analysis

diff --git a/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_1_e203ba0725ab5b38472dee8e3c0f9256._comment b/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_1_e203ba0725ab5b38472dee8e3c0f9256._comment
new file mode 100644 (file)
index 0000000..847e9f4
--- /dev/null
+++ b/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_1_e203ba0725ab5b38472dee8e3c0f9256._comment
@@ -0,0 +1,19 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 1"""
+ date="2024-07-31T14:20:51Z"
+ content="""
+Note that this does not affect the number of commits made by `addurl` generally
+eg when adding multiple urls with --batch from the web.
+
+Also, I don't think that the commits you picked out and showed necessarily
+correspond to one-another. The state being recorded in the commit in the 1st
+run is not the same as the state that gets recorded by the two commits in the
+2nd run. Unless, there is an actual behavior change that eg, leaves the file
+present in a repository that it was not present in before. 
+
+In the first run the commit shows key
+MD5E-s5--db87ebcba59a8c9f34b68e713c08a718.dat ends up recorded as present in
+datalad-archives but not in the local repository. In the second run, the
+commits show that the same key ends up recorded present in both repositories.
+"""]]

diff --git a/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_2_e203ba0725ab5b38472dee8e3c0f9256._comment b/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_2_e203ba0725ab5b38472dee8e3c0f9256._comment
new file mode 100644 (file)
index 0000000..1f9cd0c
--- /dev/null
+++ b/doc/bugs/change_in_beh__58___addurls_creates_multiple_commits/comment_2_e203ba0725ab5b38472dee8e3c0f9256._comment
@@ -0,0 +1,23 @@
+[[!comment format=mdwn
+ username="joey"
+ subject="""comment 2"""
+ date="2024-07-31T16:06:38Z"
+ content="""
+Bisected to [[!commit 780367200b14d532f745079dfa09ffaa214d0a84]],
+"remove dead nodes when loading the cluster log".
+
+Replacing `loadClusters` with a noop on top of that commit gets the test
+suite passing again.
+
+Since nothing in `loadClusters` involves the location log at all, I think
+this must come down to a difference in when/if git-annex starts reading
+from the git-annex branch. There could be git-annex commands that didn't
+used to read from the branch before, that now do. Which might mean merging
+in other git-annex branches at different points in time than happened
+before, which I suppose can result in an additional commit.
+
+Unfortunately, I can't avoid the early `loadClusters` for reasons explained
+in that commit.
+
+Anyway, I doubt this will result in a lot of additional commits.
+"""]]