Re: [PATCH v3 1/2] submodule: use the submodule repository to abbreviate commits
- From
- Srujith Poojary <poojarysrujith@gmail.com>
- Date
- Nov 11, 2026, 04:19 UTC
- Message-ID
- <CABh_KWjYc15OJjdZQPFPhLO0VRFhY4VJEgYxVaJoGRK284wiUQ@mail.gmail.com>
- In-Reply-To
- <xmqq7bikqvpm.fsf@gitster.g>
On Wed, Nov 11, 2026 at 1:56 AM Junio C Hamano <gitster@pobox.com> wrote:
Show 51 quoted lines
>
> "Srujith A Poojary via GitGitGadget" <gitgitgadget@gmail.com>
> writes:
>
> > diff --git a/submodule.c b/submodule.c
> > index 6fcb606f7e..1302c11ddf 100644
> > --- a/submodule.c
> > +++ b/submodule.c
> > @@ -543,6 +543,7 @@ static void show_submodule_header(struct diff_options *o,
> > struct commit **left, struct commit **right,
> > struct commit_list **merge_bases)
> > {
> > + struct repository *repo = sub ? sub : o->repo;
>
> Let's step back a bit. Use of any repository other than 'sub' is
> wrong as a way to correctly find suitable abbreviation of object
> names in the 'sub' repository. Now, the variable sub is NULL when
> the submodule is not initialized, in which case we won't know how
> many hexdigits we need to abbreviate "one" and "two" to, because
> there is no repository that has any object from that submodule to
> begin with.
>
> > @@ -599,9 +600,9 @@ static void show_submodule_header(struct diff_options *o,
> >
> > output_header:
> > strbuf_addf(&sb, "Submodule %s ", path);
> > - strbuf_add_unique_abbrev(&sb, one, DEFAULT_ABBREV);
> > + strbuf_repo_add_unique_abbrev(&sb, repo, one, DEFAULT_ABBREV);
> > strbuf_addstr(&sb, (fast_backward || fast_forward) ? ".." : "...");
> > - strbuf_add_unique_abbrev(&sb, two, DEFAULT_ABBREV);
> > + strbuf_repo_add_unique_abbrev(&sb, repo, two, DEFAULT_ABBREV);
>
> And that means the above patch is an utterly wrong rewrite, isn't
> it?
>
> Because using o->repo is just as wrong as using the_repository,
> replacing strbuf_add_unique_abbrev(), which uses the_repository,
> with a call to strbuf_repo_add_unique_abbrev() and using o->repo
> does not fix anything.
>
> So when (sub == NULL), the code probably should just grab the first
> DEFAULT_ABBREV hexdigits of one and two and be done with it.
> Guaranteeing uniqueness of the abbreviations within o->repo does not
> help anybody anyway, right?
>
> On the other hand, when (sub != NULL), the rewrite in this patch is
> absolutely the right thing to do. Thanks for working on it.
>
>
>
>Hi Junio,
Thanks for stepping back and explaining the root issue with submodule.c. You're right—when sub == NULL, o->repo doesn't hold object information for the submodule anyway, so querying o->repo for unique abbreviations is incorrect. In the next iteration, I will update show_submodule_header() to handle sub == NULL by using the standard DEFAULT_ABBREV digits directly without a repository abbreviation lookup, and keep the sub != NULL logic using the sub repository.
Thanks, Srujith