Hi there
First time I post to the mailing list here, I hope I get it right...
I came across an issue where pushing a subtree resulted in unexpected
commits in the remote repository: These commits included files that are
not part of the subtree's directory. There's a related question on
stackoverflow:
https://stackoverflow.com/questions/61150709/git-subtree-push-seems-to-push-commits-that-dont-apply-to-the-subtree
In my case, the excess commits were introduced by a merge commit of a
branch that forked off the main branch *before* the subtree was added.
This merge commit was the result of updating my fork from the upstream
repository (which is not aware of the subtree, it just lives in my fork
so far).
I found the issue to be related with the following change in git-subtree.sh:
Revision: 933cfeb90b5d03b4096db6d60494a6eedea25d03
Author: Dave Ware [off-list ref]
Date: 15.01.2016 01:41:43
Message:
contrib/subtree: fix "subtree split" skipped-merge bug
'git subtree split' can incorrectly skip a merge even when both parents
act on the subtree, provided the merge results in a tree identical to
one of the parents. Fix by copying the merge if at least one parent is
non-identical, and the non-identical parent is not an ancestor of the
identical parent.
Also, add a test case which checks that a descendant remains a
descendent on the subtree in this case.
Signed-off-by: Dave Ware <redacted>
Reviewed-by: David A. Greene <redacted>
Signed-off-by: Junio C Hamano <redacted>
----
Modified: contrib/subtree/git-subtree.sh
Modified: contrib/subtree/t/t7900-subtree.sh
copycommit=
if [ -n "$identical" ] && [ -n "$nonidentical" ]; then
extras=$(git rev-list --count $identical..$nonidentical)
if [ "$extras" -ne 0 ]; then
# we need to preserve history along the other branch
copycommit=1 // <---------------------
fi
fi
if [ -n "$identical" ] && [ -z "$copycommit" ]; then
echo $identical
else
copy_commit $rev $tree "$p" || exit $?
fi
As a workaround, removing the copycommit=1 line here results in a split
as I would expect for my case.
I attached a sample repository and log that reproduces the issue:
- From the main branch, do a "git subtree split --prefix my_subtree -b
splitted -d". This is how the branches listed below were created.
- branch "splitted_bad" shows the unexpected merge commit f7fd955 that
includes files that are not part of the subtree (9eeff05)
- branch "splitted_good" shows the result as I would expect it, without
merge commit f7fd955. It was done by commenting out the copycommit=1
Now I wonder, is this expected behavior, or could it be a regression? I
do not really understand the inner workings of git-subtree.sh, but I
feel that where it included "too few" commits before the mentioned
patch, it might now include "too many".
Regards
Daniel