Re: [PATCH 3/4] builtin/refs: add "update" subcommand
- From
Patrick Steinhardt <ps@pks.im>
- Date
- Jun 17, 2026, 07:28 UTC
- Message-ID
- <ajJMnZchqdpiuKTg@pks.im>
- In-Reply-To
- <xmqqeci6bupk.fsf@gitster.g>
On Tue, Jun 16, 2026 at 04:17:27AM -0700, Junio C Hamano wrote:
Show 24 quoted lines
> Patrick Steinhardt <ps@pks.im> writes: > > > Add a new "update" subcommand which mirrors `git update-ref <refname> > > <oldoid> <newoid>`. This follows the same reasoning as the preceding > > commit. > > > > Signed-off-by: Patrick Steinhardt <ps@pks.im> > > --- > > Documentation/git-refs.adoc | 7 ++ > > builtin/refs.c | 50 +++++++++++++ > > t/meson.build | 1 + > > t/t1465-refs-update.sh | 179 ++++++++++++++++++++++++++++++++++++++++++++ > > 4 files changed, 237 insertions(+) > > I do not offhand know (and I am still away by 2 hours from the time > I wake up and start functioning) if update-ref shares the same > issue, but with "delete, update, rename" combo, lack of "create" > feels a bit annoying. Wouldn't we want to offer an option to users > who want to ensure that the refs they create are truly new and they > are not overwriting a ref somebody has created? Either (1) drop > "delete" and take a special value (e.g. "") as <newvalue> to signal > deletion and make the same special value used as <oldvalue> signals > creation, or (2) add "create" and insist that "update" takes only an > existing ref, would make the annoyance go away, I guess.
In theory "update" can handle both updating existing references, deleting them and creating them race-free by providing NUL object IDs for either old or new value. But I agree that this is cumbersome, and adding another "create" subcommand as an easy-to-understand shortcut feels sensible.
Show 14 quoted lines
> > +test_expect_success 'update creates a new reference' ' > > + test_when_finished "rm -rf repo" && > > + setup_repo repo && > > + ( > > + cd repo && > > + A=$(git rev-parse A) && > > + git refs update refs/heads/foo $A && > > + test_ref_matches refs/heads/foo "$A" > > + ) > > +' > > Here we cannot test (and I strongly suspect that "git refs update" > and "git update-ref" lack ability to do so) a case where a creation > is attempted on an existing ref and fails.
We can:
$ git update-ref $NEW_OID $NULL_OID
$ git refs update $NEW_OID $NULL_OIDThis will verify that the reference doesn't exist before actually writing it. Will add a test.
Patrick