patch[kernel] completion: silence "fatal: Not a git repository" error

It is possible that a user is trying to run a git command and fail to realize that they are not in a git repository or working tree. When trying to complete an operation, __git_refs would fall to a degenerate case and attempt to use "git for-each-ref", which would emit the error.

Let's fix this by shunting the error message coming from "git for-each-ref".

Signed-off-by: John Szakmeister <john@szakmeister.net>
---
 contrib/completion/git-completion.bash | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

On Tue, Oct 14, 2014 at 6:49 AM, John Szakmeister <john@szakmeister.net> wrote:

Sorry for the "[kernel]" in the subject. I must have forgotten to remove that off of my format-patch invocation. If you need me to resubmit without it, I can do that.

Thanks!

-John

John Szakmeister <john@szakmeister.net> writes:

Hmph, do you mean this one?

    $ cd /var/tmp ;# not a git repository
    $ git checkout <TAB>

->

    $ git checkout fatal: Not a git repository (or any of the parent directories): .git
    HEAD 

I agree it is ugly, but would it be an improvement for the end user, who did not realize that she was not in a directory where "git checkout" makes sense, not to tell her that she is not in a git repository in some way?

On Tue, Oct 14, 2014 at 2:29 PM, Junio C Hamano <gitster@pobox.com> wrote:

I had thought about that too, but I think--for me--it comes down to two things:

1) We're not intentionally trying to inform the user anywhere else
that they are not in a git repo.  We simply fail to complete anything,
which I think is an established behavior.
2) It mingles with the stuff already on the command line, making it
confusing to know what you typed.  Then you end up ctrl-c'ing your way
out of it and starting over--which is the frustrating part.

For me, I thought it better to just be more well-behaved. I've also run across this issue when I legitimately wanted to do something--I wish I could remember what it was--with a remote repo and didn't happen to be in a git working tree. It was frustrating to see this error message then too, for the same reason as above. I use tab completion quite extensively, so spitting things like this out making it difficult to move forward is a problem.

Would it be better to check that "$dir" is non-empty and then provide the extra bits of information? We could then avoid giving the user anything in that case.

-John

John Szakmeister <john@szakmeister.net> writes:

It is not that I am unsympathetic. It's just it looks to me that the patch is potentially adding one more failed step by hiding the error message to further frustrate the user.

    $ git checkout <TAB>
    ... completes nothing; puzzled but decides not to be worried for now
    $ git checkout master<RET>
    fatal: not a git repository

As you noticed, however, we do not show the ugly error message by design. It is not done consistently, either (happens only when we try to complete refnames).

I was just hoping that somebody (not necessarily you) could suggest a way to do better than hide the error message only because it looks ugly (iow, perhaps show it not in the middle of the command line, and do so more consistently). Yes I would imagine it would be a lot harder, but the end user experience _might_ become so much better to make it worthwhile. I dunno.

I am not strongly opposed to queuing the patch.