Re: [PATCH v3 1/1] completion: graduate out of contrib

[Felipe Contreras <felipe.contreras@gmail.com>]

Ævar Arnfjörð Bjarmason wrote:
> On Thu, Jul 15 2021, Felipe Contreras wrote:
> > Ævar Arnfjörð Bjarmason wrote:
> >> On Thu, Jul 15 2021, Ævar Arnfjörð Bjarmason wrote:
> >> > On Wed, Jul 14 2021, Felipe Contreras wrote:
> >> >> [...]
> >> >> @@ -2399,7 +2399,7 @@ test_expect_success 'sourcing the completion script clears cached --options' '
> >> >>  	verbose test -n "$__gitcomp_builtin_checkout" &&
> >> >>  	__gitcomp_builtin notes_edit &&
> >> >>  	verbose test -n "$__gitcomp_builtin_notes_edit" &&
> >> >> -	. "$GIT_BUILD_DIR/contrib/completion/git-completion.bash" &&
> >> >> +	. "$GIT_BUILD_DIR/extra/completion/git-completion.bash" &&
> >> >>  	verbose test -z "$__gitcomp_builtin_checkout" &&
> >> >>  	verbose test -z "$__gitcomp_builtin_notes_edit"
> >> >>  '
> >> >>
> >> >> diff --git a/t/t9903-bash-prompt.sh b/t/t9903-bash-prompt.sh
> >> >> index bbd513bab0..784e523fd4 100755
> >> >> --- a/t/t9903-bash-prompt.sh
> >> >> +++ b/t/t9903-bash-prompt.sh
> >> >> @@ -10,7 +10,7 @@ export GIT_TEST_DEFAULT_INITIAL_BRANCH_NAME
> >> >>  
> >> >>  . ./lib-bash.sh
> >> >>  
> >> >> -. "$GIT_BUILD_DIR/contrib/completion/git-prompt.sh"
> >> >> +. "$GIT_BUILD_DIR/extra/completion/git-prompt.sh"
> >> >>  
> >> >>  actual="$TRASH_DIRECTORY/actual"
> >> >>  c_red='\\[\\e[31m\\]'
> >> >
> >> > It's more of a "for bonus points", but a nic way to round-trip this
> >> > would be to make this work with GIT_TEST_INSTALLED.
> >> >
> >> > I.e. source these relative to GIT_EXEC_PATH, not $GIT_BUILD_DIR, I think
> >> > that just sourcing them as e.g.:
> >> >
> >> >     . git-completion.bash
> >> >
> >> > But the GIT_TEST_INSTALLED case is tricker, maybe we'd need to add a
> >> > "git --share-path" :(
> >> 
> >> I forgot to include this not-working patch, i.e. I've got no idea what
> >> the "something" should be other than the harder thing of compiling
> >> "sharedir" into git and making "git --share-path" work.
> >
> > I don't think there's a "something" that would make sense.
> >
> > It would be something like '/opt/git/bin/completion'?
> 
> I think more importantly if they're going to be "first-class" components
> that we have some native way of getting them to the user.

Yes, I already said --share-path would make sense [1] in your other
reply.

> I.e. once you install git being able to load them in your shell as:
> 
>     . git-path
>     . git-completion
> 
> Or whatever, which means either putting them in $PATH (i.e. we'd drop
> them in bin/ along with the non-dashed-built-ins like git-upload-pack
> etc.), or something like:
> 
>     . "$(git --completion-path)"/bash
>     . "$(git --completion-path)"/zsh
> 
> Or maybe:
> 
>     . "$(git --extras-path)"/completion/bash.sh
>     . "$(git --extras-path)"/prompt.sh
> 
> ?

But that's not how completions should work.

There's a standard location for bash completions in order to be picked
by bash-completion:

  /usr/share/bash-completion/completions

This will be the case regardless of what --extras-path is.

The user *should not* be sourcing anything from there directly,
bash-completion does that automatically.

This location depends by distribution, and the standard way to figure it
out is:

  pkg-config --variable=completionsdir bash-completion

If you don't have bash-completion insalled, *then* you would need to
source the completion file directly, but you cannot use pkg-config to
figure out that location, so you would need to use the default
(/usr/share/bash-completion/completions).

We could install the completion twice (or provide a symlink):

  /usr/share/bash-completion/completions/git
  /usr/share/git-core/completion/bash.sh

So if the user doesn't have bash-completion installed, can do:

  source "$(git --extras-path)"/completion/bash.sh

Instead of the standard:

  source /usr/share/bash-completion/completions/git

But *right now* the standard location is standard, distributions are
using it, and users are using it.

Moreover, it seems a bit wasteful to have --exec-path, --html-path,
--man-path, --info-path, --extra-path. Why not have --path=extra
instead?

Once again, I think this is a good idea, but it should be done
separately, and you yourself said "for bonus points".


Not to mention my strong feeling that *even* if I implement this nice
feature, Junio will still ignore this series (as he ignores everything I
send), so I will be simply wasting my time implementing something that
will never be merged (just like he stood on the sidelines watching us
implement 6 versions of the man pager colorize patch just to say "no
thanks" [2]).

I'd be more than happy to add any other suggestions you might have, but
let's be honest: this is a patch we all will have be carrying ourselves
(I have dozens of those), so I'd rather not add more complexity for a
perfect solution that won't be applied.

I'd rather aim for something that is good enough.

Cheers.

[1] https://lore.kernel.org/git/60f0859399369_519c2083c@natae.notmuch/
[2] https://lore.kernel.org/git/xmqq4ke8pig9.fsf@gitster.g/

-- 
Felipe Contreras