v3contrib/completion: fix zsh completion regression from 59d85a2a05

A recent change to make git-completion.bash use $__git_cmd_idx in more places broke a number of completions on zsh because it modified __git_main but did not update __git_zsh_main.

Notably, completions for "add", "branch", "mv" and "push" were broken as a result of this change.

In addition to the undefined variable usage, "git mv <tab>" also prints the following error:

	__git_count_arguments:7: bad math expression:
	operand expected at `"1"'

	_git_mv:[:7: unknown condition: -gt

Remove the quotes around $__git_cmd_idx in __git_count_arguments and set __git_cmd_idx=1 early in __git_zsh_main to fix the regressions from 59d85a2a05.

Add "git" to the "words" array in _git_zsh_main to guarantee that "git" is at least always in the completion list. It may not be completely correct because __git_cmd_idx is not always 1, but it is enough to fix the regression.

This was tested on zsh 5.7.1 (x86_64-apple-darwin19.0).

Helped-by: Felipe Contreras <felipe.contreras@gmail.com>
Signed-off-by: David Aguilar <davvid@gmail.com>
---
 contrib/completion/git-completion.bash | 2 +-
 contrib/completion/git-completion.zsh  | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

David Aguilar wrote:

Hm, no. The current code already guarantees "git" is always at the start of the completion list. In [1] I suggested to add git *if* $words is used instead of $orig_words.

If you add "git" to $orig_words you end up with something like "git git mv", so the __git_cmd_idx is definitely not 1.

You should probably try to test yourself:

  words=( git ${words[@]} )
  echo "$words" >> /tmp/words-log.txt

The problem is that zsh's _arguments eats all the words it finds, so for example if you type:

  git mv --force <tab>

$words will be 'mv --force'.

It's better to use $words because in case there's arguments beforehand, like:

  git --git-dir=/tmp/test/.git mv --force

$words will be 'mv --force', so we can get the proper index by just adding 'git' beforehand.

But it doesn't work for arguments not in _arguments, like:

  git --foo mv --force

Which returns:

  --foo mv --force

And unfortunately upstream's version of the wrapper doesn't understand many arguments, like -c, or -C. git-completion does have all of them

It's better to just leave the code as it is and just fix the regression by adding __git_cmd_idx=1.

I mean I kind of wrote 2 of the 3 lines you sent, can I get a Suggested-by?

This is needed.

This is wrong. The current code is fine.

Cheers.

[1] https://lore.kernel.org/git/60b3c2d7557bd_be762089a@natae.notmuch/

On Tue, Jun 1, 2021 at 12:15 PM Felipe Contreras <felipe.contreras@gmail.com> wrote:

The v4 I just sent is now basically the same as v2 modulo the
Suggested-by: trailer update.

Thanks for the detailed explanation.

Just so I'm understanding this correctly.. if this was instead..

    words=( git ${words[@]} )

(instead of orig_words like I mistakenly included in v3) would that be an improvement, no-op or would it be worse? It sounds like additional changes are needed to make it properly support options between "git" and the sub-command name, hence the patch is fine as-is in v4, correct?

Hopefully in the future it can be extended to cover eg. "git -c foo.bar -C some-dir <sub-command>" as well. Thanks for your patience.

cheers,

David Aguilar wrote:

It would be an improvement, but it's orthogonal to the regression you are trying to fix.

I would just fix the regression for v2.32, and then afterwards try to do the improvement.

I have a testing framework for the zsh completion in my git-completion project, so I would be much more confident about this change if all the tests pass. Alas I have not yet merged any v2.32.0-rc* so it's not straightforward to run the tests now.

I mean there's git options, and git command options. I don't know how many changes are needed to make all the interactions work correctly, but I wouldn't have confidence in any of them so close to a release, especially considering git.git doesn't have any zsh tests.

So yes, v4 is fine.

That will work correctly on git-completion once I merge your v4 patch (and v2.32).