Re: [PATCH v3] contrib/completion: fix zsh completion regression from 59d85a2a05

[David Aguilar <davvid@gmail.com>]

On Tue, Jun 1, 2021 at 12:15 PM Felipe Contreras
<felipe.contreras@gmail.com> wrote:
>
> David Aguilar wrote:
> > A recent change to make git-completion.bash use $__git_cmd_idx
> > Add "git" to the "words" array in _git_zsh_main to guarantee
> > that "git" is at least always in the completion list.
>
> 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.
>
> > Helped-by: Felipe Contreras <felipe.contreras@gmail.com>
>
> I mean I kind of wrote 2 of the 3 lines you sent, can I get a
> Suggested-by?


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


>
> > --- a/contrib/completion/git-completion.zsh
> > +++ b/contrib/completion/git-completion.zsh
> > @@ -251,7 +251,7 @@ __git_zsh_main ()
> >               done
> >               ;;
> >       (arg)
> > -             local command="${words[1]}" __git_dir
> > +             local command="${words[1]}" __git_dir __git_cmd_idx=1
>
> This is needed.
>
> >
> >               if (( $+opt_args[--bare] )); then
> >                       __git_dir='.'
> > @@ -261,7 +261,7 @@ __git_zsh_main ()
> >
> >               (( $+opt_args[--help] )) && command='help'
> >
> > -             words=( ${orig_words[@]} )
> > +             words=( git ${orig_words[@]} )
>
> This is wrong. The current code is fine.
>
> Cheers.
>
> [1] https://lore.kernel.org/git/60b3c2d7557bd_be762089a@natae.notmuch/

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