Re: Regression in 'git pull --rebase --autostash' since v2.32.0

[Philippe Blain <levraiphilippeblain@gmail.com>]

Le 2021-07-17 à 11:29, Philippe Blain a écrit :
> Hi Felipe,
> 
> Your recent clean-up of 'git pull --autostash' seems to unfortunately have made things
> worse if the pull brings new files that conflict with existing untracked files,
> which makes the pull abort,
> and there are tracked files with modifications (so --autostash comes into play).
> 
> Before your change, 'git pull --no-rebase --autostash' did *not* apply the autostash
> after the pull failed, thus loosing modifications to tracked files (and it did not save the
> stash entry !). 'git pull --rebase --autostash' correctly applied the autostash, but ended with
> a strange "error: could not detach HEAD".
> 
> After your change, both 'git pull --no-rebase --autostash' and 'git pull --rebase --autostash'
> have the same buggy behaviour: they do not apply the autostash and do not save it in the stash list.

The change below seem to fix both cases:

diff --git a/builtin/merge.c b/builtin/merge.c
index a8a843b1f5..b2ad70c50a 100644
--- a/builtin/merge.c
+++ b/builtin/merge.c
@@ -1560,6 +1560,7 @@ int cmd_merge(int argc, const char **argv, const char *prefix)
  					  &head_commit->object.oid,
  					  &commit->object.oid,
  					  overwrite_ignore)) {
+			apply_autostash(git_path_merge_autostash(the_repository));
  			ret = 1;
  			goto done;
  		}


*But* from a quick audit of 'cmd_merge', there are still two code paths that
call 'create_autostash' but then fail to apply it before calling 'goto done':

1. the branch 'if (automerge_was_ok)' (line 1693)
2. the branch 'if (!best_strategy)' (line 1704)


Cheers,
Philippe.