Re: [RFC] Case-insensitive path completion in _git

[Daniel Shahaf <d.s@xxxxxxxxxxxxxxxxxx>]

m0viefreak wrote on Sun, Dec 13, 2020 at 14:55:55 +0100:
> 
> 
> On 13.12.2020 05:40, dana wrote:
> > _git has issues completing files case-insensitively (if you have matcher-list
> > 'm:{a-zA-Z}={A-Za-z}' or whatever). Looking into it, i found that __git_files
> > is trying to pass a glob pattern to `git ls-files`, and this fails if there's
> > not an exact case match, since ls-files is always case-sensitive.
> 
> > There is a fall-back to `git ls-files` with no path, but this doesn't always
> > work either, because it defaults to the CWD, and the file you're trying to
> > complete may not be under the CWD. Even when the fall-back succeeds, it's not
> > ideal, because it'll pass every single file in the tree to _multi_parts, which
> > can be slow.
> > 
> > The following hack solves the problem for me, but it might be too silly to
> > commit. Can anyone think of a more proper fix? If not, would the hack be
> > viable (probably gated behind a style)?
> > 
> > dana
> > 
> > ...
> > +  pref=${pref//(#m)[[:alpha:]]/\[${(U)MATCH}${(L)MATCH}\]}
> > ...
> 
> This solution only handles simple upper-lower-case conversion matcher-list styles,
> but it will not work for more complicated styles like 'r:|[._-]=* r:|=*'.
> That's probably fine for most use-cases.

What about the default use-case — no matchspec and case-sensitive
matching?  The patch transforms a plain string to a list of character
classes, and it's not clear to me what costs that might incur on large
trees.

@Lawrence Thanks for fishing these out of the archives ☺

Cheers,

Daniel

> I have been running with the following patch applied locally for quite some time:
> 
> 
> diff --git a/Completion/Unix/Command/_git b/Completion/Unix/Command/_git
> index 81a060e4d..1a6619e31 100644
> --- a/Completion/Unix/Command/_git
> +++ b/Completion/Unix/Command/_git
> @@ -7153,7 +7153,7 @@ __git_files () {
>    # TODO: --directory should probably be added to $opts when --others is given.
> 
>    local pref=${(Q)${~PREFIX}}
> -  [[ $pref[1] == '/' ]] || pref=$gittoplevel$gitprefix$pref
> +  [[ $pref[1] == '/' ]] || pref=$gittoplevel$gitprefix
> 
>    # First allow ls-files to pattern-match in case of remote repository
>    files=(${(0)"$(_call_program files git ls-files -z --exclude-standard ${(q)opts} -- ${(q)${pref:+$pref\*}:-.} 2>/dev/null)"})
> 
> 
> It just lets _multi_parts do the matching.
> Performance-wise, it's obviously worse than than your proposed solution, but
> still better than the fallback, since it still passes $gitprefix to ls-files,
> which should also avoid the CWD problem.
> 
> The trigger of the fallback is broken anyways, isn't it?
> 
> >  # If ls-files succeeded but returned nothing, try again with no pattern
> >  if [[ -z "$files" && -n "$pref" ]]; then
> 
> It just considers the case where no match was found at all, but it does not
> handle the case where *some* matches were found, but not all.
> 
>