Fix timeout handling in _hurd_select
authorRichard Braun <rbraun@sceen.net>
Thu, 28 Feb 2019 05:56:01 +0000 (05:56 +0000)
committerSamuel Thibault <sthibault@debian.org>
Thu, 28 Feb 2019 05:56:01 +0000 (05:56 +0000)
commitb375ddea8f14a06296071be513df74bbb5cb3e35
tree44b58280e5d3e87e57cf6775e7f0e25afd86611d
parentaaaf707c2997955e99e105dd8d06f054b19255d0
Fix timeout handling in _hurd_select

Rely on servers to implement timeouts, so that very short values (including
0) don't make mach_msg return before valid replies can be received. The
purpose of this scheme is to guarantee a full client-server round-trip,
whatever the timeout value.

This change depends on the new io_select_timeout RPC being implemented by
servers.

* hurd/Makefile (user-interfaces): Add io_reply and io_request.
* hurd/hurdselect.c: Include <sys/time.h>, <hurd/io_request.h> and <limits.h>.
(_hurd_select): Replace the call to __io_select with either __io_select_request
or __io_select_timeout_request, depending on the timeout. Count the number of
ready descriptors (replies for which at least one type bit is set). Implement
the timeout locally when there is no file descriptor.

TODO: see XXX

Gbp-Pq: Topic hurd-i386
Gbp-Pq: Name tg-io_select_timeout.diff
hurd/Makefile
hurd/hurdselect.c