[RFH] "git -C there add foo" completes, s/add/diff/ does not

[RFH] "git -C there add foo" completes, s/add/diff/ does not

[Junio C Hamano]

As some of you may already know, I keep an untracked directory
called "Meta" at the top-level of the working tree of the Git
source tree.  This "Meta" directory is actually a single-branch
clone of the git.kernel.org/pub/scm/git/git.git that checks out
its "todo" branch, where files like whats-cooking.txt lives.

So, what I often would do is

    $ git -C Meta add whats-cooking.txt

after updating the draft of the next issue of the "What's cooking"
report.  The command line completion support for "git add" knows how
to complete this when I stopped typing the above after whats-" and
hit <TAB>.  It seems that __git_find_repo_path helper function that
notices "-C there" and discovers the $GIT_DIR, and _git_add helper
uses __git_complete_index_file that honors the discovered $GIT_DIR
to find paths in the correct index, which is wonderful.

But the same does not work for the step before I can decide to
actually "add" the contents, which is to "diff", i.e.

    $ git -C Meta diff whats-<TAB>

does not complete.

Anybody wants to take a crack at it?

Thanks.

Re: [RFH] "git -C there add foo" completes, s/add/diff/ does not

[Kyle Lippincott]

On Wed, Feb 7, 2024 at 10:35 AM Junio C Hamano <gitster@pobox.com> wrote:
>
> As some of you may already know, I keep an untracked directory
> called "Meta" at the top-level of the working tree of the Git
> source tree.  This "Meta" directory is actually a single-branch
> clone of the git.kernel.org/pub/scm/git/git.git that checks out
> its "todo" branch, where files like whats-cooking.txt lives.
>
> So, what I often would do is
>
>     $ git -C Meta add whats-cooking.txt
>
> after updating the draft of the next issue of the "What's cooking"
> report.  The command line completion support for "git add" knows how
> to complete this when I stopped typing the above after whats-" and
> hit <TAB>.  It seems that __git_find_repo_path helper function that
> notices "-C there" and discovers the $GIT_DIR, and _git_add helper
> uses __git_complete_index_file that honors the discovered $GIT_DIR
> to find paths in the correct index, which is wonderful.
>
> But the same does not work for the step before I can decide to
> actually "add" the contents, which is to "diff", i.e.
>
>     $ git -C Meta diff whats-<TAB>
>
> does not complete.

I'm not a completions expert, but I think what's happening is that the
completions for `git diff` aren't producing anything, so it (where
"it" here might be the shell?) falls back to just doing normal path
completion. For `git add`, it's checking the `git status` output to
filter the list to things that need to be added, so it respects the
`-C` option when calling into git to get that list, but there's no
such logic for `git diff` (the git-specific logic treats the
[optional] positional argument as a ref, not a file).

>
> Anybody wants to take a crack at it?
>
> Thanks.
>
>

Re: [RFH] "git -C there add foo" completes, s/add/diff/ does not

[Junio C Hamano]

Kyle Lippincott <spectral@google.com> writes:

>> ...
>> But the same does not work for the step before I can decide to
>> actually "add" the contents, which is to "diff", i.e.
>>
>>     $ git -C Meta diff whats-<TAB>
>>
>> does not complete.
>
> I'm not a completions expert, but I think what's happening is that the
> completions for `git diff` aren't producing anything, so it (where
> "it" here might be the shell?) falls back to just doing normal path
> completion.

Yes, that seems to be the case.

> For `git add`, it's checking the `git status` output to
> filter the list to things that need to be added,...

Not exactly, but a close enough description, I think.

__git_complete_index_file does not run `git status` but asks
ls-files the paths it knows about (including "--others" so that a
path that is untracked can become a candidate to be added), then
massages the list of paths with a custom awk script.

For "git diff", depending on what two sets of contents are being
compared, the source of possible paths may differ, but the list of
paths obtained from ls-files (without --others) would be appropriate
when comparing the index and the working tree files, or comparing a
tree-ish and the working tree files.  The necessary ingredient to do
so may be pretty much the same as what is used by _git_add completion.

When comparing two tree-ishes, the candidate would ideally come from
union of paths in these two tree-ishes, but I offhand do not know if
there already is a support for choosing from such a set.

Thanks.