are willing to + // count, so as not to spend too much time in the loop below. + const int kMaxUnwind = 1000; + int num_dropped_frames = 0; + for (int j = 0; fp != nullptr && j < kMaxUnwind; j++) { + if (skip_count > 0) { + skip_count--; + } else { + num_dropped_frames++; + } + fp = NextStackFrame(fp, ucp, stack_low, + stack_high); + } + *min_dropped_frames = num_dropped_frames; + } + return n; +} + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { +bool StackTraceWorksForTest() { + return true; +} +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_INTERNAL_STACKTRACE_X86_INL_INC_ diff --git a/src/absl/debugging/internal/symbolize.h b/src/absl/debugging/internal/symbolize.h new file mode 100644 index 0000000..27d5e65 --- /dev/null +++ b/src/absl/debugging/internal/symbolize.h @@ -0,0 +1,153 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This file contains internal parts of the Abseil symbolizer. +// Do not depend on the anything in this file, it may change at anytime. + +#ifndef ABSL_DEBUGGING_INTERNAL_SYMBOLIZE_H_ +#define ABSL_DEBUGGING_INTERNAL_SYMBOLIZE_H_ + +#ifdef __cplusplus + +#include +#include + +#include "absl/base/config.h" +#include "absl/strings/string_view.h" + +#ifdef ABSL_INTERNAL_HAVE_ELF_SYMBOLIZE +#error ABSL_INTERNAL_HAVE_ELF_SYMBOLIZE cannot be directly set +#elif defined(__ELF__) && defined(__GLIBC__) && !defined(__native_client__) \ + && !defined(__asmjs__) && !defined(__wasm__) +#define ABSL_INTERNAL_HAVE_ELF_SYMBOLIZE 1 + +#include +#include // For ElfW() macro. +#include +#include + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// Iterates over all sections, invoking callback on each with the section name +// and the section header. +// +// Returns true on success; otherwise returns false in case of errors. +// +// This is not async-signal-safe. +bool ForEachSection(int fd, + const std::function& callback); + +// Gets the section header for the given name, if it exists. Returns true on +// success. Otherwise, returns false. +bool GetSectionHeaderByName(int fd, const char *name, size_t name_len, + ElfW(Shdr) *out); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_INTERNAL_HAVE_ELF_SYMBOLIZE + +#ifdef ABSL_INTERNAL_HAVE_DARWIN_SYMBOLIZE +#error ABSL_INTERNAL_HAVE_DARWIN_SYMBOLIZE cannot be directly set +#elif defined(__APPLE__) +#define ABSL_INTERNAL_HAVE_DARWIN_SYMBOLIZE 1 +#endif + +#ifdef ABSL_INTERNAL_HAVE_EMSCRIPTEN_SYMBOLIZE +#error ABSL_INTERNAL_HAVE_EMSCRIPTEN_SYMBOLIZE cannot be directly set +#elif defined(__EMSCRIPTEN__) +#define ABSL_INTERNAL_HAVE_EMSCRIPTEN_SYMBOLIZE 1 +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +struct SymbolDecoratorArgs { + // The program counter we are getting symbolic name for. + const void *pc; + // 0 for main executable, load address for shared libraries. + ptrdiff_t relocation; + // Read-only file descriptor for ELF image covering "pc", + // or -1 if no such ELF image exists in /proc/self/maps. + int fd; + // Output buffer, size. + // Note: the buffer may not be empty -- default symbolizer may have already + // produced some output, and earlier decorators may have adorned it in + // some way. You are free to replace or augment the contents (within the + // symbol_buf_size limit). + char *const symbol_buf; + size_t symbol_buf_size; + // Temporary scratch space, size. + // Use that space in preference to allocating your own stack buffer to + // conserve stack. + char *const tmp_buf; + size_t tmp_buf_size; + // User-provided argument + void* arg; +}; +using SymbolDecorator = void (*)(const SymbolDecoratorArgs *); + +// Installs a function-pointer as a decorator. Returns a value less than zero +// if the system cannot install the decorator. Otherwise, returns a unique +// identifier corresponding to the decorator. This identifier can be used to +// uninstall the decorator - See RemoveSymbolDecorator() below. +int InstallSymbolDecorator(SymbolDecorator decorator, void* arg); + +// Removes a previously installed function-pointer decorator. Parameter "ticket" +// is the return-value from calling InstallSymbolDecorator(). +bool RemoveSymbolDecorator(int ticket); + +// Remove all installed decorators. Returns true if successful, false if +// symbolization is currently in progress. +bool RemoveAllSymbolDecorators(void); + +// Registers an address range to a file mapping. +// +// Preconditions: +// start <= end +// filename != nullptr +// +// Returns true if the file was successfully registered. +bool RegisterFileMappingHint(const void* start, const void* end, + uint64_t offset, const char* filename); + +// Looks up the file mapping registered by RegisterFileMappingHint for an +// address range. If there is one, the file name is stored in *filename and +// *start and *end are modified to reflect the registered mapping. Returns +// whether any hint was found. +bool GetFileMappingHint(const void** start, const void** end, uint64_t* offset, + const char** filename); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // __cplusplus + +#include + +#ifdef __cplusplus +extern "C" +#endif // __cplusplus + + bool + AbslInternalGetFileMappingHint(const void** start, const void** end, + uint64_t* offset, const char** filename); + +#endif // ABSL_DEBUGGING_INTERNAL_SYMBOLIZE_H_ diff --git a/src/absl/debugging/internal/vdso_support.cc b/src/absl/debugging/internal/vdso_support.cc new file mode 100644 index 0000000..40eb055 --- /dev/null +++ b/src/absl/debugging/internal/vdso_support.cc @@ -0,0 +1,204 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Allow dynamic symbol lookup in the kernel VDSO page. +// +// VDSOSupport -- a class representing kernel VDSO (if present). + +#include "absl/debugging/internal/vdso_support.h" + +#ifdef ABSL_HAVE_VDSO_SUPPORT // defined in vdso_support.h + +#if !defined(__has_include) +#define __has_include(header) 0 +#endif + +#include +#include +#if __has_include() +#include +#elif __has_include() +#include +#endif +#include + +#if !defined(__UCLIBC__) && defined(__GLIBC__) && \ + (__GLIBC__ > 2 || (__GLIBC__ == 2 && __GLIBC_MINOR__ >= 16)) +#define ABSL_HAVE_GETAUXVAL +#endif + +#ifdef ABSL_HAVE_GETAUXVAL +#include +#endif + +#include "absl/base/dynamic_annotations.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/port.h" + +#ifndef AT_SYSINFO_EHDR +#define AT_SYSINFO_EHDR 33 // for crosstoolv10 +#endif + +#if defined(__NetBSD__) +using Elf32_auxv_t = Aux32Info; +using Elf64_auxv_t = Aux64Info; +#endif +#if defined(__FreeBSD__) +#if defined(__ELF_WORD_SIZE) && __ELF_WORD_SIZE == 64 +using Elf64_auxv_t = Elf64_Auxinfo; +#endif +using Elf32_auxv_t = Elf32_Auxinfo; +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +ABSL_CONST_INIT +std::atomic VDSOSupport::vdso_base_( + debugging_internal::ElfMemImage::kInvalidBase); + +ABSL_CONST_INIT std::atomic VDSOSupport::getcpu_fn_( + &InitAndGetCPU); + +VDSOSupport::VDSOSupport() + // If vdso_base_ is still set to kInvalidBase, we got here + // before VDSOSupport::Init has been called. Call it now. + : image_(vdso_base_.load(std::memory_order_relaxed) == + debugging_internal::ElfMemImage::kInvalidBase + ? Init() + : vdso_base_.load(std::memory_order_relaxed)) {} + +// NOTE: we can't use GoogleOnceInit() below, because we can be +// called by tcmalloc, and none of the *once* stuff may be functional yet. +// +// In addition, we hope that the VDSOSupportHelper constructor +// causes this code to run before there are any threads, and before +// InitGoogle() has executed any chroot or setuid calls. +// +// Finally, even if there is a race here, it is harmless, because +// the operation should be idempotent. +const void *VDSOSupport::Init() { + const auto kInvalidBase = debugging_internal::ElfMemImage::kInvalidBase; +#ifdef ABSL_HAVE_GETAUXVAL + if (vdso_base_.load(std::memory_order_relaxed) == kInvalidBase) { + errno = 0; + const void *const sysinfo_ehdr = + reinterpret_cast(getauxval(AT_SYSINFO_EHDR)); + if (errno == 0) { + vdso_base_.store(sysinfo_ehdr, std::memory_order_relaxed); + } + } +#endif // ABSL_HAVE_GETAUXVAL + if (vdso_base_.load(std::memory_order_relaxed) == kInvalidBase) { + int fd = open("/proc/self/auxv", O_RDONLY); + if (fd == -1) { + // Kernel too old to have a VDSO. + vdso_base_.store(nullptr, std::memory_order_relaxed); + getcpu_fn_.store(&GetCPUViaSyscall, std::memory_order_relaxed); + return nullptr; + } + ElfW(auxv_t) aux; + while (read(fd, &aux, sizeof(aux)) == sizeof(aux)) { + if (aux.a_type == AT_SYSINFO_EHDR) { +#if defined(__NetBSD__) + vdso_base_.store(reinterpret_cast(aux.a_v), + std::memory_order_relaxed); +#else + vdso_base_.store(reinterpret_cast(aux.a_un.a_val), + std::memory_order_relaxed); +#endif + break; + } + } + close(fd); + if (vdso_base_.load(std::memory_order_relaxed) == kInvalidBase) { + // Didn't find AT_SYSINFO_EHDR in auxv[]. + vdso_base_.store(nullptr, std::memory_order_relaxed); + } + } + GetCpuFn fn = &GetCPUViaSyscall; // default if VDSO not present. + if (vdso_base_.load(std::memory_order_relaxed)) { + VDSOSupport vdso; + SymbolInfo info; + if (vdso.LookupSymbol("__vdso_getcpu", "LINUX_2.6", STT_FUNC, &info)) { + fn = reinterpret_cast(const_cast(info.address)); + } + } + // Subtle: this code runs outside of any locks; prevent compiler + // from assigning to getcpu_fn_ more than once. + getcpu_fn_.store(fn, std::memory_order_relaxed); + return vdso_base_.load(std::memory_order_relaxed); +} + +const void *VDSOSupport::SetBase(const void *base) { + ABSL_RAW_CHECK(base != debugging_internal::ElfMemImage::kInvalidBase, + "internal error"); + const void *old_base = vdso_base_.load(std::memory_order_relaxed); + vdso_base_.store(base, std::memory_order_relaxed); + image_.Init(base); + // Also reset getcpu_fn_, so GetCPU could be tested with simulated VDSO. + getcpu_fn_.store(&InitAndGetCPU, std::memory_order_relaxed); + return old_base; +} + +bool VDSOSupport::LookupSymbol(const char *name, + const char *version, + int type, + SymbolInfo *info) const { + return image_.LookupSymbol(name, version, type, info); +} + +bool VDSOSupport::LookupSymbolByAddress(const void *address, + SymbolInfo *info_out) const { + return image_.LookupSymbolByAddress(address, info_out); +} + +// NOLINT on 'long' because this routine mimics kernel api. +long VDSOSupport::GetCPUViaSyscall(unsigned *cpu, // NOLINT(runtime/int) + void *, void *) { +#ifdef SYS_getcpu + return syscall(SYS_getcpu, cpu, nullptr, nullptr); +#else + // x86_64 never implemented sys_getcpu(), except as a VDSO call. + static_cast(cpu); // Avoid an unused argument compiler warning. + errno = ENOSYS; + return -1; +#endif +} + +// Use fast __vdso_getcpu if available. +long VDSOSupport::InitAndGetCPU(unsigned *cpu, // NOLINT(runtime/int) + void *x, void *y) { + Init(); + GetCpuFn fn = getcpu_fn_.load(std::memory_order_relaxed); + ABSL_RAW_CHECK(fn != &InitAndGetCPU, "Init() did not set getcpu_fn_"); + return (*fn)(cpu, x, y); +} + +// This function must be very fast, and may be called from very +// low level (e.g. tcmalloc). Hence I avoid things like +// GoogleOnceInit() and ::operator new. +ABSL_ATTRIBUTE_NO_SANITIZE_MEMORY +int GetCPU() { + unsigned cpu; + int ret_code = (*VDSOSupport::getcpu_fn_)(&cpu, nullptr, nullptr); + return ret_code == 0 ? cpu : ret_code; +} + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_HAVE_VDSO_SUPPORT diff --git a/src/absl/debugging/internal/vdso_support.h b/src/absl/debugging/internal/vdso_support.h new file mode 100644 index 0000000..6562c6c --- /dev/null +++ b/src/absl/debugging/internal/vdso_support.h @@ -0,0 +1,158 @@ +// +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// + +// Allow dynamic symbol lookup in the kernel VDSO page. +// +// VDSO stands for "Virtual Dynamic Shared Object" -- a page of +// executable code, which looks like a shared library, but doesn't +// necessarily exist anywhere on disk, and which gets mmap()ed into +// every process by kernels which support VDSO, such as 2.6.x for 32-bit +// executables, and 2.6.24 and above for 64-bit executables. +// +// More details could be found here: +// http://www.trilithium.com/johan/2005/08/linux-gate/ +// +// VDSOSupport -- a class representing kernel VDSO (if present). +// +// Example usage: +// VDSOSupport vdso; +// VDSOSupport::SymbolInfo info; +// typedef (*FN)(unsigned *, void *, void *); +// FN fn = nullptr; +// if (vdso.LookupSymbol("__vdso_getcpu", "LINUX_2.6", STT_FUNC, &info)) { +// fn = reinterpret_cast(info.address); +// } + +#ifndef ABSL_DEBUGGING_INTERNAL_VDSO_SUPPORT_H_ +#define ABSL_DEBUGGING_INTERNAL_VDSO_SUPPORT_H_ + +#include + +#include "absl/base/attributes.h" +#include "absl/debugging/internal/elf_mem_image.h" + +#ifdef ABSL_HAVE_ELF_MEM_IMAGE + +#ifdef ABSL_HAVE_VDSO_SUPPORT +#error ABSL_HAVE_VDSO_SUPPORT cannot be directly set +#else +#define ABSL_HAVE_VDSO_SUPPORT 1 +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace debugging_internal { + +// NOTE: this class may be used from within tcmalloc, and can not +// use any memory allocation routines. +class VDSOSupport { + public: + VDSOSupport(); + + typedef ElfMemImage::SymbolInfo SymbolInfo; + typedef ElfMemImage::SymbolIterator SymbolIterator; + + // On PowerPC64 VDSO symbols can either be of type STT_FUNC or STT_NOTYPE + // depending on how the kernel is built. The kernel is normally built with + // STT_NOTYPE type VDSO symbols. Let's make things simpler first by using a + // compile-time constant. +#ifdef __powerpc64__ + enum { kVDSOSymbolType = STT_NOTYPE }; +#else + enum { kVDSOSymbolType = STT_FUNC }; +#endif + + // Answers whether we have a vdso at all. + bool IsPresent() const { return image_.IsPresent(); } + + // Allow to iterate over all VDSO symbols. + SymbolIterator begin() const { return image_.begin(); } + SymbolIterator end() const { return image_.end(); } + + // Look up versioned dynamic symbol in the kernel VDSO. + // Returns false if VDSO is not present, or doesn't contain given + // symbol/version/type combination. + // If info_out != nullptr, additional details are filled in. + bool LookupSymbol(const char *name, const char *version, + int symbol_type, SymbolInfo *info_out) const; + + // Find info about symbol (if any) which overlaps given address. + // Returns true if symbol was found; false if VDSO isn't present + // or doesn't have a symbol overlapping given address. + // If info_out != nullptr, additional details are filled in. + bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const; + + // Used only for testing. Replace real VDSO base with a mock. + // Returns previous value of vdso_base_. After you are done testing, + // you are expected to call SetBase() with previous value, in order to + // reset state to the way it was. + const void *SetBase(const void *s); + + // Computes vdso_base_ and returns it. Should be called as early as + // possible; before any thread creation, chroot or setuid. + static const void *Init(); + + private: + // image_ represents VDSO ELF image in memory. + // image_.ehdr_ == nullptr implies there is no VDSO. + ElfMemImage image_; + + // Cached value of auxv AT_SYSINFO_EHDR, computed once. + // This is a tri-state: + // kInvalidBase => value hasn't been determined yet. + // 0 => there is no VDSO. + // else => vma of VDSO Elf{32,64}_Ehdr. + // + // When testing with mock VDSO, low bit is set. + // The low bit is always available because vdso_base_ is + // page-aligned. + static std::atomic vdso_base_; + + // NOLINT on 'long' because these routines mimic kernel api. + // The 'cache' parameter may be used by some versions of the kernel, + // and should be nullptr or point to a static buffer containing at + // least two 'long's. + static long InitAndGetCPU(unsigned *cpu, void *cache, // NOLINT 'long'. + void *unused); + static long GetCPUViaSyscall(unsigned *cpu, void *cache, // NOLINT 'long'. + void *unused); + typedef long (*GetCpuFn)(unsigned *cpu, void *cache, // NOLINT 'long'. + void *unused); + + // This function pointer may point to InitAndGetCPU, + // GetCPUViaSyscall, or __vdso_getcpu at different stages of initialization. + ABSL_CONST_INIT static std::atomic getcpu_fn_; + + friend int GetCPU(void); // Needs access to getcpu_fn_. + + VDSOSupport(const VDSOSupport&) = delete; + VDSOSupport& operator=(const VDSOSupport&) = delete; +}; + +// Same as sched_getcpu() on later glibc versions. +// Return current CPU, using (fast) __vdso_getcpu@LINUX_2.6 if present, +// otherwise use syscall(SYS_getcpu,...). +// May return -1 with errno == ENOSYS if the kernel doesn't +// support SYS_getcpu. +int GetCPU(); + +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_HAVE_ELF_MEM_IMAGE + +#endif // ABSL_DEBUGGING_INTERNAL_VDSO_SUPPORT_H_ diff --git a/src/absl/debugging/leak_check.cc b/src/absl/debugging/leak_check.cc new file mode 100644 index 0000000..195e82b --- /dev/null +++ b/src/absl/debugging/leak_check.cc @@ -0,0 +1,73 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Wrappers around lsan_interface functions. +// +// These are always-available run-time functions manipulating the LeakSanitizer, +// even when the lsan_interface (and LeakSanitizer) is not available. When +// LeakSanitizer is not linked in, these functions become no-op stubs. + +#include "absl/debugging/leak_check.h" + +#include "absl/base/attributes.h" +#include "absl/base/config.h" + +#if defined(ABSL_HAVE_LEAK_SANITIZER) + +#include + +#if ABSL_HAVE_ATTRIBUTE_WEAK +extern "C" ABSL_ATTRIBUTE_WEAK int __lsan_is_turned_off(); +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +bool HaveLeakSanitizer() { return true; } + +#if ABSL_HAVE_ATTRIBUTE_WEAK +bool LeakCheckerIsActive() { + return !(&__lsan_is_turned_off && __lsan_is_turned_off()); +} +#else +bool LeakCheckerIsActive() { return true; } +#endif + +bool FindAndReportLeaks() { return __lsan_do_recoverable_leak_check(); } +void DoIgnoreLeak(const void* ptr) { __lsan_ignore_object(ptr); } +void RegisterLivePointers(const void* ptr, size_t size) { + __lsan_register_root_region(ptr, size); +} +void UnRegisterLivePointers(const void* ptr, size_t size) { + __lsan_unregister_root_region(ptr, size); +} +LeakCheckDisabler::LeakCheckDisabler() { __lsan_disable(); } +LeakCheckDisabler::~LeakCheckDisabler() { __lsan_enable(); } +ABSL_NAMESPACE_END +} // namespace absl + +#else // defined(ABSL_HAVE_LEAK_SANITIZER) + +namespace absl { +ABSL_NAMESPACE_BEGIN +bool HaveLeakSanitizer() { return false; } +bool LeakCheckerIsActive() { return false; } +void DoIgnoreLeak(const void*) { } +void RegisterLivePointers(const void*, size_t) { } +void UnRegisterLivePointers(const void*, size_t) { } +LeakCheckDisabler::LeakCheckDisabler() { } +LeakCheckDisabler::~LeakCheckDisabler() { } +ABSL_NAMESPACE_END +} // namespace absl + +#endif // defined(ABSL_HAVE_LEAK_SANITIZER) diff --git a/src/absl/debugging/leak_check.h b/src/absl/debugging/leak_check.h new file mode 100644 index 0000000..eff162f --- /dev/null +++ b/src/absl/debugging/leak_check.h @@ -0,0 +1,150 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: leak_check.h +// ----------------------------------------------------------------------------- +// +// This file contains functions that affect leak checking behavior within +// targets built with the LeakSanitizer (LSan), a memory leak detector that is +// integrated within the AddressSanitizer (ASan) as an additional component, or +// which can be used standalone. LSan and ASan are included (or can be provided) +// as additional components for most compilers such as Clang, gcc and MSVC. +// Note: this leak checking API is not yet supported in MSVC. +// Leak checking is enabled by default in all ASan builds. +// +// https://clang.llvm.org/docs/LeakSanitizer.html +// https://github.com/google/sanitizers/wiki/AddressSanitizerLeakSanitizer +// +// GCC and Clang both automatically enable LeakSanitizer when AddressSanitizer +// is enabled. To use the mode, simply pass `-fsanitize=address` to both the +// compiler and linker. An example Bazel command could be +// +// $ bazel test --copt=-fsanitize=address --linkopt=-fsanitize=address ... +// +// GCC and Clang auto support a standalone LeakSanitizer mode (a mode which does +// not also use AddressSanitizer). To use the mode, simply pass +// `-fsanitize=leak` to both the compiler and linker. Since GCC does not +// currently provide a way of detecting this mode at compile-time, GCC users +// must also pass -DLEAK_SANIITIZER to the compiler. An example Bazel command +// could be +// +// $ bazel test --copt=-DLEAK_SANITIZER --copt=-fsanitize=leak +// --linkopt=-fsanitize=leak ... +// +// ----------------------------------------------------------------------------- +#ifndef ABSL_DEBUGGING_LEAK_CHECK_H_ +#define ABSL_DEBUGGING_LEAK_CHECK_H_ + +#include + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// HaveLeakSanitizer() +// +// Returns true if a leak-checking sanitizer (either ASan or standalone LSan) is +// currently built into this target. +bool HaveLeakSanitizer(); + +// LeakCheckerIsActive() +// +// Returns true if a leak-checking sanitizer (either ASan or standalone LSan) is +// currently built into this target and is turned on. +bool LeakCheckerIsActive(); + +// DoIgnoreLeak() +// +// Implements `IgnoreLeak()` below. This function should usually +// not be called directly; calling `IgnoreLeak()` is preferred. +void DoIgnoreLeak(const void* ptr); + +// IgnoreLeak() +// +// Instruct the leak sanitizer to ignore leak warnings on the object referenced +// by the passed pointer, as well as all heap objects transitively referenced +// by it. The passed object pointer can point to either the beginning of the +// object or anywhere within it. +// +// Example: +// +// static T* obj = IgnoreLeak(new T(...)); +// +// If the passed `ptr` does not point to an actively allocated object at the +// time `IgnoreLeak()` is called, the call is a no-op; if it is actively +// allocated, leak sanitizer will assume this object is referenced even if +// there is no actual reference in user memory. +// +template +T* IgnoreLeak(T* ptr) { + DoIgnoreLeak(ptr); + return ptr; +} + +// FindAndReportLeaks() +// +// If any leaks are detected, prints a leak report and returns true. This +// function may be called repeatedly, and does not affect end-of-process leak +// checking. +// +// Example: +// if (FindAndReportLeaks()) { +// ... diagnostic already printed. Exit with failure code. +// exit(1) +// } +bool FindAndReportLeaks(); + +// LeakCheckDisabler +// +// This helper class indicates that any heap allocations done in the code block +// covered by the scoped object, which should be allocated on the stack, will +// not be reported as leaks. Leak check disabling will occur within the code +// block and any nested function calls within the code block. +// +// Example: +// +// void Foo() { +// LeakCheckDisabler disabler; +// ... code that allocates objects whose leaks should be ignored ... +// } +// +// REQUIRES: Destructor runs in same thread as constructor +class LeakCheckDisabler { + public: + LeakCheckDisabler(); + LeakCheckDisabler(const LeakCheckDisabler&) = delete; + LeakCheckDisabler& operator=(const LeakCheckDisabler&) = delete; + ~LeakCheckDisabler(); +}; + +// RegisterLivePointers() +// +// Registers `ptr[0,size-1]` as pointers to memory that is still actively being +// referenced and for which leak checking should be ignored. This function is +// useful if you store pointers in mapped memory, for memory ranges that we know +// are correct but for which normal analysis would flag as leaked code. +void RegisterLivePointers(const void* ptr, size_t size); + +// UnRegisterLivePointers() +// +// Deregisters the pointers previously marked as active in +// `RegisterLivePointers()`, enabling leak checking of those pointers. +void UnRegisterLivePointers(const void* ptr, size_t size); + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_LEAK_CHECK_H_ diff --git a/src/absl/debugging/stacktrace.cc b/src/absl/debugging/stacktrace.cc new file mode 100644 index 0000000..ff8069f --- /dev/null +++ b/src/absl/debugging/stacktrace.cc @@ -0,0 +1,142 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Produce stack trace. +// +// There are three different ways we can try to get the stack trace: +// +// 1) Our hand-coded stack-unwinder. This depends on a certain stack +// layout, which is used by gcc (and those systems using a +// gcc-compatible ABI) on x86 systems, at least since gcc 2.95. +// It uses the frame pointer to do its work. +// +// 2) The libunwind library. This is still in development, and as a +// separate library adds a new dependency, but doesn't need a frame +// pointer. It also doesn't call malloc. +// +// 3) The gdb unwinder -- also the one used by the c++ exception code. +// It's obviously well-tested, but has a fatal flaw: it can call +// malloc() from the unwinder. This is a problem because we're +// trying to use the unwinder to instrument malloc(). +// +// Note: if you add a new implementation here, make sure it works +// correctly when absl::GetStackTrace() is called with max_depth == 0. +// Some code may do that. + +#include "absl/debugging/stacktrace.h" + +#include + +#include "absl/base/attributes.h" +#include "absl/base/port.h" +#include "absl/debugging/internal/stacktrace_config.h" + +#if defined(ABSL_STACKTRACE_INL_HEADER) +#include ABSL_STACKTRACE_INL_HEADER +#else +# error Cannot calculate stack trace: will need to write for your environment + +# include "absl/debugging/internal/stacktrace_aarch64-inl.inc" +# include "absl/debugging/internal/stacktrace_arm-inl.inc" +# include "absl/debugging/internal/stacktrace_emscripten-inl.inc" +# include "absl/debugging/internal/stacktrace_generic-inl.inc" +# include "absl/debugging/internal/stacktrace_powerpc-inl.inc" +# include "absl/debugging/internal/stacktrace_riscv-inl.inc" +# include "absl/debugging/internal/stacktrace_unimplemented-inl.inc" +# include "absl/debugging/internal/stacktrace_win32-inl.inc" +# include "absl/debugging/internal/stacktrace_x86-inl.inc" +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace { + +typedef int (*Unwinder)(void**, int*, int, int, const void*, int*); +std::atomic custom; + +template +ABSL_ATTRIBUTE_ALWAYS_INLINE inline int Unwind(void** result, int* sizes, + int max_depth, int skip_count, + const void* uc, + int* min_dropped_frames) { + Unwinder f = &UnwindImpl; + Unwinder g = custom.load(std::memory_order_acquire); + if (g != nullptr) f = g; + + // Add 1 to skip count for the unwinder function itself + int size = (*f)(result, sizes, max_depth, skip_count + 1, uc, + min_dropped_frames); + // To disable tail call to (*f)(...) + ABSL_BLOCK_TAIL_CALL_OPTIMIZATION(); + return size; +} + +} // anonymous namespace + +ABSL_ATTRIBUTE_NOINLINE ABSL_ATTRIBUTE_NO_TAIL_CALL int GetStackFrames( + void** result, int* sizes, int max_depth, int skip_count) { + return Unwind(result, sizes, max_depth, skip_count, nullptr, + nullptr); +} + +ABSL_ATTRIBUTE_NOINLINE ABSL_ATTRIBUTE_NO_TAIL_CALL int +GetStackFramesWithContext(void** result, int* sizes, int max_depth, + int skip_count, const void* uc, + int* min_dropped_frames) { + return Unwind(result, sizes, max_depth, skip_count, uc, + min_dropped_frames); +} + +ABSL_ATTRIBUTE_NOINLINE ABSL_ATTRIBUTE_NO_TAIL_CALL int GetStackTrace( + void** result, int max_depth, int skip_count) { + return Unwind(result, nullptr, max_depth, skip_count, nullptr, + nullptr); +} + +ABSL_ATTRIBUTE_NOINLINE ABSL_ATTRIBUTE_NO_TAIL_CALL int +GetStackTraceWithContext(void** result, int max_depth, int skip_count, + const void* uc, int* min_dropped_frames) { + return Unwind(result, nullptr, max_depth, skip_count, uc, + min_dropped_frames); +} + +void SetStackUnwinder(Unwinder w) { + custom.store(w, std::memory_order_release); +} + +int DefaultStackUnwinder(void** pcs, int* sizes, int depth, int skip, + const void* uc, int* min_dropped_frames) { + skip++; // For this function + Unwinder f = nullptr; + if (sizes == nullptr) { + if (uc == nullptr) { + f = &UnwindImpl; + } else { + f = &UnwindImpl; + } + } else { + if (uc == nullptr) { + f = &UnwindImpl; + } else { + f = &UnwindImpl; + } + } + volatile int x = 0; + int n = (*f)(pcs, sizes, depth, skip, uc, min_dropped_frames); + x = 1; (void) x; // To disable tail call to (*f)(...) + return n; +} + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/stacktrace.h b/src/absl/debugging/stacktrace.h new file mode 100644 index 0000000..0ec0ffd --- /dev/null +++ b/src/absl/debugging/stacktrace.h @@ -0,0 +1,231 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: stacktrace.h +// ----------------------------------------------------------------------------- +// +// This file contains routines to extract the current stack trace and associated +// stack frames. These functions are thread-safe and async-signal-safe. +// +// Note that stack trace functionality is platform dependent and requires +// additional support from the compiler/build system in most cases. (That is, +// this functionality generally only works on platforms/builds that have been +// specifically configured to support it.) +// +// Note: stack traces in Abseil that do not utilize a symbolizer will result in +// frames consisting of function addresses rather than human-readable function +// names. (See symbolize.h for information on symbolizing these values.) + +#ifndef ABSL_DEBUGGING_STACKTRACE_H_ +#define ABSL_DEBUGGING_STACKTRACE_H_ + +#include "absl/base/config.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// GetStackFrames() +// +// Records program counter values for up to `max_depth` frames, skipping the +// most recent `skip_count` stack frames, stores their corresponding values +// and sizes in `results` and `sizes` buffers, and returns the number of frames +// stored. (Note that the frame generated for the `absl::GetStackFrames()` +// routine itself is also skipped.) +// +// Example: +// +// main() { foo(); } +// foo() { bar(); } +// bar() { +// void* result[10]; +// int sizes[10]; +// int depth = absl::GetStackFrames(result, sizes, 10, 1); +// } +// +// The current stack frame would consist of three function calls: `bar()`, +// `foo()`, and then `main()`; however, since the `GetStackFrames()` call sets +// `skip_count` to `1`, it will skip the frame for `bar()`, the most recently +// invoked function call. It will therefore return 2 and fill `result` with +// program counters within the following functions: +// +// result[0] foo() +// result[1] main() +// +// (Note: in practice, a few more entries after `main()` may be added to account +// for startup processes.) +// +// Corresponding stack frame sizes will also be recorded: +// +// sizes[0] 16 +// sizes[1] 16 +// +// (Stack frame sizes of `16` above are just for illustration purposes.) +// +// Stack frame sizes of 0 or less indicate that those frame sizes couldn't +// be identified. +// +// This routine may return fewer stack frame entries than are +// available. Also note that `result` and `sizes` must both be non-null. +extern int GetStackFrames(void** result, int* sizes, int max_depth, + int skip_count); + +// GetStackFramesWithContext() +// +// Records program counter values obtained from a signal handler. Records +// program counter values for up to `max_depth` frames, skipping the most recent +// `skip_count` stack frames, stores their corresponding values and sizes in +// `results` and `sizes` buffers, and returns the number of frames stored. (Note +// that the frame generated for the `absl::GetStackFramesWithContext()` routine +// itself is also skipped.) +// +// The `uc` parameter, if non-null, should be a pointer to a `ucontext_t` value +// passed to a signal handler registered via the `sa_sigaction` field of a +// `sigaction` struct. (See +// http://man7.org/linux/man-pages/man2/sigaction.2.html.) The `uc` value may +// help a stack unwinder to provide a better stack trace under certain +// conditions. `uc` may safely be null. +// +// The `min_dropped_frames` output parameter, if non-null, points to the +// location to note any dropped stack frames, if any, due to buffer limitations +// or other reasons. (This value will be set to `0` if no frames were dropped.) +// The number of total stack frames is guaranteed to be >= skip_count + +// max_depth + *min_dropped_frames. +extern int GetStackFramesWithContext(void** result, int* sizes, int max_depth, + int skip_count, const void* uc, + int* min_dropped_frames); + +// GetStackTrace() +// +// Records program counter values for up to `max_depth` frames, skipping the +// most recent `skip_count` stack frames, stores their corresponding values +// in `results`, and returns the number of frames +// stored. Note that this function is similar to `absl::GetStackFrames()` +// except that it returns the stack trace only, and not stack frame sizes. +// +// Example: +// +// main() { foo(); } +// foo() { bar(); } +// bar() { +// void* result[10]; +// int depth = absl::GetStackTrace(result, 10, 1); +// } +// +// This produces: +// +// result[0] foo +// result[1] main +// .... ... +// +// `result` must not be null. +extern int GetStackTrace(void** result, int max_depth, int skip_count); + +// GetStackTraceWithContext() +// +// Records program counter values obtained from a signal handler. Records +// program counter values for up to `max_depth` frames, skipping the most recent +// `skip_count` stack frames, stores their corresponding values in `results`, +// and returns the number of frames stored. (Note that the frame generated for +// the `absl::GetStackFramesWithContext()` routine itself is also skipped.) +// +// The `uc` parameter, if non-null, should be a pointer to a `ucontext_t` value +// passed to a signal handler registered via the `sa_sigaction` field of a +// `sigaction` struct. (See +// http://man7.org/linux/man-pages/man2/sigaction.2.html.) The `uc` value may +// help a stack unwinder to provide a better stack trace under certain +// conditions. `uc` may safely be null. +// +// The `min_dropped_frames` output parameter, if non-null, points to the +// location to note any dropped stack frames, if any, due to buffer limitations +// or other reasons. (This value will be set to `0` if no frames were dropped.) +// The number of total stack frames is guaranteed to be >= skip_count + +// max_depth + *min_dropped_frames. +extern int GetStackTraceWithContext(void** result, int max_depth, + int skip_count, const void* uc, + int* min_dropped_frames); + +// SetStackUnwinder() +// +// Provides a custom function for unwinding stack frames that will be used in +// place of the default stack unwinder when invoking the static +// GetStack{Frames,Trace}{,WithContext}() functions above. +// +// The arguments passed to the unwinder function will match the +// arguments passed to `absl::GetStackFramesWithContext()` except that sizes +// will be non-null iff the caller is interested in frame sizes. +// +// If unwinder is set to null, we revert to the default stack-tracing behavior. +// +// ***************************************************************************** +// WARNING +// ***************************************************************************** +// +// absl::SetStackUnwinder is not suitable for general purpose use. It is +// provided for custom runtimes. +// Some things to watch out for when calling `absl::SetStackUnwinder()`: +// +// (a) The unwinder may be called from within signal handlers and +// therefore must be async-signal-safe. +// +// (b) Even after a custom stack unwinder has been unregistered, other +// threads may still be in the process of using that unwinder. +// Therefore do not clean up any state that may be needed by an old +// unwinder. +// ***************************************************************************** +extern void SetStackUnwinder(int (*unwinder)(void** pcs, int* sizes, + int max_depth, int skip_count, + const void* uc, + int* min_dropped_frames)); + +// DefaultStackUnwinder() +// +// Records program counter values of up to `max_depth` frames, skipping the most +// recent `skip_count` stack frames, and stores their corresponding values in +// `pcs`. (Note that the frame generated for this call itself is also skipped.) +// This function acts as a generic stack-unwinder; prefer usage of the more +// specific `GetStack{Trace,Frames}{,WithContext}()` functions above. +// +// If you have set your own stack unwinder (with the `SetStackUnwinder()` +// function above, you can still get the default stack unwinder by calling +// `DefaultStackUnwinder()`, which will ignore any previously set stack unwinder +// and use the default one instead. +// +// Because this function is generic, only `pcs` is guaranteed to be non-null +// upon return. It is legal for `sizes`, `uc`, and `min_dropped_frames` to all +// be null when called. +// +// The semantics are the same as the corresponding `GetStack*()` function in the +// case where `absl::SetStackUnwinder()` was never called. Equivalents are: +// +// null sizes | non-nullptr sizes +// |==========================================================| +// null uc | GetStackTrace() | GetStackFrames() | +// non-null uc | GetStackTraceWithContext() | GetStackFramesWithContext() | +// |==========================================================| +extern int DefaultStackUnwinder(void** pcs, int* sizes, int max_depth, + int skip_count, const void* uc, + int* min_dropped_frames); + +namespace debugging_internal { +// Returns true for platforms which are expected to have functioning stack trace +// implementations. Intended to be used for tests which want to exclude +// verification of logic known to be broken because stack traces are not +// working. +extern bool StackTraceWorksForTest(); +} // namespace debugging_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_STACKTRACE_H_ diff --git a/src/absl/debugging/symbolize.cc b/src/absl/debugging/symbolize.cc new file mode 100644 index 0000000..de4c276 --- /dev/null +++ b/src/absl/debugging/symbolize.cc @@ -0,0 +1,17 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "absl/debugging/symbolize.h" + +#include "absl/debugging/symbolize_unimplemented.inc" diff --git a/src/absl/debugging/symbolize.h b/src/absl/debugging/symbolize.h new file mode 100644 index 0000000..43d93a8 --- /dev/null +++ b/src/absl/debugging/symbolize.h @@ -0,0 +1,99 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: symbolize.h +// ----------------------------------------------------------------------------- +// +// This file configures the Abseil symbolizer for use in converting instruction +// pointer addresses (program counters) into human-readable names (function +// calls, etc.) within Abseil code. +// +// The symbolizer may be invoked from several sources: +// +// * Implicitly, through the installation of an Abseil failure signal handler. +// (See failure_signal_handler.h for more information.) +// * By calling `Symbolize()` directly on a program counter you obtain through +// `absl::GetStackTrace()` or `absl::GetStackFrames()`. (See stacktrace.h +// for more information. +// * By calling `Symbolize()` directly on a program counter you obtain through +// other means (which would be platform-dependent). +// +// In all of the above cases, the symbolizer must first be initialized before +// any program counter values can be symbolized. If you are installing a failure +// signal handler, initialize the symbolizer before you do so. +// +// Example: +// +// int main(int argc, char** argv) { +// // Initialize the Symbolizer before installing the failure signal handler +// absl::InitializeSymbolizer(argv[0]); +// +// // Now you may install the failure signal handler +// absl::FailureSignalHandlerOptions options; +// absl::InstallFailureSignalHandler(options); +// +// // Start running your main program +// ... +// return 0; +// } +// +#ifndef ABSL_DEBUGGING_SYMBOLIZE_H_ +#define ABSL_DEBUGGING_SYMBOLIZE_H_ + +#include "absl/debugging/internal/symbolize.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// InitializeSymbolizer() +// +// Initializes the program counter symbolizer, given the path of the program +// (typically obtained through `main()`s `argv[0]`). The Abseil symbolizer +// allows you to read program counters (instruction pointer values) using their +// human-readable names within output such as stack traces. +// +// Example: +// +// int main(int argc, char *argv[]) { +// absl::InitializeSymbolizer(argv[0]); +// // Now you can use the symbolizer +// } +void InitializeSymbolizer(const char* argv0); +// +// Symbolize() +// +// Symbolizes a program counter (instruction pointer value) `pc` and, on +// success, writes the name to `out`. The symbol name is demangled, if possible. +// Note that the symbolized name may be truncated and will be NUL-terminated. +// Demangling is supported for symbols generated by GCC 3.x or newer). Returns +// `false` on failure. +// +// Example: +// +// // Print a program counter and its symbol name. +// static void DumpPCAndSymbol(void *pc) { +// char tmp[1024]; +// const char *symbol = "(unknown)"; +// if (absl::Symbolize(pc, tmp, sizeof(tmp))) { +// symbol = tmp; +// } +// absl::PrintF("%p %s\n", pc, symbol); +// } +bool Symbolize(const void *pc, char *out, int out_size); + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_DEBUGGING_SYMBOLIZE_H_ diff --git a/src/absl/debugging/symbolize_darwin.inc b/src/absl/debugging/symbolize_darwin.inc new file mode 100644 index 0000000..443ce9e --- /dev/null +++ b/src/absl/debugging/symbolize_darwin.inc @@ -0,0 +1,101 @@ +// Copyright 2020 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include + +#include +#include + +#include "absl/base/internal/raw_logging.h" +#include "absl/debugging/internal/demangle.h" +#include "absl/strings/numbers.h" +#include "absl/strings/str_cat.h" +#include "absl/strings/string_view.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +void InitializeSymbolizer(const char*) {} + +namespace debugging_internal { +namespace { + +static std::string GetSymbolString(absl::string_view backtrace_line) { + // Example Backtrace lines: + // 0 libimaging_shared.dylib 0x018c152a + // _ZNSt11_Deque_baseIN3nik7mediadb4PageESaIS2_EE17_M_initialize_mapEm + 3478 + // + // or + // 0 libimaging_shared.dylib 0x0000000001895c39 + // _ZN3nik4util19register_shared_ptrINS_3gpu7TextureEEEvPKvS5_ + 39 + // + // or + // 0 mysterious_app 0x0124000120120009 main + 17 + auto address_pos = backtrace_line.find(" 0x"); + if (address_pos == absl::string_view::npos) return std::string(); + absl::string_view symbol_view = backtrace_line.substr(address_pos + 1); + + auto space_pos = symbol_view.find(" "); + if (space_pos == absl::string_view::npos) return std::string(); + symbol_view = symbol_view.substr(space_pos + 1); // to mangled symbol + + auto plus_pos = symbol_view.find(" + "); + if (plus_pos == absl::string_view::npos) return std::string(); + symbol_view = symbol_view.substr(0, plus_pos); // strip remainng + + return std::string(symbol_view); +} + +} // namespace +} // namespace debugging_internal + +bool Symbolize(const void* pc, char* out, int out_size) { + if (out_size <= 0 || pc == nullptr) { + out = nullptr; + return false; + } + + // This allocates a char* array. + char** frame_strings = backtrace_symbols(const_cast(&pc), 1); + + if (frame_strings == nullptr) return false; + + std::string symbol = debugging_internal::GetSymbolString(frame_strings[0]); + free(frame_strings); + + char tmp_buf[1024]; + if (debugging_internal::Demangle(symbol.c_str(), tmp_buf, sizeof(tmp_buf))) { + size_t len = strlen(tmp_buf); + if (len + 1 <= static_cast(out_size)) { // +1 for '\0' + assert(len < sizeof(tmp_buf)); + memmove(out, tmp_buf, len + 1); + } + } else { + strncpy(out, symbol.c_str(), out_size); + } + + if (out[out_size - 1] != '\0') { + // strncpy() does not '\0' terminate when it truncates. + static constexpr char kEllipsis[] = "..."; + int ellipsis_size = std::min(sizeof(kEllipsis) - 1, out_size - 1); + memcpy(out + out_size - ellipsis_size - 1, kEllipsis, ellipsis_size); + out[out_size - 1] = '\0'; + } + + return true; +} + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/symbolize_elf.inc b/src/absl/debugging/symbolize_elf.inc new file mode 100644 index 0000000..f453272 --- /dev/null +++ b/src/absl/debugging/symbolize_elf.inc @@ -0,0 +1,1613 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// This library provides Symbolize() function that symbolizes program +// counters to their corresponding symbol names on linux platforms. +// This library has a minimal implementation of an ELF symbol table +// reader (i.e. it doesn't depend on libelf, etc.). +// +// The algorithm used in Symbolize() is as follows. +// +// 1. Go through a list of maps in /proc/self/maps and find the map +// containing the program counter. +// +// 2. Open the mapped file and find a regular symbol table inside. +// Iterate over symbols in the symbol table and look for the symbol +// containing the program counter. If such a symbol is found, +// obtain the symbol name, and demangle the symbol if possible. +// If the symbol isn't found in the regular symbol table (binary is +// stripped), try the same thing with a dynamic symbol table. +// +// Note that Symbolize() is originally implemented to be used in +// signal handlers, hence it doesn't use malloc() and other unsafe +// operations. It should be both thread-safe and async-signal-safe. +// +// Implementation note: +// +// We don't use heaps but only use stacks. We want to reduce the +// stack consumption so that the symbolizer can run on small stacks. +// +// Here are some numbers collected with GCC 4.1.0 on x86: +// - sizeof(Elf32_Sym) = 16 +// - sizeof(Elf32_Shdr) = 40 +// - sizeof(Elf64_Sym) = 24 +// - sizeof(Elf64_Shdr) = 64 +// +// This implementation is intended to be async-signal-safe but uses some +// functions which are not guaranteed to be so, such as memchr() and +// memmove(). We assume they are async-signal-safe. + +#include +#include +#include +#include // For ElfW() macro. +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/casts.h" +#include "absl/base/dynamic_annotations.h" +#include "absl/base/internal/low_level_alloc.h" +#include "absl/base/internal/raw_logging.h" +#include "absl/base/internal/spinlock.h" +#include "absl/base/port.h" +#include "absl/debugging/internal/demangle.h" +#include "absl/debugging/internal/vdso_support.h" +#include "absl/strings/string_view.h" + +#if defined(__FreeBSD__) && !defined(ElfW) +#define ElfW(x) __ElfN(x) +#endif + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// Value of argv[0]. Used by MaybeInitializeObjFile(). +static char *argv0_value = nullptr; + +void InitializeSymbolizer(const char *argv0) { +#ifdef ABSL_HAVE_VDSO_SUPPORT + // We need to make sure VDSOSupport::Init() is called before any setuid or + // chroot calls, so InitializeSymbolizer() should be called very early in the + // life of a program. + absl::debugging_internal::VDSOSupport::Init(); +#endif + if (argv0_value != nullptr) { + free(argv0_value); + argv0_value = nullptr; + } + if (argv0 != nullptr && argv0[0] != '\0') { + argv0_value = strdup(argv0); + } +} + +namespace debugging_internal { +namespace { + +// Re-runs fn until it doesn't cause EINTR. +#define NO_INTR(fn) \ + do { \ + } while ((fn) < 0 && errno == EINTR) + +// On Linux, ELF_ST_* are defined in . To make this portable +// we define our own ELF_ST_BIND and ELF_ST_TYPE if not available. +#ifndef ELF_ST_BIND +#define ELF_ST_BIND(info) (((unsigned char)(info)) >> 4) +#endif + +#ifndef ELF_ST_TYPE +#define ELF_ST_TYPE(info) (((unsigned char)(info)) & 0xF) +#endif + +// Some platforms use a special .opd section to store function pointers. +const char kOpdSectionName[] = ".opd"; + +#if (defined(__powerpc__) && !(_CALL_ELF > 1)) || defined(__ia64) +// Use opd section for function descriptors on these platforms, the function +// address is the first word of the descriptor. +enum { kPlatformUsesOPDSections = 1 }; +#else // not PPC or IA64 +enum { kPlatformUsesOPDSections = 0 }; +#endif + +// This works for PowerPC & IA64 only. A function descriptor consist of two +// pointers and the first one is the function's entry. +const size_t kFunctionDescriptorSize = sizeof(void *) * 2; + +const int kMaxDecorators = 10; // Seems like a reasonable upper limit. + +struct InstalledSymbolDecorator { + SymbolDecorator fn; + void *arg; + int ticket; +}; + +int g_num_decorators; +InstalledSymbolDecorator g_decorators[kMaxDecorators]; + +struct FileMappingHint { + const void *start; + const void *end; + uint64_t offset; + const char *filename; +}; + +// Protects g_decorators. +// We are using SpinLock and not a Mutex here, because we may be called +// from inside Mutex::Lock itself, and it prohibits recursive calls. +// This happens in e.g. base/stacktrace_syscall_unittest. +// Moreover, we are using only TryLock(), if the decorator list +// is being modified (is busy), we skip all decorators, and possibly +// loose some info. Sorry, that's the best we could do. +ABSL_CONST_INIT absl::base_internal::SpinLock g_decorators_mu( + absl::kConstInit, absl::base_internal::SCHEDULE_KERNEL_ONLY); + +const int kMaxFileMappingHints = 8; +int g_num_file_mapping_hints; +FileMappingHint g_file_mapping_hints[kMaxFileMappingHints]; +// Protects g_file_mapping_hints. +ABSL_CONST_INIT absl::base_internal::SpinLock g_file_mapping_mu( + absl::kConstInit, absl::base_internal::SCHEDULE_KERNEL_ONLY); + +// Async-signal-safe function to zero a buffer. +// memset() is not guaranteed to be async-signal-safe. +static void SafeMemZero(void* p, size_t size) { + unsigned char *c = static_cast(p); + while (size--) { + *c++ = 0; + } +} + +struct ObjFile { + ObjFile() + : filename(nullptr), + start_addr(nullptr), + end_addr(nullptr), + offset(0), + fd(-1), + elf_type(-1) { + SafeMemZero(&elf_header, sizeof(elf_header)); + SafeMemZero(&phdr[0], sizeof(phdr)); + } + + char *filename; + const void *start_addr; + const void *end_addr; + uint64_t offset; + + // The following fields are initialized on the first access to the + // object file. + int fd; + int elf_type; + ElfW(Ehdr) elf_header; + + // PT_LOAD program header describing executable code. + // Normally we expect just one, but SWIFT binaries have two. + std::array phdr; +}; + +// Build 4-way associative cache for symbols. Within each cache line, symbols +// are replaced in LRU order. +enum { + ASSOCIATIVITY = 4, +}; +struct SymbolCacheLine { + const void *pc[ASSOCIATIVITY]; + char *name[ASSOCIATIVITY]; + + // age[i] is incremented when a line is accessed. it's reset to zero if the + // i'th entry is read. + uint32_t age[ASSOCIATIVITY]; +}; + +// --------------------------------------------------------------- +// An async-signal-safe arena for LowLevelAlloc +static std::atomic g_sig_safe_arena; + +static base_internal::LowLevelAlloc::Arena *SigSafeArena() { + return g_sig_safe_arena.load(std::memory_order_acquire); +} + +static void InitSigSafeArena() { + if (SigSafeArena() == nullptr) { + base_internal::LowLevelAlloc::Arena *new_arena = + base_internal::LowLevelAlloc::NewArena( + base_internal::LowLevelAlloc::kAsyncSignalSafe); + base_internal::LowLevelAlloc::Arena *old_value = nullptr; + if (!g_sig_safe_arena.compare_exchange_strong(old_value, new_arena, + std::memory_order_release, + std::memory_order_relaxed)) { + // We lost a race to allocate an arena; deallocate. + base_internal::LowLevelAlloc::DeleteArena(new_arena); + } + } +} + +// --------------------------------------------------------------- +// An AddrMap is a vector of ObjFile, using SigSafeArena() for allocation. + +class AddrMap { + public: + AddrMap() : size_(0), allocated_(0), obj_(nullptr) {} + ~AddrMap() { base_internal::LowLevelAlloc::Free(obj_); } + int Size() const { return size_; } + ObjFile *At(int i) { return &obj_[i]; } + ObjFile *Add(); + void Clear(); + + private: + int size_; // count of valid elements (<= allocated_) + int allocated_; // count of allocated elements + ObjFile *obj_; // array of allocated_ elements + AddrMap(const AddrMap &) = delete; + AddrMap &operator=(const AddrMap &) = delete; +}; + +void AddrMap::Clear() { + for (int i = 0; i != size_; i++) { + At(i)->~ObjFile(); + } + size_ = 0; +} + +ObjFile *AddrMap::Add() { + if (size_ == allocated_) { + int new_allocated = allocated_ * 2 + 50; + ObjFile *new_obj_ = + static_cast(base_internal::LowLevelAlloc::AllocWithArena( + new_allocated * sizeof(*new_obj_), SigSafeArena())); + if (obj_) { + memcpy(new_obj_, obj_, allocated_ * sizeof(*new_obj_)); + base_internal::LowLevelAlloc::Free(obj_); + } + obj_ = new_obj_; + allocated_ = new_allocated; + } + return new (&obj_[size_++]) ObjFile; +} + +// --------------------------------------------------------------- + +enum FindSymbolResult { SYMBOL_NOT_FOUND = 1, SYMBOL_TRUNCATED, SYMBOL_FOUND }; + +class Symbolizer { + public: + Symbolizer(); + ~Symbolizer(); + const char *GetSymbol(const void *const pc); + + private: + char *CopyString(const char *s) { + int len = strlen(s); + char *dst = static_cast( + base_internal::LowLevelAlloc::AllocWithArena(len + 1, SigSafeArena())); + ABSL_RAW_CHECK(dst != nullptr, "out of memory"); + memcpy(dst, s, len + 1); + return dst; + } + ObjFile *FindObjFile(const void *const start, + size_t size) ABSL_ATTRIBUTE_NOINLINE; + static bool RegisterObjFile(const char *filename, + const void *const start_addr, + const void *const end_addr, uint64_t offset, + void *arg); + SymbolCacheLine *GetCacheLine(const void *const pc); + const char *FindSymbolInCache(const void *const pc); + const char *InsertSymbolInCache(const void *const pc, const char *name); + void AgeSymbols(SymbolCacheLine *line); + void ClearAddrMap(); + FindSymbolResult GetSymbolFromObjectFile(const ObjFile &obj, + const void *const pc, + const ptrdiff_t relocation, + char *out, int out_size, + char *tmp_buf, int tmp_buf_size); + const char *GetUncachedSymbol(const void *pc); + + enum { + SYMBOL_BUF_SIZE = 3072, + TMP_BUF_SIZE = 1024, + SYMBOL_CACHE_LINES = 128, + }; + + AddrMap addr_map_; + + bool ok_; + bool addr_map_read_; + + char symbol_buf_[SYMBOL_BUF_SIZE]; + + // tmp_buf_ will be used to store arrays of ElfW(Shdr) and ElfW(Sym) + // so we ensure that tmp_buf_ is properly aligned to store either. + alignas(16) char tmp_buf_[TMP_BUF_SIZE]; + static_assert(alignof(ElfW(Shdr)) <= 16, + "alignment of tmp buf too small for Shdr"); + static_assert(alignof(ElfW(Sym)) <= 16, + "alignment of tmp buf too small for Sym"); + + SymbolCacheLine symbol_cache_[SYMBOL_CACHE_LINES]; +}; + +static std::atomic g_cached_symbolizer; + +} // namespace + +static int SymbolizerSize() { +#if defined(__wasm__) || defined(__asmjs__) + int pagesize = getpagesize(); +#else + int pagesize = sysconf(_SC_PAGESIZE); +#endif + return ((sizeof(Symbolizer) - 1) / pagesize + 1) * pagesize; +} + +// Return (and set null) g_cached_symbolized_state if it is not null. +// Otherwise return a new symbolizer. +static Symbolizer *AllocateSymbolizer() { + InitSigSafeArena(); + Symbolizer *symbolizer = + g_cached_symbolizer.exchange(nullptr, std::memory_order_acquire); + if (symbolizer != nullptr) { + return symbolizer; + } + return new (base_internal::LowLevelAlloc::AllocWithArena( + SymbolizerSize(), SigSafeArena())) Symbolizer(); +} + +// Set g_cached_symbolize_state to s if it is null, otherwise +// delete s. +static void FreeSymbolizer(Symbolizer *s) { + Symbolizer *old_cached_symbolizer = nullptr; + if (!g_cached_symbolizer.compare_exchange_strong(old_cached_symbolizer, s, + std::memory_order_release, + std::memory_order_relaxed)) { + s->~Symbolizer(); + base_internal::LowLevelAlloc::Free(s); + } +} + +Symbolizer::Symbolizer() : ok_(true), addr_map_read_(false) { + for (SymbolCacheLine &symbol_cache_line : symbol_cache_) { + for (size_t j = 0; j < ABSL_ARRAYSIZE(symbol_cache_line.name); ++j) { + symbol_cache_line.pc[j] = nullptr; + symbol_cache_line.name[j] = nullptr; + symbol_cache_line.age[j] = 0; + } + } +} + +Symbolizer::~Symbolizer() { + for (SymbolCacheLine &symbol_cache_line : symbol_cache_) { + for (char *s : symbol_cache_line.name) { + base_internal::LowLevelAlloc::Free(s); + } + } + ClearAddrMap(); +} + +// We don't use assert() since it's not guaranteed to be +// async-signal-safe. Instead we define a minimal assertion +// macro. So far, we don't need pretty printing for __FILE__, etc. +#define SAFE_ASSERT(expr) ((expr) ? static_cast(0) : throw std::runtime_error("abort()")) + +// Read up to "count" bytes from file descriptor "fd" into the buffer +// starting at "buf" while handling short reads and EINTR. On +// success, return the number of bytes read. Otherwise, return -1. +static ssize_t ReadPersistent(int fd, void *buf, size_t count) { + SAFE_ASSERT(fd >= 0); + SAFE_ASSERT(count <= SSIZE_MAX); + char *buf0 = reinterpret_cast(buf); + size_t num_bytes = 0; + while (num_bytes < count) { + ssize_t len; + NO_INTR(len = read(fd, buf0 + num_bytes, count - num_bytes)); + if (len < 0) { // There was an error other than EINTR. + ABSL_RAW_LOG(WARNING, "read failed: errno=%d", errno); + return -1; + } + if (len == 0) { // Reached EOF. + break; + } + num_bytes += len; + } + SAFE_ASSERT(num_bytes <= count); + return static_cast(num_bytes); +} + +// Read up to "count" bytes from "offset" in the file pointed by file +// descriptor "fd" into the buffer starting at "buf". On success, +// return the number of bytes read. Otherwise, return -1. +static ssize_t ReadFromOffset(const int fd, void *buf, const size_t count, + const off_t offset) { + off_t off = lseek(fd, offset, SEEK_SET); + if (off == (off_t)-1) { + ABSL_RAW_LOG(WARNING, "lseek(%d, %ju, SEEK_SET) failed: errno=%d", fd, + static_cast(offset), errno); + return -1; + } + return ReadPersistent(fd, buf, count); +} + +// Try reading exactly "count" bytes from "offset" bytes in a file +// pointed by "fd" into the buffer starting at "buf" while handling +// short reads and EINTR. On success, return true. Otherwise, return +// false. +static bool ReadFromOffsetExact(const int fd, void *buf, const size_t count, + const off_t offset) { + ssize_t len = ReadFromOffset(fd, buf, count, offset); + return len >= 0 && static_cast(len) == count; +} + +// Returns elf_header.e_type if the file pointed by fd is an ELF binary. +static int FileGetElfType(const int fd) { + ElfW(Ehdr) elf_header; + if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { + return -1; + } + if (memcmp(elf_header.e_ident, ELFMAG, SELFMAG) != 0) { + return -1; + } + return elf_header.e_type; +} + +// Read the section headers in the given ELF binary, and if a section +// of the specified type is found, set the output to this section header +// and return true. Otherwise, return false. +// To keep stack consumption low, we would like this function to not get +// inlined. +static ABSL_ATTRIBUTE_NOINLINE bool GetSectionHeaderByType( + const int fd, ElfW(Half) sh_num, const off_t sh_offset, ElfW(Word) type, + ElfW(Shdr) * out, char *tmp_buf, int tmp_buf_size) { + ElfW(Shdr) *buf = reinterpret_cast(tmp_buf); + const int buf_entries = tmp_buf_size / sizeof(buf[0]); + const int buf_bytes = buf_entries * sizeof(buf[0]); + + for (int i = 0; i < sh_num;) { + const ssize_t num_bytes_left = (sh_num - i) * sizeof(buf[0]); + const ssize_t num_bytes_to_read = + (buf_bytes > num_bytes_left) ? num_bytes_left : buf_bytes; + const off_t offset = sh_offset + i * sizeof(buf[0]); + const ssize_t len = ReadFromOffset(fd, buf, num_bytes_to_read, offset); + if (len % sizeof(buf[0]) != 0) { + ABSL_RAW_LOG( + WARNING, + "Reading %zd bytes from offset %ju returned %zd which is not a " + "multiple of %zu.", + num_bytes_to_read, static_cast(offset), len, + sizeof(buf[0])); + return false; + } + const ssize_t num_headers_in_buf = len / sizeof(buf[0]); + SAFE_ASSERT(num_headers_in_buf <= buf_entries); + for (int j = 0; j < num_headers_in_buf; ++j) { + if (buf[j].sh_type == type) { + *out = buf[j]; + return true; + } + } + i += num_headers_in_buf; + } + return false; +} + +// There is no particular reason to limit section name to 63 characters, +// but there has (as yet) been no need for anything longer either. +const int kMaxSectionNameLen = 64; + +bool ForEachSection(int fd, + const std::function &callback) { + ElfW(Ehdr) elf_header; + if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { + return false; + } + + ElfW(Shdr) shstrtab; + off_t shstrtab_offset = + (elf_header.e_shoff + elf_header.e_shentsize * elf_header.e_shstrndx); + if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) { + return false; + } + + for (int i = 0; i < elf_header.e_shnum; ++i) { + ElfW(Shdr) out; + off_t section_header_offset = + (elf_header.e_shoff + elf_header.e_shentsize * i); + if (!ReadFromOffsetExact(fd, &out, sizeof(out), section_header_offset)) { + return false; + } + off_t name_offset = shstrtab.sh_offset + out.sh_name; + char header_name[kMaxSectionNameLen]; + ssize_t n_read = + ReadFromOffset(fd, &header_name, kMaxSectionNameLen, name_offset); + if (n_read == -1) { + return false; + } else if (n_read > kMaxSectionNameLen) { + // Long read? + return false; + } + + absl::string_view name(header_name, strnlen(header_name, n_read)); + if (!callback(name, out)) { + break; + } + } + return true; +} + +// name_len should include terminating '\0'. +bool GetSectionHeaderByName(int fd, const char *name, size_t name_len, + ElfW(Shdr) * out) { + char header_name[kMaxSectionNameLen]; + if (sizeof(header_name) < name_len) { + ABSL_RAW_LOG(WARNING, + "Section name '%s' is too long (%zu); " + "section will not be found (even if present).", + name, name_len); + // No point in even trying. + return false; + } + + ElfW(Ehdr) elf_header; + if (!ReadFromOffsetExact(fd, &elf_header, sizeof(elf_header), 0)) { + return false; + } + + ElfW(Shdr) shstrtab; + off_t shstrtab_offset = + (elf_header.e_shoff + elf_header.e_shentsize * elf_header.e_shstrndx); + if (!ReadFromOffsetExact(fd, &shstrtab, sizeof(shstrtab), shstrtab_offset)) { + return false; + } + + for (int i = 0; i < elf_header.e_shnum; ++i) { + off_t section_header_offset = + (elf_header.e_shoff + elf_header.e_shentsize * i); + if (!ReadFromOffsetExact(fd, out, sizeof(*out), section_header_offset)) { + return false; + } + off_t name_offset = shstrtab.sh_offset + out->sh_name; + ssize_t n_read = ReadFromOffset(fd, &header_name, name_len, name_offset); + if (n_read < 0) { + return false; + } else if (static_cast(n_read) != name_len) { + // Short read -- name could be at end of file. + continue; + } + if (memcmp(header_name, name, name_len) == 0) { + return true; + } + } + return false; +} + +// Compare symbols at in the same address. +// Return true if we should pick symbol1. +static bool ShouldPickFirstSymbol(const ElfW(Sym) & symbol1, + const ElfW(Sym) & symbol2) { + // If one of the symbols is weak and the other is not, pick the one + // this is not a weak symbol. + char bind1 = ELF_ST_BIND(symbol1.st_info); + char bind2 = ELF_ST_BIND(symbol1.st_info); + if (bind1 == STB_WEAK && bind2 != STB_WEAK) return false; + if (bind2 == STB_WEAK && bind1 != STB_WEAK) return true; + + // If one of the symbols has zero size and the other is not, pick the + // one that has non-zero size. + if (symbol1.st_size != 0 && symbol2.st_size == 0) { + return true; + } + if (symbol1.st_size == 0 && symbol2.st_size != 0) { + return false; + } + + // If one of the symbols has no type and the other is not, pick the + // one that has a type. + char type1 = ELF_ST_TYPE(symbol1.st_info); + char type2 = ELF_ST_TYPE(symbol1.st_info); + if (type1 != STT_NOTYPE && type2 == STT_NOTYPE) { + return true; + } + if (type1 == STT_NOTYPE && type2 != STT_NOTYPE) { + return false; + } + + // Pick the first one, if we still cannot decide. + return true; +} + +// Return true if an address is inside a section. +static bool InSection(const void *address, const ElfW(Shdr) * section) { + const char *start = reinterpret_cast(section->sh_addr); + size_t size = static_cast(section->sh_size); + return start <= address && address < (start + size); +} + +static const char *ComputeOffset(const char *base, ptrdiff_t offset) { + // Note: cast to uintptr_t to avoid undefined behavior when base evaluates to + // zero and offset is non-zero. + return reinterpret_cast( + reinterpret_cast(base) + offset); +} + +// Read a symbol table and look for the symbol containing the +// pc. Iterate over symbols in a symbol table and look for the symbol +// containing "pc". If the symbol is found, and its name fits in +// out_size, the name is written into out and SYMBOL_FOUND is returned. +// If the name does not fit, truncated name is written into out, +// and SYMBOL_TRUNCATED is returned. Out is NUL-terminated. +// If the symbol is not found, SYMBOL_NOT_FOUND is returned; +// To keep stack consumption low, we would like this function to not get +// inlined. +static ABSL_ATTRIBUTE_NOINLINE FindSymbolResult FindSymbol( + const void *const pc, const int fd, char *out, int out_size, + ptrdiff_t relocation, const ElfW(Shdr) * strtab, const ElfW(Shdr) * symtab, + const ElfW(Shdr) * opd, char *tmp_buf, int tmp_buf_size) { + if (symtab == nullptr) { + return SYMBOL_NOT_FOUND; + } + + // Read multiple symbols at once to save read() calls. + ElfW(Sym) *buf = reinterpret_cast(tmp_buf); + const int buf_entries = tmp_buf_size / sizeof(buf[0]); + + const int num_symbols = symtab->sh_size / symtab->sh_entsize; + + // On platforms using an .opd section (PowerPC & IA64), a function symbol + // has the address of a function descriptor, which contains the real + // starting address. However, we do not always want to use the real + // starting address because we sometimes want to symbolize a function + // pointer into the .opd section, e.g. FindSymbol(&foo,...). + const bool pc_in_opd = + kPlatformUsesOPDSections && opd != nullptr && InSection(pc, opd); + const bool deref_function_descriptor_pointer = + kPlatformUsesOPDSections && opd != nullptr && !pc_in_opd; + + ElfW(Sym) best_match; + SafeMemZero(&best_match, sizeof(best_match)); + bool found_match = false; + for (int i = 0; i < num_symbols;) { + off_t offset = symtab->sh_offset + i * symtab->sh_entsize; + const int num_remaining_symbols = num_symbols - i; + const int entries_in_chunk = std::min(num_remaining_symbols, buf_entries); + const int bytes_in_chunk = entries_in_chunk * sizeof(buf[0]); + const ssize_t len = ReadFromOffset(fd, buf, bytes_in_chunk, offset); + SAFE_ASSERT(len % sizeof(buf[0]) == 0); + const ssize_t num_symbols_in_buf = len / sizeof(buf[0]); + SAFE_ASSERT(num_symbols_in_buf <= entries_in_chunk); + for (int j = 0; j < num_symbols_in_buf; ++j) { + const ElfW(Sym) &symbol = buf[j]; + + // For a DSO, a symbol address is relocated by the loading address. + // We keep the original address for opd redirection below. + const char *const original_start_address = + reinterpret_cast(symbol.st_value); + const char *start_address = + ComputeOffset(original_start_address, relocation); + +#ifdef __arm__ + // ARM functions are always aligned to multiples of two bytes; the + // lowest-order bit in start_address is ignored by the CPU and indicates + // whether the function contains ARM (0) or Thumb (1) code. We don't care + // about what encoding is being used; we just want the real start address + // of the function. + start_address = reinterpret_cast( + reinterpret_cast(start_address) & ~1); +#endif + + if (deref_function_descriptor_pointer && + InSection(original_start_address, opd)) { + // The opd section is mapped into memory. Just dereference + // start_address to get the first double word, which points to the + // function entry. + start_address = *reinterpret_cast(start_address); + } + + // If pc is inside the .opd section, it points to a function descriptor. + const size_t size = pc_in_opd ? kFunctionDescriptorSize : symbol.st_size; + const void *const end_address = ComputeOffset(start_address, size); + if (symbol.st_value != 0 && // Skip null value symbols. + symbol.st_shndx != 0 && // Skip undefined symbols. +#ifdef STT_TLS + ELF_ST_TYPE(symbol.st_info) != STT_TLS && // Skip thread-local data. +#endif // STT_TLS + ((start_address <= pc && pc < end_address) || + (start_address == pc && pc == end_address))) { + if (!found_match || ShouldPickFirstSymbol(symbol, best_match)) { + found_match = true; + best_match = symbol; + } + } + } + i += num_symbols_in_buf; + } + + if (found_match) { + const size_t off = strtab->sh_offset + best_match.st_name; + const ssize_t n_read = ReadFromOffset(fd, out, out_size, off); + if (n_read <= 0) { + // This should never happen. + ABSL_RAW_LOG(WARNING, + "Unable to read from fd %d at offset %zu: n_read = %zd", fd, + off, n_read); + return SYMBOL_NOT_FOUND; + } + ABSL_RAW_CHECK(n_read <= out_size, "ReadFromOffset read too much data."); + + // strtab->sh_offset points into .strtab-like section that contains + // NUL-terminated strings: '\0foo\0barbaz\0...". + // + // sh_offset+st_name points to the start of symbol name, but we don't know + // how long the symbol is, so we try to read as much as we have space for, + // and usually over-read (i.e. there is a NUL somewhere before n_read). + if (memchr(out, '\0', n_read) == nullptr) { + // Either out_size was too small (n_read == out_size and no NUL), or + // we tried to read past the EOF (n_read < out_size) and .strtab is + // corrupt (missing terminating NUL; should never happen for valid ELF). + out[n_read - 1] = '\0'; + return SYMBOL_TRUNCATED; + } + return SYMBOL_FOUND; + } + + return SYMBOL_NOT_FOUND; +} + +// Get the symbol name of "pc" from the file pointed by "fd". Process +// both regular and dynamic symbol tables if necessary. +// See FindSymbol() comment for description of return value. +FindSymbolResult Symbolizer::GetSymbolFromObjectFile( + const ObjFile &obj, const void *const pc, const ptrdiff_t relocation, + char *out, int out_size, char *tmp_buf, int tmp_buf_size) { + ElfW(Shdr) symtab; + ElfW(Shdr) strtab; + ElfW(Shdr) opd; + ElfW(Shdr) *opd_ptr = nullptr; + + // On platforms using an .opd sections for function descriptor, read + // the section header. The .opd section is in data segment and should be + // loaded but we check that it is mapped just to be extra careful. + if (kPlatformUsesOPDSections) { + if (GetSectionHeaderByName(obj.fd, kOpdSectionName, + sizeof(kOpdSectionName) - 1, &opd) && + FindObjFile(reinterpret_cast(opd.sh_addr) + relocation, + opd.sh_size) != nullptr) { + opd_ptr = &opd; + } else { + return SYMBOL_NOT_FOUND; + } + } + + // Consult a regular symbol table, then fall back to the dynamic symbol table. + for (const auto symbol_table_type : {SHT_SYMTAB, SHT_DYNSYM}) { + if (!GetSectionHeaderByType(obj.fd, obj.elf_header.e_shnum, + obj.elf_header.e_shoff, symbol_table_type, + &symtab, tmp_buf, tmp_buf_size)) { + continue; + } + if (!ReadFromOffsetExact( + obj.fd, &strtab, sizeof(strtab), + obj.elf_header.e_shoff + symtab.sh_link * sizeof(symtab))) { + continue; + } + const FindSymbolResult rc = + FindSymbol(pc, obj.fd, out, out_size, relocation, &strtab, &symtab, + opd_ptr, tmp_buf, tmp_buf_size); + if (rc != SYMBOL_NOT_FOUND) { + return rc; + } + } + + return SYMBOL_NOT_FOUND; +} + +namespace { +// Thin wrapper around a file descriptor so that the file descriptor +// gets closed for sure. +class FileDescriptor { + public: + explicit FileDescriptor(int fd) : fd_(fd) {} + FileDescriptor(const FileDescriptor &) = delete; + FileDescriptor &operator=(const FileDescriptor &) = delete; + + ~FileDescriptor() { + if (fd_ >= 0) { + NO_INTR(close(fd_)); + } + } + + int get() const { return fd_; } + + private: + const int fd_; +}; + +// Helper class for reading lines from file. +// +// Note: we don't use ProcMapsIterator since the object is big (it has +// a 5k array member) and uses async-unsafe functions such as sscanf() +// and snprintf(). +class LineReader { + public: + explicit LineReader(int fd, char *buf, int buf_len) + : fd_(fd), + buf_len_(buf_len), + buf_(buf), + bol_(buf), + eol_(buf), + eod_(buf) {} + + LineReader(const LineReader &) = delete; + LineReader &operator=(const LineReader &) = delete; + + // Read '\n'-terminated line from file. On success, modify "bol" + // and "eol", then return true. Otherwise, return false. + // + // Note: if the last line doesn't end with '\n', the line will be + // dropped. It's an intentional behavior to make the code simple. + bool ReadLine(const char **bol, const char **eol) { + if (BufferIsEmpty()) { // First time. + const ssize_t num_bytes = ReadPersistent(fd_, buf_, buf_len_); + if (num_bytes <= 0) { // EOF or error. + return false; + } + eod_ = buf_ + num_bytes; + bol_ = buf_; + } else { + bol_ = eol_ + 1; // Advance to the next line in the buffer. + SAFE_ASSERT(bol_ <= eod_); // "bol_" can point to "eod_". + if (!HasCompleteLine()) { + const int incomplete_line_length = eod_ - bol_; + // Move the trailing incomplete line to the beginning. + memmove(buf_, bol_, incomplete_line_length); + // Read text from file and append it. + char *const append_pos = buf_ + incomplete_line_length; + const int capacity_left = buf_len_ - incomplete_line_length; + const ssize_t num_bytes = + ReadPersistent(fd_, append_pos, capacity_left); + if (num_bytes <= 0) { // EOF or error. + return false; + } + eod_ = append_pos + num_bytes; + bol_ = buf_; + } + } + eol_ = FindLineFeed(); + if (eol_ == nullptr) { // '\n' not found. Malformed line. + return false; + } + *eol_ = '\0'; // Replace '\n' with '\0'. + + *bol = bol_; + *eol = eol_; + return true; + } + + private: + char *FindLineFeed() const { + return reinterpret_cast(memchr(bol_, '\n', eod_ - bol_)); + } + + bool BufferIsEmpty() const { return buf_ == eod_; } + + bool HasCompleteLine() const { + return !BufferIsEmpty() && FindLineFeed() != nullptr; + } + + const int fd_; + const int buf_len_; + char *const buf_; + char *bol_; + char *eol_; + const char *eod_; // End of data in "buf_". +}; +} // namespace + +// Place the hex number read from "start" into "*hex". The pointer to +// the first non-hex character or "end" is returned. +static const char *GetHex(const char *start, const char *end, + uint64_t *const value) { + uint64_t hex = 0; + const char *p; + for (p = start; p < end; ++p) { + int ch = *p; + if ((ch >= '0' && ch <= '9') || (ch >= 'A' && ch <= 'F') || + (ch >= 'a' && ch <= 'f')) { + hex = (hex << 4) | (ch < 'A' ? ch - '0' : (ch & 0xF) + 9); + } else { // Encountered the first non-hex character. + break; + } + } + SAFE_ASSERT(p <= end); + *value = hex; + return p; +} + +static const char *GetHex(const char *start, const char *end, + const void **const addr) { + uint64_t hex = 0; + const char *p = GetHex(start, end, &hex); + *addr = reinterpret_cast(hex); + return p; +} + +// Normally we are only interested in "r?x" maps. +// On the PowerPC, function pointers point to descriptors in the .opd +// section. The descriptors themselves are not executable code, so +// we need to relax the check below to "r??". +static bool ShouldUseMapping(const char *const flags) { + return flags[0] == 'r' && (kPlatformUsesOPDSections || flags[2] == 'x'); +} + +// Read /proc/self/maps and run "callback" for each mmapped file found. If +// "callback" returns false, stop scanning and return true. Else continue +// scanning /proc/self/maps. Return true if no parse error is found. +static ABSL_ATTRIBUTE_NOINLINE bool ReadAddrMap( + bool (*callback)(const char *filename, const void *const start_addr, + const void *const end_addr, uint64_t offset, void *arg), + void *arg, void *tmp_buf, int tmp_buf_size) { + // Use /proc/self/task//maps instead of /proc/self/maps. The latter + // requires kernel to stop all threads, and is significantly slower when there + // are 1000s of threads. + char maps_path[80]; + snprintf(maps_path, sizeof(maps_path), "/proc/self/task/%d/maps", getpid()); + + int maps_fd; + NO_INTR(maps_fd = open(maps_path, O_RDONLY)); + FileDescriptor wrapped_maps_fd(maps_fd); + if (wrapped_maps_fd.get() < 0) { + ABSL_RAW_LOG(WARNING, "%s: errno=%d", maps_path, errno); + return false; + } + + // Iterate over maps and look for the map containing the pc. Then + // look into the symbol tables inside. + LineReader reader(wrapped_maps_fd.get(), static_cast(tmp_buf), + tmp_buf_size); + while (true) { + const char *cursor; + const char *eol; + if (!reader.ReadLine(&cursor, &eol)) { // EOF or malformed line. + break; + } + + const char *line = cursor; + const void *start_address; + // Start parsing line in /proc/self/maps. Here is an example: + // + // 08048000-0804c000 r-xp 00000000 08:01 2142121 /bin/cat + // + // We want start address (08048000), end address (0804c000), flags + // (r-xp) and file name (/bin/cat). + + // Read start address. + cursor = GetHex(cursor, eol, &start_address); + if (cursor == eol || *cursor != '-') { + ABSL_RAW_LOG(WARNING, "Corrupt /proc/self/maps line: %s", line); + return false; + } + ++cursor; // Skip '-'. + + // Read end address. + const void *end_address; + cursor = GetHex(cursor, eol, &end_address); + if (cursor == eol || *cursor != ' ') { + ABSL_RAW_LOG(WARNING, "Corrupt /proc/self/maps line: %s", line); + return false; + } + ++cursor; // Skip ' '. + + // Read flags. Skip flags until we encounter a space or eol. + const char *const flags_start = cursor; + while (cursor < eol && *cursor != ' ') { + ++cursor; + } + // We expect at least four letters for flags (ex. "r-xp"). + if (cursor == eol || cursor < flags_start + 4) { + ABSL_RAW_LOG(WARNING, "Corrupt /proc/self/maps: %s", line); + return false; + } + + // Check flags. + if (!ShouldUseMapping(flags_start)) { + continue; // We skip this map. + } + ++cursor; // Skip ' '. + + // Read file offset. + uint64_t offset; + cursor = GetHex(cursor, eol, &offset); + ++cursor; // Skip ' '. + + // Skip to file name. "cursor" now points to dev. We need to skip at least + // two spaces for dev and inode. + int num_spaces = 0; + while (cursor < eol) { + if (*cursor == ' ') { + ++num_spaces; + } else if (num_spaces >= 2) { + // The first non-space character after skipping two spaces + // is the beginning of the file name. + break; + } + ++cursor; + } + + // Check whether this entry corresponds to our hint table for the true + // filename. + bool hinted = + GetFileMappingHint(&start_address, &end_address, &offset, &cursor); + if (!hinted && (cursor == eol || cursor[0] == '[')) { + // not an object file, typically [vdso] or [vsyscall] + continue; + } + if (!callback(cursor, start_address, end_address, offset, arg)) break; + } + return true; +} + +// Find the objfile mapped in address region containing [addr, addr + len). +ObjFile *Symbolizer::FindObjFile(const void *const addr, size_t len) { + for (int i = 0; i < 2; ++i) { + if (!ok_) return nullptr; + + // Read /proc/self/maps if necessary + if (!addr_map_read_) { + addr_map_read_ = true; + if (!ReadAddrMap(RegisterObjFile, this, tmp_buf_, TMP_BUF_SIZE)) { + ok_ = false; + return nullptr; + } + } + + int lo = 0; + int hi = addr_map_.Size(); + while (lo < hi) { + int mid = (lo + hi) / 2; + if (addr < addr_map_.At(mid)->end_addr) { + hi = mid; + } else { + lo = mid + 1; + } + } + if (lo != addr_map_.Size()) { + ObjFile *obj = addr_map_.At(lo); + SAFE_ASSERT(obj->end_addr > addr); + if (addr >= obj->start_addr && + reinterpret_cast(addr) + len <= obj->end_addr) + return obj; + } + + // The address mapping may have changed since it was last read. Retry. + ClearAddrMap(); + } + return nullptr; +} + +void Symbolizer::ClearAddrMap() { + for (int i = 0; i != addr_map_.Size(); i++) { + ObjFile *o = addr_map_.At(i); + base_internal::LowLevelAlloc::Free(o->filename); + if (o->fd >= 0) { + NO_INTR(close(o->fd)); + } + } + addr_map_.Clear(); + addr_map_read_ = false; +} + +// Callback for ReadAddrMap to register objfiles in an in-memory table. +bool Symbolizer::RegisterObjFile(const char *filename, + const void *const start_addr, + const void *const end_addr, uint64_t offset, + void *arg) { + Symbolizer *impl = static_cast(arg); + + // Files are supposed to be added in the increasing address order. Make + // sure that's the case. + int addr_map_size = impl->addr_map_.Size(); + if (addr_map_size != 0) { + ObjFile *old = impl->addr_map_.At(addr_map_size - 1); + if (old->end_addr > end_addr) { + ABSL_RAW_LOG(ERROR, + "Unsorted addr map entry: 0x%" PRIxPTR ": %s <-> 0x%" PRIxPTR + ": %s", + reinterpret_cast(end_addr), filename, + reinterpret_cast(old->end_addr), old->filename); + return true; + } else if (old->end_addr == end_addr) { + // The same entry appears twice. This sometimes happens for [vdso]. + if (old->start_addr != start_addr || + strcmp(old->filename, filename) != 0) { + ABSL_RAW_LOG(ERROR, + "Duplicate addr 0x%" PRIxPTR ": %s <-> 0x%" PRIxPTR ": %s", + reinterpret_cast(end_addr), filename, + reinterpret_cast(old->end_addr), old->filename); + } + return true; + } else if (old->end_addr == start_addr && + reinterpret_cast(old->start_addr) - old->offset == + reinterpret_cast(start_addr) - offset && + strcmp(old->filename, filename) == 0) { + // Two contiguous map entries that span a contiguous region of the file, + // perhaps because some part of the file was mlock()ed. Combine them. + old->end_addr = end_addr; + return true; + } + } + ObjFile *obj = impl->addr_map_.Add(); + obj->filename = impl->CopyString(filename); + obj->start_addr = start_addr; + obj->end_addr = end_addr; + obj->offset = offset; + obj->elf_type = -1; // filled on demand + obj->fd = -1; // opened on demand + return true; +} + +// This function wraps the Demangle function to provide an interface +// where the input symbol is demangled in-place. +// To keep stack consumption low, we would like this function to not +// get inlined. +static ABSL_ATTRIBUTE_NOINLINE void DemangleInplace(char *out, int out_size, + char *tmp_buf, + int tmp_buf_size) { + if (Demangle(out, tmp_buf, tmp_buf_size)) { + // Demangling succeeded. Copy to out if the space allows. + int len = strlen(tmp_buf); + if (len + 1 <= out_size) { // +1 for '\0'. + SAFE_ASSERT(len < tmp_buf_size); + memmove(out, tmp_buf, len + 1); + } + } +} + +SymbolCacheLine *Symbolizer::GetCacheLine(const void *const pc) { + uintptr_t pc0 = reinterpret_cast(pc); + pc0 >>= 3; // drop the low 3 bits + + // Shuffle bits. + pc0 ^= (pc0 >> 6) ^ (pc0 >> 12) ^ (pc0 >> 18); + return &symbol_cache_[pc0 % SYMBOL_CACHE_LINES]; +} + +void Symbolizer::AgeSymbols(SymbolCacheLine *line) { + for (uint32_t &age : line->age) { + ++age; + } +} + +const char *Symbolizer::FindSymbolInCache(const void *const pc) { + if (pc == nullptr) return nullptr; + + SymbolCacheLine *line = GetCacheLine(pc); + for (size_t i = 0; i < ABSL_ARRAYSIZE(line->pc); ++i) { + if (line->pc[i] == pc) { + AgeSymbols(line); + line->age[i] = 0; + return line->name[i]; + } + } + return nullptr; +} + +const char *Symbolizer::InsertSymbolInCache(const void *const pc, + const char *name) { + SAFE_ASSERT(pc != nullptr); + + SymbolCacheLine *line = GetCacheLine(pc); + uint32_t max_age = 0; + int oldest_index = -1; + for (size_t i = 0; i < ABSL_ARRAYSIZE(line->pc); ++i) { + if (line->pc[i] == nullptr) { + AgeSymbols(line); + line->pc[i] = pc; + line->name[i] = CopyString(name); + line->age[i] = 0; + return line->name[i]; + } + if (line->age[i] >= max_age) { + max_age = line->age[i]; + oldest_index = i; + } + } + + AgeSymbols(line); + ABSL_RAW_CHECK(oldest_index >= 0, "Corrupt cache"); + base_internal::LowLevelAlloc::Free(line->name[oldest_index]); + line->pc[oldest_index] = pc; + line->name[oldest_index] = CopyString(name); + line->age[oldest_index] = 0; + return line->name[oldest_index]; +} + +static void MaybeOpenFdFromSelfExe(ObjFile *obj) { + if (memcmp(obj->start_addr, ELFMAG, SELFMAG) != 0) { + return; + } + int fd = open("/proc/self/exe", O_RDONLY); + if (fd == -1) { + return; + } + // Verify that contents of /proc/self/exe matches in-memory image of + // the binary. This can fail if the "deleted" binary is in fact not + // the main executable, or for binaries that have the first PT_LOAD + // segment smaller than 4K. We do it in four steps so that the + // buffer is smaller and we don't consume too much stack space. + const char *mem = reinterpret_cast(obj->start_addr); + for (int i = 0; i < 4; ++i) { + char buf[1024]; + ssize_t n = read(fd, buf, sizeof(buf)); + if (n != sizeof(buf) || memcmp(buf, mem, sizeof(buf)) != 0) { + close(fd); + return; + } + mem += sizeof(buf); + } + obj->fd = fd; +} + +static bool MaybeInitializeObjFile(ObjFile *obj) { + if (obj->fd < 0) { + obj->fd = open(obj->filename, O_RDONLY); + + if (obj->fd < 0) { + // Getting /proc/self/exe here means that we were hinted. + if (strcmp(obj->filename, "/proc/self/exe") == 0) { + // /proc/self/exe may be inaccessible (due to setuid, etc.), so try + // accessing the binary via argv0. + if (argv0_value != nullptr) { + obj->fd = open(argv0_value, O_RDONLY); + } + } else { + MaybeOpenFdFromSelfExe(obj); + } + } + + if (obj->fd < 0) { + ABSL_RAW_LOG(WARNING, "%s: open failed: errno=%d", obj->filename, errno); + return false; + } + obj->elf_type = FileGetElfType(obj->fd); + if (obj->elf_type < 0) { + ABSL_RAW_LOG(WARNING, "%s: wrong elf type: %d", obj->filename, + obj->elf_type); + return false; + } + + if (!ReadFromOffsetExact(obj->fd, &obj->elf_header, sizeof(obj->elf_header), + 0)) { + ABSL_RAW_LOG(WARNING, "%s: failed to read elf header", obj->filename); + return false; + } + const int phnum = obj->elf_header.e_phnum; + const int phentsize = obj->elf_header.e_phentsize; + size_t phoff = obj->elf_header.e_phoff; + size_t num_executable_load_segments = 0; + for (int j = 0; j < phnum; j++) { + ElfW(Phdr) phdr; + if (!ReadFromOffsetExact(obj->fd, &phdr, sizeof(phdr), phoff)) { + ABSL_RAW_LOG(WARNING, "%s: failed to read program header %d", + obj->filename, j); + return false; + } + phoff += phentsize; + constexpr int rx = PF_X | PF_R; + if (phdr.p_type != PT_LOAD || (phdr.p_flags & rx) != rx) { + // Not a LOAD segment, or not executable code. + continue; + } + if (num_executable_load_segments < obj->phdr.size()) { + memcpy(&obj->phdr[num_executable_load_segments++], &phdr, sizeof(phdr)); + } else { + ABSL_RAW_LOG(WARNING, "%s: too many executable LOAD segments", + obj->filename); + break; + } + } + if (num_executable_load_segments == 0) { + // This object has no "r-x" LOAD segments. That's unexpected. + ABSL_RAW_LOG(WARNING, "%s: no executable LOAD segments", obj->filename); + return false; + } + } + return true; +} + +// The implementation of our symbolization routine. If it +// successfully finds the symbol containing "pc" and obtains the +// symbol name, returns pointer to that symbol. Otherwise, returns nullptr. +// If any symbol decorators have been installed via InstallSymbolDecorator(), +// they are called here as well. +// To keep stack consumption low, we would like this function to not +// get inlined. +const char *Symbolizer::GetUncachedSymbol(const void *pc) { + ObjFile *const obj = FindObjFile(pc, 1); + ptrdiff_t relocation = 0; + int fd = -1; + if (obj != nullptr) { + if (MaybeInitializeObjFile(obj)) { + const size_t start_addr = reinterpret_cast(obj->start_addr); + if (obj->elf_type == ET_DYN && start_addr >= obj->offset) { + // This object was relocated. + // + // For obj->offset > 0, adjust the relocation since a mapping at offset + // X in the file will have a start address of [true relocation]+X. + relocation = start_addr - obj->offset; + + // Note: some binaries have multiple "rx" LOAD segments. We must + // find the right one. + ElfW(Phdr) *phdr = nullptr; + for (size_t j = 0; j < obj->phdr.size(); j++) { + ElfW(Phdr) &p = obj->phdr[j]; + if (p.p_type != PT_LOAD) { + // We only expect PT_LOADs. This must be PT_NULL that we didn't + // write over (i.e. we exhausted all interesting PT_LOADs). + ABSL_RAW_CHECK(p.p_type == PT_NULL, "unexpected p_type"); + break; + } + if (pc < reinterpret_cast(start_addr + p.p_memsz)) { + phdr = &p; + break; + } + } + if (phdr == nullptr) { + // That's unexpected. Hope for the best. + ABSL_RAW_LOG( + WARNING, + "%s: unable to find LOAD segment for pc: %p, start_addr: %zx", + obj->filename, pc, start_addr); + } else { + // Adjust relocation in case phdr.p_vaddr != 0. + // This happens for binaries linked with `lld --rosegment`, and for + // binaries linked with BFD `ld -z separate-code`. + relocation -= phdr->p_vaddr - phdr->p_offset; + } + } + + fd = obj->fd; + if (GetSymbolFromObjectFile(*obj, pc, relocation, symbol_buf_, + sizeof(symbol_buf_), tmp_buf_, + sizeof(tmp_buf_)) == SYMBOL_FOUND) { + // Only try to demangle the symbol name if it fit into symbol_buf_. + DemangleInplace(symbol_buf_, sizeof(symbol_buf_), tmp_buf_, + sizeof(tmp_buf_)); + } + } + } else { +#if ABSL_HAVE_VDSO_SUPPORT + VDSOSupport vdso; + if (vdso.IsPresent()) { + VDSOSupport::SymbolInfo symbol_info; + if (vdso.LookupSymbolByAddress(pc, &symbol_info)) { + // All VDSO symbols are known to be short. + size_t len = strlen(symbol_info.name); + ABSL_RAW_CHECK(len + 1 < sizeof(symbol_buf_), + "VDSO symbol unexpectedly long"); + memcpy(symbol_buf_, symbol_info.name, len + 1); + } + } +#endif + } + + if (g_decorators_mu.TryLock()) { + if (g_num_decorators > 0) { + SymbolDecoratorArgs decorator_args = { + pc, relocation, fd, symbol_buf_, sizeof(symbol_buf_), + tmp_buf_, sizeof(tmp_buf_), nullptr}; + for (int i = 0; i < g_num_decorators; ++i) { + decorator_args.arg = g_decorators[i].arg; + g_decorators[i].fn(&decorator_args); + } + } + g_decorators_mu.Unlock(); + } + if (symbol_buf_[0] == '\0') { + return nullptr; + } + symbol_buf_[sizeof(symbol_buf_) - 1] = '\0'; // Paranoia. + return InsertSymbolInCache(pc, symbol_buf_); +} + +const char *Symbolizer::GetSymbol(const void *pc) { + const char *entry = FindSymbolInCache(pc); + if (entry != nullptr) { + return entry; + } + symbol_buf_[0] = '\0'; + +#ifdef __hppa__ + { + // In some contexts (e.g., return addresses), PA-RISC uses the lowest two + // bits of the address to indicate the privilege level. Clear those bits + // before trying to symbolize. + const auto pc_bits = reinterpret_cast(pc); + const auto address = pc_bits & ~0x3; + entry = GetUncachedSymbol(reinterpret_cast(address)); + if (entry != nullptr) { + return entry; + } + + // In some contexts, PA-RISC also uses bit 1 of the address to indicate that + // this is a cross-DSO function pointer. Such function pointers actually + // point to a procedure label, a struct whose first 32-bit (pointer) element + // actually points to the function text. With no symbol found for this + // address so far, try interpreting it as a cross-DSO function pointer and + // see how that goes. + if (pc_bits & 0x2) { + return GetUncachedSymbol(*reinterpret_cast(address)); + } + + return nullptr; + } +#else + return GetUncachedSymbol(pc); +#endif +} + +bool RemoveAllSymbolDecorators(void) { + if (!g_decorators_mu.TryLock()) { + // Someone else is using decorators. Get out. + return false; + } + g_num_decorators = 0; + g_decorators_mu.Unlock(); + return true; +} + +bool RemoveSymbolDecorator(int ticket) { + if (!g_decorators_mu.TryLock()) { + // Someone else is using decorators. Get out. + return false; + } + for (int i = 0; i < g_num_decorators; ++i) { + if (g_decorators[i].ticket == ticket) { + while (i < g_num_decorators - 1) { + g_decorators[i] = g_decorators[i + 1]; + ++i; + } + g_num_decorators = i; + break; + } + } + g_decorators_mu.Unlock(); + return true; // Decorator is known to be removed. +} + +int InstallSymbolDecorator(SymbolDecorator decorator, void *arg) { + static int ticket = 0; + + if (!g_decorators_mu.TryLock()) { + // Someone else is using decorators. Get out. + return -2; + } + int ret = ticket; + if (g_num_decorators >= kMaxDecorators) { + ret = -1; + } else { + g_decorators[g_num_decorators] = {decorator, arg, ticket++}; + ++g_num_decorators; + } + g_decorators_mu.Unlock(); + return ret; +} + +bool RegisterFileMappingHint(const void *start, const void *end, uint64_t offset, + const char *filename) { + SAFE_ASSERT(start <= end); + SAFE_ASSERT(filename != nullptr); + + InitSigSafeArena(); + + if (!g_file_mapping_mu.TryLock()) { + return false; + } + + bool ret = true; + if (g_num_file_mapping_hints >= kMaxFileMappingHints) { + ret = false; + } else { + // TODO(ckennelly): Move this into a string copy routine. + int len = strlen(filename); + char *dst = static_cast( + base_internal::LowLevelAlloc::AllocWithArena(len + 1, SigSafeArena())); + ABSL_RAW_CHECK(dst != nullptr, "out of memory"); + memcpy(dst, filename, len + 1); + + auto &hint = g_file_mapping_hints[g_num_file_mapping_hints++]; + hint.start = start; + hint.end = end; + hint.offset = offset; + hint.filename = dst; + } + + g_file_mapping_mu.Unlock(); + return ret; +} + +bool GetFileMappingHint(const void **start, const void **end, uint64_t *offset, + const char **filename) { + if (!g_file_mapping_mu.TryLock()) { + return false; + } + bool found = false; + for (int i = 0; i < g_num_file_mapping_hints; i++) { + if (g_file_mapping_hints[i].start <= *start && + *end <= g_file_mapping_hints[i].end) { + // We assume that the start_address for the mapping is the base + // address of the ELF section, but when [start_address,end_address) is + // not strictly equal to [hint.start, hint.end), that assumption is + // invalid. + // + // This uses the hint's start address (even though hint.start is not + // necessarily equal to start_address) to ensure the correct + // relocation is computed later. + *start = g_file_mapping_hints[i].start; + *end = g_file_mapping_hints[i].end; + *offset = g_file_mapping_hints[i].offset; + *filename = g_file_mapping_hints[i].filename; + found = true; + break; + } + } + g_file_mapping_mu.Unlock(); + return found; +} + +} // namespace debugging_internal + +bool Symbolize(const void *pc, char *out, int out_size) { + // Symbolization is very slow under tsan. + ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN(); + SAFE_ASSERT(out_size >= 0); + debugging_internal::Symbolizer *s = debugging_internal::AllocateSymbolizer(); + const char *name = s->GetSymbol(pc); + bool ok = false; + if (name != nullptr && out_size > 0) { + strncpy(out, name, out_size); + ok = true; + if (out[out_size - 1] != '\0') { + // strncpy() does not '\0' terminate when it truncates. Do so, with + // trailing ellipsis. + static constexpr char kEllipsis[] = "..."; + int ellipsis_size = + std::min(implicit_cast(strlen(kEllipsis)), out_size - 1); + memcpy(out + out_size - ellipsis_size - 1, kEllipsis, ellipsis_size); + out[out_size - 1] = '\0'; + } + } + debugging_internal::FreeSymbolizer(s); + ABSL_ANNOTATE_IGNORE_READS_AND_WRITES_END(); + return ok; +} + +ABSL_NAMESPACE_END +} // namespace absl + +extern "C" bool AbslInternalGetFileMappingHint(const void **start, + const void **end, uint64_t *offset, + const char **filename) { + return absl::debugging_internal::GetFileMappingHint(start, end, offset, + filename); +} diff --git a/src/absl/debugging/symbolize_emscripten.inc b/src/absl/debugging/symbolize_emscripten.inc new file mode 100644 index 0000000..c226c45 --- /dev/null +++ b/src/absl/debugging/symbolize_emscripten.inc @@ -0,0 +1,72 @@ +// Copyright 2020 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include + +#include +#include + +#include "absl/base/internal/raw_logging.h" +#include "absl/debugging/internal/demangle.h" +#include "absl/strings/numbers.h" +#include "absl/strings/str_cat.h" +#include "absl/strings/string_view.h" + +extern "C" { +const char* emscripten_pc_get_function(const void* pc); +} + +// clang-format off +EM_JS(bool, HaveOffsetConverter, (), + { return typeof wasmOffsetConverter !== 'undefined'; }); +// clang-format on + +namespace absl { +ABSL_NAMESPACE_BEGIN + +void InitializeSymbolizer(const char*) { + if (!HaveOffsetConverter()) { + ABSL_RAW_LOG(INFO, + "Symbolization unavailable. Rebuild with -sWASM=1 " + "and -sUSE_OFFSET_CONVERTER=1."); + } +} + +bool Symbolize(const void* pc, char* out, int out_size) { + // Check if we have the offset converter necessary for pc_get_function. + // Without it, the program will abort(). + if (!HaveOffsetConverter()) { + return false; + } + const char* func_name = emscripten_pc_get_function(pc); + if (func_name == nullptr) { + return false; + } + + strncpy(out, func_name, out_size); + + if (out[out_size - 1] != '\0') { + // strncpy() does not '\0' terminate when it truncates. + static constexpr char kEllipsis[] = "..."; + int ellipsis_size = std::min(sizeof(kEllipsis) - 1, out_size - 1); + memcpy(out + out_size - ellipsis_size - 1, kEllipsis, ellipsis_size); + out[out_size - 1] = '\0'; + } + + return true; +} + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/symbolize_unimplemented.inc b/src/absl/debugging/symbolize_unimplemented.inc new file mode 100644 index 0000000..db24456 --- /dev/null +++ b/src/absl/debugging/symbolize_unimplemented.inc @@ -0,0 +1,40 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include + +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +namespace debugging_internal { + +int InstallSymbolDecorator(SymbolDecorator, void*) { return -1; } +bool RemoveSymbolDecorator(int) { return false; } +bool RemoveAllSymbolDecorators(void) { return false; } +bool RegisterFileMappingHint(const void *, const void *, uint64_t, const char *) { + return false; +} +bool GetFileMappingHint(const void **, const void **, uint64_t *, const char **) { + return false; +} + +} // namespace debugging_internal + +void InitializeSymbolizer(const char*) {} +bool Symbolize(const void *, char *, int) { return false; } + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/debugging/symbolize_win32.inc b/src/absl/debugging/symbolize_win32.inc new file mode 100644 index 0000000..c3df46f --- /dev/null +++ b/src/absl/debugging/symbolize_win32.inc @@ -0,0 +1,81 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// See "Retrieving Symbol Information by Address": +// https://msdn.microsoft.com/en-us/library/windows/desktop/ms680578(v=vs.85).aspx + +#include + +// MSVC header dbghelp.h has a warning for an ignored typedef. +#pragma warning(push) +#pragma warning(disable:4091) +#include +#pragma warning(pop) + +#pragma comment(lib, "dbghelp.lib") + +#include +#include + +#include "absl/base/internal/raw_logging.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +static HANDLE process = NULL; + +void InitializeSymbolizer(const char*) { + if (process != nullptr) { + return; + } + process = GetCurrentProcess(); + + // Symbols are not loaded until a reference is made requiring the + // symbols be loaded. This is the fastest, most efficient way to use + // the symbol handler. + SymSetOptions(SYMOPT_DEFERRED_LOADS | SYMOPT_UNDNAME); + if (!SymInitialize(process, nullptr, true)) { + // GetLastError() returns a Win32 DWORD, but we assign to + // unsigned long long to simplify the ABSL_RAW_LOG case below. The uniform + // initialization guarantees this is not a narrowing conversion. + const unsigned long long error{GetLastError()}; // NOLINT(runtime/int) + ABSL_RAW_LOG(FATAL, "SymInitialize() failed: %llu", error); + } +} + +bool Symbolize(const void* pc, char* out, int out_size) { + if (out_size <= 0) { + return false; + } + alignas(SYMBOL_INFO) char buf[sizeof(SYMBOL_INFO) + MAX_SYM_NAME]; + SYMBOL_INFO* symbol = reinterpret_cast(buf); + symbol->SizeOfStruct = sizeof(SYMBOL_INFO); + symbol->MaxNameLen = MAX_SYM_NAME; + if (!SymFromAddr(process, reinterpret_cast(pc), nullptr, symbol)) { + return false; + } + strncpy(out, symbol->Name, out_size); + if (out[out_size - 1] != '\0') { + // strncpy() does not '\0' terminate when it truncates. + static constexpr char kEllipsis[] = "..."; + int ellipsis_size = + std::min(sizeof(kEllipsis) - 1, out_size - 1); + memcpy(out + out_size - ellipsis_size - 1, kEllipsis, ellipsis_size); + out[out_size - 1] = '\0'; + } + return true; +} + +ABSL_NAMESPACE_END +} // namespace absl diff --git a/src/absl/functional/any_invocable.h b/src/absl/functional/any_invocable.h new file mode 100644 index 0000000..0c5faca --- /dev/null +++ b/src/absl/functional/any_invocable.h @@ -0,0 +1,313 @@ +// Copyright 2022 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: any_invocable.h +// ----------------------------------------------------------------------------- +// +// This header file defines an `absl::AnyInvocable` type that assumes ownership +// and wraps an object of an invocable type. (Invocable types adhere to the +// concept specified in https://en.cppreference.com/w/cpp/concepts/invocable.) +// +// In general, prefer `absl::AnyInvocable` when you need a type-erased +// function parameter that needs to take ownership of the type. +// +// NOTE: `absl::AnyInvocable` is similar to the C++23 `std::move_only_function` +// abstraction, but has a slightly different API and is not designed to be a +// drop-in replacement or C++11-compatible backfill of that type. + +#ifndef ABSL_FUNCTIONAL_ANY_INVOCABLE_H_ +#define ABSL_FUNCTIONAL_ANY_INVOCABLE_H_ + +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/functional/internal/any_invocable.h" +#include "absl/meta/type_traits.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// absl::AnyInvocable +// +// `absl::AnyInvocable` is a functional wrapper type, like `std::function`, that +// assumes ownership of an invocable object. Unlike `std::function`, an +// `absl::AnyInvocable` is more type-safe and provides the following additional +// benefits: +// +// * Properly adheres to const correctness of the underlying type +// * Is move-only so avoids concurrency problems with copied invocables and +// unnecessary copies in general. +// * Supports reference qualifiers allowing it to perform unique actions (noted +// below). +// +// `absl::AnyInvocable` is a template, and an `absl::AnyInvocable` instantiation +// may wrap any invocable object with a compatible function signature, e.g. +// having arguments and return types convertible to types matching the +// `absl::AnyInvocable` signature, and also matching any stated reference +// qualifiers, as long as that type is moveable. It therefore provides broad +// type erasure for functional objects. +// +// An `absl::AnyInvocable` is typically used as a type-erased function parameter +// for accepting various functional objects: +// +// // Define a function taking an AnyInvocable parameter. +// void my_func(absl::AnyInvocable f) { +// ... +// }; +// +// // That function can accept any invocable type: +// +// // Accept a function reference. We don't need to move a reference. +// int func1() { return 0; }; +// my_func(func1); +// +// // Accept a lambda. We use std::move here because otherwise my_func would +// // copy the lambda. +// auto lambda = []() { return 0; }; +// my_func(std::move(lambda)); +// +// // Accept a function pointer. We don't need to move a function pointer. +// func2 = &func1; +// my_func(func2); +// +// // Accept an std::function by moving it. Note that the lambda is copyable +// // (satisfying std::function requirements) and moveable (satisfying +// // absl::AnyInvocable requirements). +// std::function func6 = []() { return 0; }; +// my_func(std::move(func6)); +// +// `AnyInvocable` also properly respects `const` qualifiers, reference +// qualifiers, and the `noexcept` specification (only in C++ 17 and beyond) as +// part of the user-specified function type (e.g. +// `AnyInvocable`). These qualifiers will be applied to +// the `AnyInvocable` object's `operator()`, and the underlying invocable must +// be compatible with those qualifiers. +// +// Comparison of const and non-const function types: +// +// // Store a closure inside of `func` with the function type `int()`. +// // Note that we have made `func` itself `const`. +// const AnyInvocable func = [](){ return 0; }; +// +// func(); // Compile-error: the passed type `int()` isn't `const`. +// +// // Store a closure inside of `const_func` with the function type +// // `int() const`. +// // Note that we have also made `const_func` itself `const`. +// const AnyInvocable const_func = [](){ return 0; }; +// +// const_func(); // Fine: `int() const` is `const`. +// +// In the above example, the call `func()` would have compiled if +// `std::function` were used even though the types are not const compatible. +// This is a bug, and using `absl::AnyInvocable` properly detects that bug. +// +// In addition to affecting the signature of `operator()`, the `const` and +// reference qualifiers of the function type also appropriately constrain which +// kinds of invocable objects you are allowed to place into the `AnyInvocable` +// instance. If you specify a function type that is const-qualified, then +// anything that you attempt to put into the `AnyInvocable` must be callable on +// a `const` instance of that type. +// +// Constraint example: +// +// // Fine because the lambda is callable when `const`. +// AnyInvocable func = [=](){ return 0; }; +// +// // This is a compile-error because the lambda isn't callable when `const`. +// AnyInvocable error = [=]() mutable { return 0; }; +// +// An `&&` qualifier can be used to express that an `absl::AnyInvocable` +// instance should be invoked at most once: +// +// // Invokes `continuation` with the logical result of an operation when +// // that operation completes (common in asynchronous code). +// void CallOnCompletion(AnyInvocable continuation) { +// int result_of_foo = foo(); +// +// // `std::move` is required because the `operator()` of `continuation` is +// // rvalue-reference qualified. +// std::move(continuation)(result_of_foo); +// } +// +// Credits to Matt Calabrese (https://github.com/mattcalabrese) for the original +// implementation. +template +class AnyInvocable : private internal_any_invocable::Impl { + private: + static_assert( + std::is_function::value, + "The template argument of AnyInvocable must be a function type."); + + using Impl = internal_any_invocable::Impl; + + public: + // The return type of Sig + using result_type = typename Impl::result_type; + + // Constructors + + // Constructs the `AnyInvocable` in an empty state. + AnyInvocable() noexcept = default; + AnyInvocable(std::nullptr_t) noexcept {} // NOLINT + + // Constructs the `AnyInvocable` from an existing `AnyInvocable` by a move. + // Note that `f` is not guaranteed to be empty after move-construction, + // although it may be. + AnyInvocable(AnyInvocable&& /*f*/) noexcept = default; + + // Constructs an `AnyInvocable` from an invocable object. + // + // Upon construction, `*this` is only empty if `f` is a function pointer or + // member pointer type and is null, or if `f` is an `AnyInvocable` that is + // empty. + template ::value>> + AnyInvocable(F&& f) // NOLINT + : Impl(internal_any_invocable::ConversionConstruct(), + std::forward(f)) {} + + // Constructs an `AnyInvocable` that holds an invocable object of type `T`, + // which is constructed in-place from the given arguments. + // + // Example: + // + // AnyInvocable func( + // absl::in_place_type, arg1, arg2); + // + template ::value>> + explicit AnyInvocable(absl::in_place_type_t, Args&&... args) + : Impl(absl::in_place_type>, + std::forward(args)...) { + static_assert(std::is_same>::value, + "The explicit template argument of in_place_type is required " + "to be an unqualified object type."); + } + + // Overload of the above constructor to support list-initialization. + template &, Args...>::value>> + explicit AnyInvocable(absl::in_place_type_t, + std::initializer_list ilist, Args&&... args) + : Impl(absl::in_place_type>, ilist, + std::forward(args)...) { + static_assert(std::is_same>::value, + "The explicit template argument of in_place_type is required " + "to be an unqualified object type."); + } + + // Assignment Operators + + // Assigns an `AnyInvocable` through move-assignment. + // Note that `f` is not guaranteed to be empty after move-assignment + // although it may be. + AnyInvocable& operator=(AnyInvocable&& /*f*/) noexcept = default; + + // Assigns an `AnyInvocable` from a nullptr, clearing the `AnyInvocable`. If + // not empty, destroys the target, putting `*this` into an empty state. + AnyInvocable& operator=(std::nullptr_t) noexcept { + this->Clear(); + return *this; + } + + // Assigns an `AnyInvocable` from an existing `AnyInvocable` instance. + // + // Upon assignment, `*this` is only empty if `f` is a function pointer or + // member pointer type and is null, or if `f` is an `AnyInvocable` that is + // empty. + template ::value>> + AnyInvocable& operator=(F&& f) { + *this = AnyInvocable(std::forward(f)); + return *this; + } + + // Assigns an `AnyInvocable` from a reference to an invocable object. + // Upon assignment, stores a reference to the invocable object in the + // `AnyInvocable` instance. + template < + class F, + typename = absl::enable_if_t< + internal_any_invocable::CanAssignReferenceWrapper::value>> + AnyInvocable& operator=(std::reference_wrapper f) noexcept { + *this = AnyInvocable(f); + return *this; + } + + // Destructor + + // If not empty, destroys the target. + ~AnyInvocable() = default; + + // absl::AnyInvocable::swap() + // + // Exchanges the targets of `*this` and `other`. + void swap(AnyInvocable& other) noexcept { std::swap(*this, other); } + + // abl::AnyInvocable::operator bool() + // + // Returns `true` if `*this` is not empty. + explicit operator bool() const noexcept { return this->HasValue(); } + + // Invokes the target object of `*this`. `*this` must not be empty. + // + // Note: The signature of this function call operator is the same as the + // template parameter `Sig`. + using Impl::operator(); + + // Equality operators + + // Returns `true` if `*this` is empty. + friend bool operator==(const AnyInvocable& f, std::nullptr_t) noexcept { + return !f.HasValue(); + } + + // Returns `true` if `*this` is empty. + friend bool operator==(std::nullptr_t, const AnyInvocable& f) noexcept { + return !f.HasValue(); + } + + // Returns `false` if `*this` is empty. + friend bool operator!=(const AnyInvocable& f, std::nullptr_t) noexcept { + return f.HasValue(); + } + + // Returns `false` if `*this` is empty. + friend bool operator!=(std::nullptr_t, const AnyInvocable& f) noexcept { + return f.HasValue(); + } + + // swap() + // + // Exchanges the targets of `f1` and `f2`. + friend void swap(AnyInvocable& f1, AnyInvocable& f2) noexcept { f1.swap(f2); } + + private: + // Friending other instantiations is necessary for conversions. + template + friend class internal_any_invocable::CoreImpl; +}; + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_ANY_INVOCABLE_H_ diff --git a/src/absl/functional/bind_front.h b/src/absl/functional/bind_front.h new file mode 100644 index 0000000..f9075bd --- /dev/null +++ b/src/absl/functional/bind_front.h @@ -0,0 +1,193 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: bind_front.h +// ----------------------------------------------------------------------------- +// +// `absl::bind_front()` returns a functor by binding a number of arguments to +// the front of a provided (usually more generic) functor. Unlike `std::bind`, +// it does not require the use of argument placeholders. The simpler syntax of +// `absl::bind_front()` allows you to avoid known misuses with `std::bind()`. +// +// `absl::bind_front()` is meant as a drop-in replacement for C++20's upcoming +// `std::bind_front()`, which similarly resolves these issues with +// `std::bind()`. Both `bind_front()` alternatives, unlike `std::bind()`, allow +// partial function application. (See +// https://en.wikipedia.org/wiki/Partial_application). + +#ifndef ABSL_FUNCTIONAL_BIND_FRONT_H_ +#define ABSL_FUNCTIONAL_BIND_FRONT_H_ + +#if defined(__cpp_lib_bind_front) && __cpp_lib_bind_front >= 201907L +#include // For std::bind_front. +#endif // defined(__cpp_lib_bind_front) && __cpp_lib_bind_front >= 201907L + +#include "absl/functional/internal/front_binder.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// bind_front() +// +// Binds the first N arguments of an invocable object and stores them by value. +// +// Like `std::bind()`, `absl::bind_front()` is implicitly convertible to +// `std::function`. In particular, it may be used as a simpler replacement for +// `std::bind()` in most cases, as it does not require placeholders to be +// specified. More importantly, it provides more reliable correctness guarantees +// than `std::bind()`; while `std::bind()` will silently ignore passing more +// parameters than expected, for example, `absl::bind_front()` will report such +// mis-uses as errors. In C++20, `absl::bind_front` is replaced by +// `std::bind_front`. +// +// absl::bind_front(a...) can be seen as storing the results of +// std::make_tuple(a...). +// +// Example: Binding a free function. +// +// int Minus(int a, int b) { return a - b; } +// +// assert(absl::bind_front(Minus)(3, 2) == 3 - 2); +// assert(absl::bind_front(Minus, 3)(2) == 3 - 2); +// assert(absl::bind_front(Minus, 3, 2)() == 3 - 2); +// +// Example: Binding a member function. +// +// struct Math { +// int Double(int a) const { return 2 * a; } +// }; +// +// Math math; +// +// assert(absl::bind_front(&Math::Double)(&math, 3) == 2 * 3); +// // Stores a pointer to math inside the functor. +// assert(absl::bind_front(&Math::Double, &math)(3) == 2 * 3); +// // Stores a copy of math inside the functor. +// assert(absl::bind_front(&Math::Double, math)(3) == 2 * 3); +// // Stores std::unique_ptr inside the functor. +// assert(absl::bind_front(&Math::Double, +// std::unique_ptr(new Math))(3) == 2 * 3); +// +// Example: Using `absl::bind_front()`, instead of `std::bind()`, with +// `std::function`. +// +// class FileReader { +// public: +// void ReadFileAsync(const std::string& filename, std::string* content, +// const std::function& done) { +// // Calls Executor::Schedule(std::function). +// Executor::DefaultExecutor()->Schedule( +// absl::bind_front(&FileReader::BlockingRead, this, +// filename, content, done)); +// } +// +// private: +// void BlockingRead(const std::string& filename, std::string* content, +// const std::function& done) { +// CHECK_OK(file::GetContents(filename, content, {})); +// done(); +// } +// }; +// +// `absl::bind_front()` stores bound arguments explicitly using the type passed +// rather than implicitly based on the type accepted by its functor. +// +// Example: Binding arguments explicitly. +// +// void LogStringView(absl::string_view sv) { +// LOG(INFO) << sv; +// } +// +// Executor* e = Executor::DefaultExecutor(); +// std::string s = "hello"; +// absl::string_view sv = s; +// +// // absl::bind_front(LogStringView, arg) makes a copy of arg and stores it. +// e->Schedule(absl::bind_front(LogStringView, sv)); // ERROR: dangling +// // string_view. +// +// e->Schedule(absl::bind_front(LogStringView, s)); // OK: stores a copy of +// // s. +// +// To store some of the arguments passed to `absl::bind_front()` by reference, +// use std::ref()` and `std::cref()`. +// +// Example: Storing some of the bound arguments by reference. +// +// class Service { +// public: +// void Serve(const Request& req, std::function* done) { +// // The request protocol buffer won't be deleted until done is called. +// // It's safe to store a reference to it inside the functor. +// Executor::DefaultExecutor()->Schedule( +// absl::bind_front(&Service::BlockingServe, this, std::cref(req), +// done)); +// } +// +// private: +// void BlockingServe(const Request& req, std::function* done); +// }; +// +// Example: Storing bound arguments by reference. +// +// void Print(const std::string& a, const std::string& b) { +// std::cerr << a << b; +// } +// +// std::string hi = "Hello, "; +// std::vector names = {"Chuk", "Gek"}; +// // Doesn't copy hi. +// for_each(names.begin(), names.end(), +// absl::bind_front(Print, std::ref(hi))); +// +// // DO NOT DO THIS: the functor may outlive "hi", resulting in +// // dangling references. +// foo->DoInFuture(absl::bind_front(Print, std::ref(hi), "Guest")); // BAD! +// auto f = absl::bind_front(Print, std::ref(hi), "Guest"); // BAD! +// +// Example: Storing reference-like types. +// +// void Print(absl::string_view a, const std::string& b) { +// std::cerr << a << b; +// } +// +// std::string hi = "Hello, "; +// // Copies "hi". +// absl::bind_front(Print, hi)("Chuk"); +// +// // Compile error: std::reference_wrapper is not implicitly +// // convertible to string_view. +// // absl::bind_front(Print, std::cref(hi))("Chuk"); +// +// // Doesn't copy "hi". +// absl::bind_front(Print, absl::string_view(hi))("Chuk"); +// +#if defined(__cpp_lib_bind_front) && __cpp_lib_bind_front >= 201907L +using std::bind_front; +#else // defined(__cpp_lib_bind_front) && __cpp_lib_bind_front >= 201907L +template +constexpr functional_internal::bind_front_t bind_front( + F&& func, BoundArgs&&... args) { + return functional_internal::bind_front_t( + absl::in_place, absl::forward(func), + absl::forward(args)...); +} +#endif // defined(__cpp_lib_bind_front) && __cpp_lib_bind_front >= 201907L + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_BIND_FRONT_H_ diff --git a/src/absl/functional/function_ref.h b/src/absl/functional/function_ref.h new file mode 100644 index 0000000..f977960 --- /dev/null +++ b/src/absl/functional/function_ref.h @@ -0,0 +1,143 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: function_ref.h +// ----------------------------------------------------------------------------- +// +// This header file defines the `absl::FunctionRef` type for holding a +// non-owning reference to an object of any invocable type. This function +// reference is typically most useful as a type-erased argument type for +// accepting function types that neither take ownership nor copy the type; using +// the reference type in this case avoids a copy and an allocation. Best +// practices of other non-owning reference-like objects (such as +// `absl::string_view`) apply here. +// +// An `absl::FunctionRef` is similar in usage to a `std::function` but has the +// following differences: +// +// * It doesn't own the underlying object. +// * It doesn't have a null or empty state. +// * It never performs deep copies or allocations. +// * It's much faster and cheaper to construct. +// * It's trivially copyable and destructable. +// +// Generally, `absl::FunctionRef` should not be used as a return value, data +// member, or to initialize a `std::function`. Such usages will often lead to +// problematic lifetime issues. Once you convert something to an +// `absl::FunctionRef` you cannot make a deep copy later. +// +// This class is suitable for use wherever a "const std::function<>&" +// would be used without making a copy. ForEach functions and other versions of +// the visitor pattern are a good example of when this class should be used. +// +// This class is trivial to copy and should be passed by value. +#ifndef ABSL_FUNCTIONAL_FUNCTION_REF_H_ +#define ABSL_FUNCTIONAL_FUNCTION_REF_H_ + +#include +#include +#include + +#include "absl/base/attributes.h" +#include "absl/functional/internal/function_ref.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// FunctionRef +// +// Dummy class declaration to allow the partial specialization based on function +// types below. +template +class FunctionRef; + +// FunctionRef +// +// An `absl::FunctionRef` is a lightweight wrapper to any invokable object with +// a compatible signature. Generally, an `absl::FunctionRef` should only be used +// as an argument type and should be preferred as an argument over a const +// reference to a `std::function`. `absl::FunctionRef` itself does not allocate, +// although the wrapped invokable may. +// +// Example: +// +// // The following function takes a function callback by const reference +// bool Visitor(const std::function& callback); +// +// // Assuming that the function is not stored or otherwise copied, it can be +// // replaced by an `absl::FunctionRef`: +// bool Visitor(absl::FunctionRef +// callback); +// +// Note: the assignment operator within an `absl::FunctionRef` is intentionally +// deleted to prevent misuse; because the `absl::FunctionRef` does not own the +// underlying type, assignment likely indicates misuse. +template +class FunctionRef { + private: + // Used to disable constructors for objects that are not compatible with the + // signature of this FunctionRef. + template > + using EnableIfCompatible = + typename std::enable_if::value || + std::is_convertible::value>::type; + + public: + // Constructs a FunctionRef from any invokable type. + template > + // NOLINTNEXTLINE(runtime/explicit) + FunctionRef(const F& f ABSL_ATTRIBUTE_LIFETIME_BOUND) + : invoker_(&absl::functional_internal::InvokeObject) { + absl::functional_internal::AssertNonNull(f); + ptr_.obj = &f; + } + + // Overload for function pointers. This eliminates a level of indirection that + // would happen if the above overload was used (it lets us store the pointer + // instead of a pointer to a pointer). + // + // This overload is also used for references to functions, since references to + // functions can decay to function pointers implicitly. + template < + typename F, typename = EnableIfCompatible, + absl::functional_internal::EnableIf::value> = 0> + FunctionRef(F* f) // NOLINT(runtime/explicit) + : invoker_(&absl::functional_internal::InvokeFunction) { + assert(f != nullptr); + ptr_.fun = reinterpret_cast(f); + } + + // To help prevent subtle lifetime bugs, FunctionRef is not assignable. + // Typically, it should only be used as an argument type. + FunctionRef& operator=(const FunctionRef& rhs) = delete; + FunctionRef(const FunctionRef& rhs) = default; + + // Call the underlying object. + R operator()(Args... args) const { + return invoker_(ptr_, std::forward(args)...); + } + + private: + absl::functional_internal::VoidPtr ptr_; + absl::functional_internal::Invoker invoker_; +}; + +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_FUNCTION_REF_H_ diff --git a/src/absl/functional/internal/any_invocable.h b/src/absl/functional/internal/any_invocable.h new file mode 100644 index 0000000..0b9a621 --- /dev/null +++ b/src/absl/functional/internal/any_invocable.h @@ -0,0 +1,857 @@ +// Copyright 2022 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// Implementation details for `absl::AnyInvocable` + +#ifndef ABSL_FUNCTIONAL_INTERNAL_ANY_INVOCABLE_H_ +#define ABSL_FUNCTIONAL_INTERNAL_ANY_INVOCABLE_H_ + +//////////////////////////////////////////////////////////////////////////////// +// // +// This implementation of the proposed `any_invocable` uses an approach that // +// chooses between local storage and remote storage for the contained target // +// object based on the target object's size, alignment requirements, and // +// whether or not it has a nothrow move constructor. Additional optimizations // +// are performed when the object is a trivially copyable type [basic.types]. // +// // +// There are three datamembers per `AnyInvocable` instance // +// // +// 1) A union containing either // +// - A pointer to the target object referred to via a void*, or // +// - the target object, emplaced into a raw char buffer // +// // +// 2) A function pointer to a "manager" function operation that takes a // +// discriminator and logically branches to either perform a move operation // +// or destroy operation based on that discriminator. // +// // +// 3) A function pointer to an "invoker" function operation that invokes the // +// target object, directly returning the result. // +// // +// When in the logically empty state, the manager function is an empty // +// function and the invoker function is one that would be undefined-behavior // +// to call. // +// // +// An additional optimization is performed when converting from one // +// AnyInvocable to another where only the noexcept specification and/or the // +// cv/ref qualifiers of the function type differ. In these cases, the // +// conversion works by "moving the guts", similar to if they were the same // +// exact type, as opposed to having to perform an additional layer of // +// wrapping through remote storage. // +// // +//////////////////////////////////////////////////////////////////////////////// + +// IWYU pragma: private, include "absl/functional/any_invocable.h" + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "absl/base/config.h" +#include "absl/base/internal/invoke.h" +#include "absl/base/macros.h" +#include "absl/meta/type_traits.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// Helper macro used to prevent spelling `noexcept` in language versions older +// than C++17, where it is not part of the type system, in order to avoid +// compilation failures and internal compiler errors. +#if ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L +#define ABSL_INTERNAL_NOEXCEPT_SPEC(noex) noexcept(noex) +#else +#define ABSL_INTERNAL_NOEXCEPT_SPEC(noex) +#endif + +// Defined in functional/any_invocable.h +template +class AnyInvocable; + +namespace internal_any_invocable { + +// Constants relating to the small-object-storage for AnyInvocable +enum StorageProperty : std::size_t { + kAlignment = alignof(std::max_align_t), // The alignment of the storage + kStorageSize = sizeof(void*) * 2 // The size of the storage +}; + +//////////////////////////////////////////////////////////////////////////////// +// +// A metafunction for checking if a type is an AnyInvocable instantiation. +// This is used during conversion operations. +template +struct IsAnyInvocable : std::false_type {}; + +template +struct IsAnyInvocable> : std::true_type {}; +// +//////////////////////////////////////////////////////////////////////////////// + +// A type trait that tells us whether or not a target function type should be +// stored locally in the small object optimization storage +template +using IsStoredLocally = std::integral_constant< + bool, sizeof(T) <= kStorageSize && alignof(T) <= kAlignment && + kAlignment % alignof(T) == 0 && + std::is_nothrow_move_constructible::value>; + +// An implementation of std::remove_cvref_t of C++20. +template +using RemoveCVRef = + typename std::remove_cv::type>::type; + +//////////////////////////////////////////////////////////////////////////////// +// +// An implementation of the C++ standard INVOKE pseudo-macro, operation is +// equivalent to std::invoke except that it forces an implicit conversion to the +// specified return type. If "R" is void, the function is executed and the +// return value is simply ignored. +template ::value>> +void InvokeR(F&& f, P&&... args) { + absl::base_internal::invoke(std::forward(f), std::forward

(args)...); +} + +template ::value, int> = 0> +ReturnType InvokeR(F&& f, P&&... args) { + return absl::base_internal::invoke(std::forward(f), + std::forward

(args)...); +} + +// +//////////////////////////////////////////////////////////////////////////////// + +//////////////////////////////////////////////////////////////////////////////// +/// +// A metafunction that takes a "T" corresponding to a parameter type of the +// user's specified function type, and yields the parameter type to use for the +// type-erased invoker. In order to prevent observable moves, this must be +// either a reference or, if the type is trivial, the original parameter type +// itself. Since the parameter type may be incomplete at the point that this +// metafunction is used, we can only do this optimization for scalar types +// rather than for any trivial type. +template +T ForwardImpl(std::true_type); + +template +T&& ForwardImpl(std::false_type); + +// NOTE: We deliberately use an intermediate struct instead of a direct alias, +// as a workaround for b/206991861 on MSVC versions < 1924. +template +struct ForwardedParameter { + using type = decltype(( + ForwardImpl)(std::integral_constant::value>())); +}; + +template +using ForwardedParameterType = typename ForwardedParameter::type; +// +//////////////////////////////////////////////////////////////////////////////// + +// A discriminator when calling the "manager" function that describes operation +// type-erased operation should be invoked. +// +// "relocate_from_to" specifies that the manager should perform a move. +// +// "dispose" specifies that the manager should perform a destroy. +enum class FunctionToCall : bool { relocate_from_to, dispose }; + +// The portion of `AnyInvocable` state that contains either a pointer to the +// target object or the object itself in local storage +union TypeErasedState { + struct { + // A pointer to the type-erased object when remotely stored + void* target; + // The size of the object for `RemoteManagerTrivial` + std::size_t size; + } remote; + + // Local-storage for the type-erased object when small and trivial enough + alignas(kAlignment) char storage[kStorageSize]; +}; + +// A typed accessor for the object in `TypeErasedState` storage +template +T& ObjectInLocalStorage(TypeErasedState* const state) { + // We launder here because the storage may be reused with the same type. +#if ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L + return *std::launder(reinterpret_cast(&state->storage)); +#elif ABSL_HAVE_BUILTIN(__builtin_launder) + return *__builtin_launder(reinterpret_cast(&state->storage)); +#else + + // When `std::launder` or equivalent are not available, we rely on undefined + // behavior, which works as intended on Abseil's officially supported + // platforms as of Q2 2022. +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic ignored "-Wstrict-aliasing" +// #pragma GCC diagnostic push +#endif + return *reinterpret_cast(&state->storage); +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic pop +#endif + +#endif +} + +// The type for functions issuing lifetime-related operations: move and dispose +// A pointer to such a function is contained in each `AnyInvocable` instance. +// NOTE: When specifying `FunctionToCall::`dispose, the same state must be +// passed as both "from" and "to". +using ManagerType = void(FunctionToCall /*operation*/, + TypeErasedState* /*from*/, TypeErasedState* /*to*/) + ABSL_INTERNAL_NOEXCEPT_SPEC(true); + +// The type for functions issuing the actual invocation of the object +// A pointer to such a function is contained in each AnyInvocable instance. +template +using InvokerType = ReturnType(TypeErasedState*, ForwardedParameterType

...) + ABSL_INTERNAL_NOEXCEPT_SPEC(SigIsNoexcept); + +// The manager that is used when AnyInvocable is empty +inline void EmptyManager(FunctionToCall /*operation*/, + TypeErasedState* /*from*/, + TypeErasedState* /*to*/) noexcept {} + +// The manager that is used when a target function is in local storage and is +// a trivially copyable type. +inline void LocalManagerTrivial(FunctionToCall /*operation*/, + TypeErasedState* const from, + TypeErasedState* const to) noexcept { + // This single statement without branching handles both possible operations. + // + // For FunctionToCall::dispose, "from" and "to" point to the same state, and + // so this assignment logically would do nothing. + // + // Note: Correctness here relies on http://wg21.link/p0593, which has only + // become standard in C++20, though implementations do not break it in + // practice for earlier versions of C++. + // + // The correct way to do this without that paper is to first placement-new a + // default-constructed T in "to->storage" prior to the memmove, but doing so + // requires a different function to be created for each T that is stored + // locally, which can cause unnecessary bloat and be less cache friendly. + *to = *from; + + // Note: Because the type is trivially copyable, the destructor does not need + // to be called ("trivially copyable" requires a trivial destructor). +} + +// The manager that is used when a target function is in local storage and is +// not a trivially copyable type. +template +void LocalManagerNontrivial(FunctionToCall operation, + TypeErasedState* const from, + TypeErasedState* const to) noexcept { + static_assert(IsStoredLocally::value, + "Local storage must only be used for supported types."); + static_assert(!std::is_trivially_copyable::value, + "Locally stored types must be trivially copyable."); + + T& from_object = (ObjectInLocalStorage)(from); + + switch (operation) { + case FunctionToCall::relocate_from_to: + // NOTE: Requires that the left-hand operand is already empty. + ::new (static_cast(&to->storage)) T(std::move(from_object)); + ABSL_FALLTHROUGH_INTENDED; + case FunctionToCall::dispose: + from_object.~T(); // Must not throw. // NOLINT + return; + } + ABSL_INTERNAL_UNREACHABLE; +} + +// The invoker that is used when a target function is in local storage +// Note: QualTRef here is the target function type along with cv and reference +// qualifiers that must be used when calling the function. +template +ReturnType LocalInvoker( + TypeErasedState* const state, + ForwardedParameterType

... args) noexcept(SigIsNoexcept) { + using RawT = RemoveCVRef; + static_assert( + IsStoredLocally::value, + "Target object must be in local storage in order to be invoked from it."); + + auto& f = (ObjectInLocalStorage)(state); + return (InvokeR)(static_cast(f), + static_cast>(args)...); +} + +// The manager that is used when a target function is in remote storage and it +// has a trivial destructor +inline void RemoteManagerTrivial(FunctionToCall operation, + TypeErasedState* const from, + TypeErasedState* const to) noexcept { + switch (operation) { + case FunctionToCall::relocate_from_to: + // NOTE: Requires that the left-hand operand is already empty. + to->remote = from->remote; + return; + case FunctionToCall::dispose: +#if defined(__cpp_sized_deallocation) + ::operator delete(from->remote.target, from->remote.size); +#else // __cpp_sized_deallocation + ::operator delete(from->remote.target); +#endif // __cpp_sized_deallocation + return; + } + ABSL_INTERNAL_UNREACHABLE; +} + +// The manager that is used when a target function is in remote storage and the +// destructor of the type is not trivial +template +void RemoteManagerNontrivial(FunctionToCall operation, + TypeErasedState* const from, + TypeErasedState* const to) noexcept { + static_assert(!IsStoredLocally::value, + "Remote storage must only be used for types that do not " + "qualify for local storage."); + + switch (operation) { + case FunctionToCall::relocate_from_to: + // NOTE: Requires that the left-hand operand is already empty. + to->remote.target = from->remote.target; + return; + case FunctionToCall::dispose: + ::delete static_cast(from->remote.target); // Must not throw. + return; + } + ABSL_INTERNAL_UNREACHABLE; +} + +// The invoker that is used when a target function is in remote storage +template +ReturnType RemoteInvoker( + TypeErasedState* const state, + ForwardedParameterType

... args) noexcept(SigIsNoexcept) { + using RawT = RemoveCVRef; + static_assert(!IsStoredLocally::value, + "Target object must be in remote storage in order to be " + "invoked from it."); + + auto& f = *static_cast(state->remote.target); + return (InvokeR)(static_cast(f), + static_cast>(args)...); +} + +//////////////////////////////////////////////////////////////////////////////// +// +// A metafunction that checks if a type T is an instantiation of +// absl::in_place_type_t (needed for constructor constraints of AnyInvocable). +template +struct IsInPlaceType : std::false_type {}; + +template +struct IsInPlaceType> : std::true_type {}; +// +//////////////////////////////////////////////////////////////////////////////// + +// A constructor name-tag used with CoreImpl (below) to request the +// conversion-constructor. QualDecayedTRef is the decayed-type of the object to +// wrap, along with the cv and reference qualifiers that must be applied when +// performing an invocation of the wrapped object. +template +struct TypedConversionConstruct {}; + +// A helper base class for all core operations of AnyInvocable. Most notably, +// this class creates the function call operator and constraint-checkers so that +// the top-level class does not have to be a series of partial specializations. +// +// Note: This definition exists (as opposed to being a declaration) so that if +// the user of the top-level template accidentally passes a template argument +// that is not a function type, they will get a static_assert in AnyInvocable's +// class body rather than an error stating that Impl is not defined. +template +class Impl {}; // Note: This is partially-specialized later. + +// A std::unique_ptr deleter that deletes memory allocated via ::operator new. +#if defined(__cpp_sized_deallocation) +class TrivialDeleter { + public: + explicit TrivialDeleter(std::size_t size) : size_(size) {} + + void operator()(void* target) const { + ::operator delete(target, size_); + } + + private: + std::size_t size_; +}; +#else // __cpp_sized_deallocation +class TrivialDeleter { + public: + explicit TrivialDeleter(std::size_t) {} + + void operator()(void* target) const { ::operator delete(target); } +}; +#endif // __cpp_sized_deallocation + +template +class CoreImpl; + +constexpr bool IsCompatibleConversion(void*, void*) { return false; } +template +constexpr bool IsCompatibleConversion(CoreImpl*, + CoreImpl*) { + return !NoExceptDest || NoExceptSrc; +} + +// A helper base class for all core operations of AnyInvocable that do not +// depend on the cv/ref qualifiers of the function type. +template +class CoreImpl { + public: + using result_type = ReturnType; + + CoreImpl() noexcept : manager_(EmptyManager), invoker_(nullptr) {} + + enum class TargetType : int { + kPointer = 0, + kCompatibleAnyInvocable = 1, + kIncompatibleAnyInvocable = 2, + kOther = 3, + }; + + // Note: QualDecayedTRef here includes the cv-ref qualifiers associated with + // the invocation of the Invocable. The unqualified type is the target object + // type to be stored. + template + explicit CoreImpl(TypedConversionConstruct, F&& f) { + using DecayedT = RemoveCVRef; + + constexpr TargetType kTargetType = + (std::is_pointer::value || + std::is_member_pointer::value) + ? TargetType::kPointer + : IsCompatibleAnyInvocable::value + ? TargetType::kCompatibleAnyInvocable + : IsAnyInvocable::value + ? TargetType::kIncompatibleAnyInvocable + : TargetType::kOther; + // NOTE: We only use integers instead of enums as template parameters in + // order to work around a bug on C++14 under MSVC 2017. + // See b/236131881. + Initialize(kTargetType), QualDecayedTRef>( + std::forward(f)); + } + + // Note: QualTRef here includes the cv-ref qualifiers associated with the + // invocation of the Invocable. The unqualified type is the target object + // type to be stored. + template + explicit CoreImpl(absl::in_place_type_t, Args&&... args) { + InitializeStorage(std::forward(args)...); + } + + CoreImpl(CoreImpl&& other) noexcept { + other.manager_(FunctionToCall::relocate_from_to, &other.state_, &state_); + manager_ = other.manager_; + invoker_ = other.invoker_; + other.manager_ = EmptyManager; + other.invoker_ = nullptr; + } + + CoreImpl& operator=(CoreImpl&& other) noexcept { + // Put the left-hand operand in an empty state. + // + // Note: A full reset that leaves us with an object that has its invariants + // intact is necessary in order to handle self-move. This is required by + // types that are used with certain operations of the standard library, such + // as the default definition of std::swap when both operands target the same + // object. + Clear(); + + // Perform the actual move/destory operation on the target function. + other.manager_(FunctionToCall::relocate_from_to, &other.state_, &state_); + manager_ = other.manager_; + invoker_ = other.invoker_; + other.manager_ = EmptyManager; + other.invoker_ = nullptr; + + return *this; + } + + ~CoreImpl() { manager_(FunctionToCall::dispose, &state_, &state_); } + + // Check whether or not the AnyInvocable is in the empty state. + bool HasValue() const { return invoker_ != nullptr; } + + // Effects: Puts the object into its empty state. + void Clear() { + manager_(FunctionToCall::dispose, &state_, &state_); + manager_ = EmptyManager; + invoker_ = nullptr; + } + + template = 0> + void Initialize(F&& f) { +// This condition handles types that decay into pointers, which includes +// function references. Since function references cannot be null, GCC warns +// against comparing their decayed form with nullptr. +// Since this is template-heavy code, we prefer to disable these warnings +// locally instead of adding yet another overload of this function. +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic ignored "-Wpragmas" +// #pragma GCC diagnostic ignored "-Waddress" +// #pragma GCC diagnostic ignored "-Wnonnull-compare" +// #pragma GCC diagnostic push +#endif + if (static_cast>(f) == nullptr) { +#if !defined(__clang__) && defined(__GNUC__) +// #pragma GCC diagnostic pop +#endif + manager_ = EmptyManager; + invoker_ = nullptr; + return; + } + InitializeStorage(std::forward(f)); + } + + template = 0> + void Initialize(F&& f) { + // In this case we can "steal the guts" of the other AnyInvocable. + f.manager_(FunctionToCall::relocate_from_to, &f.state_, &state_); + manager_ = f.manager_; + invoker_ = f.invoker_; + + f.manager_ = EmptyManager; + f.invoker_ = nullptr; + } + + template = 0> + void Initialize(F&& f) { + if (f.HasValue()) { + InitializeStorage(std::forward(f)); + } else { + manager_ = EmptyManager; + invoker_ = nullptr; + } + } + + template > + void Initialize(F&& f) { + InitializeStorage(std::forward(f)); + } + + // Use local (inline) storage for applicable target object types. + template >::value>> + void InitializeStorage(Args&&... args) { + using RawT = RemoveCVRef; + ::new (static_cast(&state_.storage)) + RawT(std::forward(args)...); + + invoker_ = LocalInvoker; + // We can simplify our manager if we know the type is trivially copyable. + InitializeLocalManager(); + } + + // Use remote storage for target objects that cannot be stored locally. + template >::value, + int> = 0> + void InitializeStorage(Args&&... args) { + InitializeRemoteManager>(std::forward(args)...); + // This is set after everything else in case an exception is thrown in an + // earlier step of the initialization. + invoker_ = RemoteInvoker; + } + + template ::value>> + void InitializeLocalManager() { + manager_ = LocalManagerTrivial; + } + + template ::value, int> = 0> + void InitializeLocalManager() { + manager_ = LocalManagerNontrivial; + } + + template + using HasTrivialRemoteStorage = + std::integral_constant::value && + alignof(T) <= + ABSL_INTERNAL_DEFAULT_NEW_ALIGNMENT>; + + template ::value>> + void InitializeRemoteManager(Args&&... args) { + // unique_ptr is used for exception-safety in case construction throws. + std::unique_ptr uninitialized_target( + ::operator new(sizeof(T)), TrivialDeleter(sizeof(T))); + ::new (uninitialized_target.get()) T(std::forward(args)...); + state_.remote.target = uninitialized_target.release(); + state_.remote.size = sizeof(T); + manager_ = RemoteManagerTrivial; + } + + template ::value, int> = 0> + void InitializeRemoteManager(Args&&... args) { + state_.remote.target = ::new T(std::forward(args)...); + manager_ = RemoteManagerNontrivial; + } + + ////////////////////////////////////////////////////////////////////////////// + // + // Type trait to determine if the template argument is an AnyInvocable whose + // function type is compatible enough with ours such that we can + // "move the guts" out of it when moving, rather than having to place a new + // object into remote storage. + + template + struct IsCompatibleAnyInvocable { + static constexpr bool value = false; + }; + + template + struct IsCompatibleAnyInvocable> { + static constexpr bool value = + (IsCompatibleConversion)(static_cast< + typename AnyInvocable::CoreImpl*>( + nullptr), + static_cast(nullptr)); + }; + + // + ////////////////////////////////////////////////////////////////////////////// + + TypeErasedState state_; + ManagerType* manager_; + InvokerType* invoker_; +}; + +// A constructor name-tag used with Impl to request the +// conversion-constructor +struct ConversionConstruct {}; + +//////////////////////////////////////////////////////////////////////////////// +// +// A metafunction that is normally an identity metafunction except that when +// given a std::reference_wrapper, it yields T&. This is necessary because +// currently std::reference_wrapper's operator() is not conditionally noexcept, +// so when checking if such an Invocable is nothrow-invocable, we must pull out +// the underlying type. +template +struct UnwrapStdReferenceWrapperImpl { + using type = T; +}; + +template +struct UnwrapStdReferenceWrapperImpl> { + using type = T&; +}; + +template +using UnwrapStdReferenceWrapper = + typename UnwrapStdReferenceWrapperImpl::type; +// +//////////////////////////////////////////////////////////////////////////////// + +// An alias that always yields std::true_type (used with constraints) where +// substitution failures happen when forming the template arguments. +template +using True = + std::integral_constant*) != 0>; + +/*SFINAE constraints for the conversion-constructor.*/ +template , AnyInvocable>::value>> +using CanConvert = + True>::value>, + absl::enable_if_t::template CallIsValid::value>, + absl::enable_if_t< + Impl::template CallIsNoexceptIfSigIsNoexcept::value>, + absl::enable_if_t, F>::value>>; + +/*SFINAE constraints for the std::in_place constructors.*/ +template +using CanEmplace = True< + absl::enable_if_t::template CallIsValid::value>, + absl::enable_if_t< + Impl::template CallIsNoexceptIfSigIsNoexcept::value>, + absl::enable_if_t, Args...>::value>>; + +/*SFINAE constraints for the conversion-assign operator.*/ +template , AnyInvocable>::value>> +using CanAssign = + True::template CallIsValid::value>, + absl::enable_if_t< + Impl::template CallIsNoexceptIfSigIsNoexcept::value>, + absl::enable_if_t, F>::value>>; + +/*SFINAE constraints for the reference-wrapper conversion-assign operator.*/ +template +using CanAssignReferenceWrapper = + True::template CallIsValid>::value>, + absl::enable_if_t::template CallIsNoexceptIfSigIsNoexcept< + std::reference_wrapper>::value>>; + +//////////////////////////////////////////////////////////////////////////////// +// +// The constraint for checking whether or not a call meets the noexcept +// callability requirements. This is a preprocessor macro because specifying it +// this way as opposed to a disjunction/branch can improve the user-side error +// messages and avoids an instantiation of std::is_nothrow_invocable_r in the +// cases where the user did not specify a noexcept function type. +// +#define ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT(inv_quals, noex) \ + ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT_##noex(inv_quals) + +// The disjunction below is because we can't rely on std::is_nothrow_invocable_r +// to give the right result when ReturnType is non-moveable in toolchains that +// don't treat non-moveable result types correctly. For example this was the +// case in libc++ before commit c3a24882 (2022-05). +#define ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT_true(inv_quals) \ + absl::enable_if_t> inv_quals, \ + P...>, \ + std::conjunction< \ + std::is_nothrow_invocable< \ + UnwrapStdReferenceWrapper> inv_quals, P...>, \ + std::is_same< \ + ReturnType, \ + absl::base_internal::invoke_result_t< \ + UnwrapStdReferenceWrapper> inv_quals, \ + P...>>>>::value> + +#define ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT_false(inv_quals) +// +//////////////////////////////////////////////////////////////////////////////// + +// A macro to generate partial specializations of Impl with the different +// combinations of supported cv/reference qualifiers and noexcept specifier. +// +// Here, `cv` are the cv-qualifiers if any, `ref` is the ref-qualifier if any, +// inv_quals is the reference type to be used when invoking the target, and +// noex is "true" if the function type is noexcept, or false if it is not. +// +// The CallIsValid condition is more complicated than simply using +// absl::base_internal::is_invocable_r because we can't rely on it to give the +// right result when ReturnType is non-moveable in toolchains that don't treat +// non-moveable result types correctly. For example this was the case in libc++ +// before commit c3a24882 (2022-05). +#define ABSL_INTERNAL_ANY_INVOCABLE_IMPL_(cv, ref, inv_quals, noex) \ + template \ + class Impl \ + : public CoreImpl { \ + public: \ + /*The base class, which contains the datamembers and core operations*/ \ + using Core = CoreImpl; \ + \ + /*SFINAE constraint to check if F is invocable with the proper signature*/ \ + template \ + using CallIsValid = True inv_quals, P...>, \ + std::is_same inv_quals, P...>>>::value>>; \ + \ + /*SFINAE constraint to check if F is nothrow-invocable when necessary*/ \ + template \ + using CallIsNoexceptIfSigIsNoexcept = \ + True; \ + \ + /*Put the AnyInvocable into an empty state.*/ \ + Impl() = default; \ + \ + /*The implementation of a conversion-constructor from "f*/ \ + /*This forwards to Core, attaching inv_quals so that the base class*/ \ + /*knows how to properly type-erase the invocation.*/ \ + template \ + explicit Impl(ConversionConstruct, F&& f) \ + : Core(TypedConversionConstruct< \ + typename std::decay::type inv_quals>(), \ + std::forward(f)) {} \ + \ + /*Forward along the in-place construction parameters.*/ \ + template \ + explicit Impl(absl::in_place_type_t, Args&&... args) \ + : Core(absl::in_place_type inv_quals>, \ + std::forward(args)...) {} \ + \ + /*The actual invocation operation with the proper signature*/ \ + ReturnType operator()(P... args) cv ref noexcept(noex) { \ + assert(this->invoker_ != nullptr); \ + return this->invoker_(const_cast(&this->state_), \ + static_cast>(args)...); \ + } \ + } + +// Define the `noexcept(true)` specialization only for C++17 and beyond, when +// `noexcept` is part of the type system. +#if ABSL_INTERNAL_CPLUSPLUS_LANG >= 201703L +// A convenience macro that defines specializations for the noexcept(true) and +// noexcept(false) forms, given the other properties. +#define ABSL_INTERNAL_ANY_INVOCABLE_IMPL(cv, ref, inv_quals) \ + ABSL_INTERNAL_ANY_INVOCABLE_IMPL_(cv, ref, inv_quals, false); \ + ABSL_INTERNAL_ANY_INVOCABLE_IMPL_(cv, ref, inv_quals, true) +#else +#define ABSL_INTERNAL_ANY_INVOCABLE_IMPL(cv, ref, inv_quals) \ + ABSL_INTERNAL_ANY_INVOCABLE_IMPL_(cv, ref, inv_quals, false) +#endif + +// Non-ref-qualified partial specializations +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(, , &); +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(const, , const&); + +// Lvalue-ref-qualified partial specializations +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(, &, &); +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(const, &, const&); + +// Rvalue-ref-qualified partial specializations +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(, &&, &&); +ABSL_INTERNAL_ANY_INVOCABLE_IMPL(const, &&, const&&); + +// Undef the detail-only macros. +#undef ABSL_INTERNAL_ANY_INVOCABLE_IMPL +#undef ABSL_INTERNAL_ANY_INVOCABLE_IMPL_ +#undef ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT_false +#undef ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT_true +#undef ABSL_INTERNAL_ANY_INVOCABLE_NOEXCEPT_CONSTRAINT +#undef ABSL_INTERNAL_NOEXCEPT_SPEC + +} // namespace internal_any_invocable +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_INTERNAL_ANY_INVOCABLE_H_ diff --git a/src/absl/functional/internal/front_binder.h b/src/absl/functional/internal/front_binder.h new file mode 100644 index 0000000..45f52de --- /dev/null +++ b/src/absl/functional/internal/front_binder.h @@ -0,0 +1,95 @@ +// Copyright 2018 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +// Implementation details for `absl::bind_front()`. + +#ifndef ABSL_FUNCTIONAL_INTERNAL_FRONT_BINDER_H_ +#define ABSL_FUNCTIONAL_INTERNAL_FRONT_BINDER_H_ + +#include +#include +#include + +#include "absl/base/internal/invoke.h" +#include "absl/container/internal/compressed_tuple.h" +#include "absl/meta/type_traits.h" +#include "absl/utility/utility.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace functional_internal { + +// Invoke the method, expanding the tuple of bound arguments. +template +R Apply(Tuple&& bound, absl::index_sequence, Args&&... free) { + return base_internal::invoke( + absl::forward(bound).template get()..., + absl::forward(free)...); +} + +template +class FrontBinder { + using BoundArgsT = absl::container_internal::CompressedTuple; + using Idx = absl::make_index_sequence; + + BoundArgsT bound_args_; + + public: + template + constexpr explicit FrontBinder(absl::in_place_t, Ts&&... ts) + : bound_args_(absl::forward(ts)...) {} + + template > + R operator()(FreeArgs&&... free_args) & { + return functional_internal::Apply(bound_args_, Idx(), + absl::forward(free_args)...); + } + + template > + R operator()(FreeArgs&&... free_args) const& { + return functional_internal::Apply(bound_args_, Idx(), + absl::forward(free_args)...); + } + + template > + R operator()(FreeArgs&&... free_args) && { + // This overload is called when *this is an rvalue. If some of the bound + // arguments are stored by value or rvalue reference, we move them. + return functional_internal::Apply(absl::move(bound_args_), Idx(), + absl::forward(free_args)...); + } + + template > + R operator()(FreeArgs&&... free_args) const&& { + // This overload is called when *this is an rvalue. If some of the bound + // arguments are stored by value or rvalue reference, we move them. + return functional_internal::Apply(absl::move(bound_args_), Idx(), + absl::forward(free_args)...); + } +}; + +template +using bind_front_t = FrontBinder, absl::decay_t...>; + +} // namespace functional_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_INTERNAL_FRONT_BINDER_H_ diff --git a/src/absl/functional/internal/function_ref.h b/src/absl/functional/internal/function_ref.h new file mode 100644 index 0000000..b5bb8b4 --- /dev/null +++ b/src/absl/functional/internal/function_ref.h @@ -0,0 +1,106 @@ +// Copyright 2019 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#ifndef ABSL_FUNCTIONAL_INTERNAL_FUNCTION_REF_H_ +#define ABSL_FUNCTIONAL_INTERNAL_FUNCTION_REF_H_ + +#include +#include +#include + +#include "absl/base/internal/invoke.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN +namespace functional_internal { + +// Like a void* that can handle function pointers as well. The standard does not +// allow function pointers to round-trip through void*, but void(*)() is fine. +// +// Note: It's important that this class remains trivial and is the same size as +// a pointer, since this allows the compiler to perform tail-call optimizations +// when the underlying function is a callable object with a matching signature. +union VoidPtr { + const void* obj; + void (*fun)(); +}; + +// Chooses the best type for passing T as an argument. +// Attempt to be close to SystemV AMD64 ABI. Objects with trivial copy ctor are +// passed by value. +template +constexpr bool PassByValue() { + return !std::is_lvalue_reference::value && + absl::is_trivially_copy_constructible::value && + absl::is_trivially_copy_assignable< + typename std::remove_cv::type>::value && + std::is_trivially_destructible::value && + sizeof(T) <= 2 * sizeof(void*); +} + +template +struct ForwardT : std::conditional(), T, T&&> {}; + +// An Invoker takes a pointer to the type-erased invokable object, followed by +// the arguments that the invokable object expects. +// +// Note: The order of arguments here is an optimization, since member functions +// have an implicit "this" pointer as their first argument, putting VoidPtr +// first allows the compiler to perform tail-call optimization in many cases. +template +using Invoker = R (*)(VoidPtr, typename ForwardT::type...); + +// +// InvokeObject and InvokeFunction provide static "Invoke" functions that can be +// used as Invokers for objects or functions respectively. +// +// static_cast handles the case the return type is void. +template +R InvokeObject(VoidPtr ptr, typename ForwardT::type... args) { + auto o = static_cast(ptr.obj); + return static_cast( + absl::base_internal::invoke(*o, std::forward(args)...)); +} + +template +R InvokeFunction(VoidPtr ptr, typename ForwardT::type... args) { + auto f = reinterpret_cast(ptr.fun); + return static_cast( + absl::base_internal::invoke(f, std::forward(args)...)); +} + +template +void AssertNonNull(const std::function& f) { + assert(f != nullptr); + (void)f; +} + +template +void AssertNonNull(const F&) {} + +template +void AssertNonNull(F C::*f) { + assert(f != nullptr); + (void)f; +} + +template +using EnableIf = typename ::std::enable_if::type; + +} // namespace functional_internal +ABSL_NAMESPACE_END +} // namespace absl + +#endif // ABSL_FUNCTIONAL_INTERNAL_FUNCTION_REF_H_ diff --git a/src/absl/memory/memory.h b/src/absl/memory/memory.h new file mode 100644 index 0000000..d633260 --- /dev/null +++ b/src/absl/memory/memory.h @@ -0,0 +1,699 @@ +// Copyright 2017 The Abseil Authors. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// https://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +// +// ----------------------------------------------------------------------------- +// File: memory.h +// ----------------------------------------------------------------------------- +// +// This header file contains utility functions for managing the creation and +// conversion of smart pointers. This file is an extension to the C++ +// standard library header file. + +#ifndef ABSL_MEMORY_MEMORY_H_ +#define ABSL_MEMORY_MEMORY_H_ + +#include +#include +#include +#include +#include +#include + +#include "absl/base/macros.h" +#include "absl/meta/type_traits.h" + +namespace absl { +ABSL_NAMESPACE_BEGIN + +// ----------------------------------------------------------------------------- +// Function Template: WrapUnique() +// ----------------------------------------------------------------------------- +// +// Adopts ownership from a raw pointer and transfers it to the returned +// `std::unique_ptr`, whose type is deduced. Because of this deduction, *do not* +// specify the template type `T` when calling `WrapUnique`. +// +// Example: +// X* NewX(int, int); +// auto x = WrapUnique(NewX(1, 2)); // 'x' is std::unique_ptr. +// +// Do not call WrapUnique with an explicit type, as in +// `WrapUnique(NewX(1, 2))`. The purpose of WrapUnique is to automatically +// deduce the pointer type. If you wish to make the type explicit, just use +// `std::unique_ptr` directly. +// +// auto x = std::unique_ptr(NewX(1, 2)); +// - or - +// std::unique_ptr x(NewX(1, 2)); +// +// While `absl::WrapUnique` is useful for capturing the output of a raw +// pointer factory, prefer 'absl::make_unique(args...)' over +// 'absl::WrapUnique(new T(args...))'. +// +// auto x = WrapUnique(new X(1, 2)); // works, but nonideal. +// auto x = make_unique(1, 2); // safer, standard, avoids raw 'new'. +// +// Note that `absl::WrapUnique(p)` is valid only if `delete p` is a valid +// expression. In particular, `absl::WrapUnique()` cannot wrap pointers to +// arrays, functions or void, and it must not be used to capture pointers +// obtained from array-new expressions (even though that would compile!). +template +std::unique_ptr WrapUnique(T* ptr) { + static_assert(!std::is_array::value, "array types are unsupported"); + static_assert(std::is_object::value, "non-object types are unsupported"); + return std::unique_ptr(ptr); +} + +namespace memory_internal { + +// Traits to select proper overload and return type for `absl::make_unique<>`. +template +struct MakeUniqueResult { + using scalar = std::unique_ptr; +}; +template +struct MakeUniqueResult { + using array = std::unique_ptr; +}; +template +struct MakeUniqueResult { + using invalid = void; +}; + +} // namespace memory_internal + +// gcc 4.8 has __cplusplus at 201301 but the libstdc++ shipped with it doesn't +// define make_unique. Other supported compilers either just define __cplusplus +// as 201103 but have make_unique (msvc), or have make_unique whenever +// __cplusplus > 201103 (clang). +#if (__cplusplus > 201103L || defined(_MSC_VER)) && \ + !(defined(__GLIBCXX__) && !defined(__cpp_lib_make_unique)) +using std::make_unique; +#else +// ----------------------------------------------------------------------------- +// Function Template: make_unique() +// ----------------------------------------------------------------------------- +// +// Creates a `std::unique_ptr<>`, while avoiding issues creating temporaries +// during the construction process. `absl::make_unique<>` also avoids redundant +// type declarations, by avoiding the need to explicitly use the `new` operator. +// +// This implementation of `absl::make_unique<>` is designed for C++11 code and +// will be replaced in C++14 by the equivalent `std::make_unique<>` abstraction. +// `absl::make_unique<>` is designed to be 100% compatible with +// `std::make_unique<>` so that the eventual migration will involve a simple +// rename operation. +// +// For more background on why `std::unique_ptr(new T(a,b))` is problematic, +// see Herb Sutter's explanation on +// (Exception-Safe Function Calls)[https://herbsutter.com/gotw/_102/]. +// (In general, reviewers should treat `new T(a,b)` with scrutiny.) +// +// Example usage: +// +// auto p = make_unique(args...); // 'p' is a std::unique_ptr +// auto pa = make_unique(5); // 'pa' is a std::unique_ptr +// +// Three overloads of `absl::make_unique` are required: +// +// - For non-array T: +// +// Allocates a T with `new T(std::forward args...)`, +// forwarding all `args` to T's constructor. +// Returns a `std::unique_ptr` owning that object. +// +// - For an array of unknown bounds T[]: +// +// `absl::make_unique<>` will allocate an array T of type U[] with +// `new U[n]()` and return a `std::unique_ptr` owning that array. +// +// Note that 'U[n]()' is different from 'U[n]', and elements will be +// value-initialized. Note as well that `std::unique_ptr` will perform its +// own destruction of the array elements upon leaving scope, even though +// the array [] does not have a default destructor. +// +// NOTE: an array of unknown bounds T[] may still be (and often will be) +// initialized to have a size, and will still use this overload. E.g: +// +// auto my_array = absl::make_unique(10); +// +// - For an array of known bounds T[N]: +// +// `absl::make_unique<>` is deleted (like with `std::make_unique<>`) as +// this overload is not useful. +// +// NOTE: an array of known bounds T[N] is not considered a useful +// construction, and may cause undefined behavior in templates. E.g: +// +// auto my_array = absl::make_unique(); +// +// In those cases, of course, you can still use the overload above and +// simply initialize it to its desired size: +// +// auto my_array = absl::make_unique(10); + +// `absl::make_unique` overload for non-array types. +template +typename memory_internal::MakeUniqueResult::scalar make_unique( + Args&&... args) { + return std::unique_ptr(new T(std::forward(args)...)); +} + +// `absl::make_unique` overload for an array T[] of unknown bounds. +// The array allocation needs to use the `new T[size]` form and cannot take +// element constructor arguments. The `std::unique_ptr` will manage destructing +// these array elements. +template +typename memory_internal::MakeUniqueResult::array make_unique(size_t n) { + return std::unique_ptr(new typename absl::remove_extent_t[n]()); +} + +// `absl::make_unique` overload for an array T[N] of known bounds. +// This construction will be rejected. +template +typename memory_internal::MakeUniqueResult::invalid make_unique( + Args&&... /* args */) = delete; +#endif + +// ----------------------------------------------------------------------------- +// Function Template: RawPtr() +// ----------------------------------------------------------------------------- +// +// Extracts the raw pointer from a pointer-like value `ptr`. `absl::RawPtr` is +// useful within templates that need to handle a complement of raw pointers, +// `std::nullptr_t`, and smart pointers. +template +auto RawPtr(T&& ptr) -> decltype(std::addressof(*ptr)) { + // ptr is a forwarding reference to support Ts with non-const operators. + return (ptr != nullptr) ? std::addressof(*ptr) : nullptr; +} +inline std::nullptr_t RawPtr(std::nullptr_t) { return nullptr; } + +// ----------------------------------------------------------------------------- +// Function Template: ShareUniquePtr() +// ----------------------------------------------------------------------------- +// +// Adopts a `std::unique_ptr` rvalue and returns a `std::shared_ptr` of deduced +// type. Ownership (if any) of the held value is transferred to the returned +// shared pointer. +// +// Example: +// +// auto up = absl::make_unique(10); +// auto sp = absl::ShareUniquePtr(std::move(up)); // shared_ptr +// CHECK_EQ(*sp, 10); +// CHECK(up == nullptr); +// +// Note that this conversion is correct even when T is an array type, and more +// generally it works for *any* deleter of the `unique_ptr` (single-object +// deleter, array deleter, or any custom deleter), since the deleter is adopted +// by the shared pointer as well. The deleter is copied (unless it is a +// reference). +// +// Implements the resolution of [LWG 2415](http://wg21.link/lwg2415), by which a +// null shared pointer does not attempt to call the deleter. +template +std::shared_ptr ShareUniquePtr(std::unique_ptr&& ptr) { + return ptr ? std::shared_ptr(std::move(ptr)) : std::shared_ptr(); +} + +// ----------------------------------------------------------------------------- +// Function Template: WeakenPtr() +// ----------------------------------------------------------------------------- +// +// Creates a weak pointer associated with a given shared pointer. The returned +// value is a `std::weak_ptr` of deduced type. +// +// Example: +// +// auto sp = std::make_shared(10); +// auto wp = absl::WeakenPtr(sp); +// CHECK_EQ(sp.get(), wp.lock().get()); +// sp.reset(); +// CHECK(wp.lock() == nullptr); +// +template +std::weak_ptr WeakenPtr(const std::shared_ptr& ptr) { + return std::weak_ptr(ptr); +} + +namespace memory_internal { + +// ExtractOr::type evaluates to E if possible. Otherwise, D. +template