Re: [PATCH v4] ls-files: introduce "--format" option

[ZheNing Hu <adlternative@gmail.com>]

Ævar Arnfjörð Bjarmason <avarab@gmail.com> 于2022年6月28日周二 02:48写道：
>
>
> On Sun, Jun 26 2022, ZheNing Hu via GitGitGadget wrote:
>
> > From: ZheNing Hu <adlternative@gmail.com>
> >
> > Add a new option --format that output index enties
> > informations with custom format, taking inspiration
> > from the option with the same name in the `git ls-tree`
> > command.
> >
> > --format cannot used with -s, -o, -k, --resolve-undo,
> > --deduplicate and --eol.
> >
> > Signed-off-by: ZheNing Hu <adlternative@gmail.com>
> > ---
> > [...]
> > +test_expect_success 'git ls-files --format with --debug' '
> > +     git ls-files --debug >expect &&
> > +     git ls-files --format="%(path)" --debug >actual &&
> > +     test_cmp expect actual
> > +'
> > +
> > +test_done
>
> I'm not sure what to make of this.
>
> In some ways I think this makes more sense than what I suggested in
> https://lore.kernel.org/git/220624.86letmi383.gmgdl@evledraar.gmail.com/;
> but I had to think for a second about what's going on here.
>
> In my version I suggested having this work with --debug, but not in this
> way, in my version you'd always emit the debug output, and the format
> output.
>
> But here e.g.:
>
>     git ls-files -t --debug
>
> Will emit "H tag.c" or whatever, but if you add --format the -t option
> is silently discarded.
>
> So the test is relying on "%(path)" being the default format.
>
> I think extending this to e.g. test what happens with "-t" would be a
> good thing, but also in general does combining --format with -t make
> sense, and are there other such options where the combination might not
> make sense?
>

Why not we just let -t incompatible with --format? Is this because -t can also
be considered a “debug” message, and we often use --debug and -t together?

If so, we can just do something like:

@@ -238,6 +335,13 @@ static void show_ce(struct repository *repo,
struct dir_struct *dir,
                                  S_ISGITLINK(ce->ce_mode))) {
                tag = get_tag(ce, tag);

+               if (format) {
+                       fputs(tag, stdout);
+                       show_ce_fmt(repo, ce, format, fullname);
+                       print_debug(ce);
+                       return;
+               }
+


> So I'm not 100% sure, but I think I'd prefer my version, but I see how
> it would get hairy to support, e.g.:
>
>     git ls-files -s --debug --format=...
>
> Should work, but you'd have to special-case the logic for erroring if -s
> is combined with --format.
>

Agree. it's really weird.

> Anyway, I think it would be fine to leave this in whatever state is
> easy, the --debug option "just for debugging".
>
> But re
> https://lore.kernel.org/git/CAOLTT8Tc95-aUE+uN2d8QjTJpGpGw6cBJfG+bpmyE55OcXTSRA@mail.gmail.com/
> I think it might be interesting to get --format to a state where we can
> remove --debug entirely.
>
> I.e. in c2a29405105 (t1091/t3705: remove 'test-tool read-cache --table',
> 2021-12-22) we could replace some similar test-only code with "git
> ls-files". I for one wouldn't mind --debug going away entirely, and have
> the t3705-add-sparse-checkout.sh tests use --format instead.
>
> Or we could keep --debug, but just have it powerful enough to do what
> print_debug() is doing now, possibly without "truly internal" stuff like
> "ce_flags".

Ah, though we just remove these little "useless" atoms, maybe we can add
them back later? (not in this patch?)

ZheNing Hu