Re: some git confusion (where git's advice didn't help)

[Jeff King <peff@peff.net>]

On Tue, May 14, 2019 at 11:54:43AM +0200, Ulrich Windl wrote:

> > git branch --track origin/f-gcc-4.8
> Branch origin/f-gcc-4.8 set up to track local branch next.
> > git fetch
> > git branch
>   f-systemd
>   integration
>   master
> * next
>   origin/f-gcc-4.8
> > git merge f-gcc-4.8
> merge: f-gcc-4.8 - not something we can merge
> 
> Did you mean this?
>         origin/f-gcc-4.8
> > git merge origin/f-gcc-4.8
> warning: refname 'origin/f-gcc-4.8' is ambiguous.
> Already up-to-date.
> 
> ### So actually this advice wasn't helpful at all. Cause of the
> problem most likely was "git branch --track origin/f-gcc-4.8" that
> "imported" the branch under the same name as the remote branch is
> referenced.

Right, that was the source of the problem. Having both
"refs/heads/origin/f-gcc-4.8" and "refs/remotes/origin/f-gcc-4.8" is
going to lead to confusion, and you're best off deleting the mistaken
branch as soon as possible.

But I agree we could be more helpful in the messages.

The "did you mean?" advice just blindly says "oh, you asked for X and
refs/remotes/ABC/X exists, so let's suggest ABC/X", without checking for
ambiguities. It should probably do this:

diff --git a/help.c b/help.c
index a9e451f2ee..108ca54af3 100644
--- a/help.c
+++ b/help.c
@@ -759,7 +759,8 @@ static int append_similar_ref(const char *refname, const struct object_id *oid,
 	/* A remote branch of the same name is deemed similar */
 	if (skip_prefix(refname, "refs/remotes/", &remote) &&
 	    !strcmp(branch, cb->base_ref))
-		string_list_append(cb->similar_refs, remote);
+		string_list_append(cb->similar_refs,
+				   shorten_unambiguous_ref(refname, 1));
 	return 0;
 }
 

which would print "ABC/X" in most cases, but "remotes/ABC/X" for your
ambiguous case. Incidentally, the existing code also has a memory
problem! It blindly skips past "refs/remotes/" in the refname and saves
the pointer away in a NODUP string-list. But that refname pointer isn't
ours, and isn't guaranteed to last past our for_each_ref() callback. The
hunk above fixes it because shorten_unambiguous_ref() always returns a
newly allocated string. :)

I also think the "warning: refname ... is ambiguous" message would
probably be a bit more helpful if it showed _which_ candidates it found
(and which one it chose!).

-Peff