Sorry, I accidentally sent v3 as a new thread. It is available here:
https://lore.kernel.org/git/20260912014609.535922-1-skariel@gmail.com/
On Fri, Sep 11, 2026 at 1:13 AM Patrick Steinhardt <ps@pks.im> wrote:
Show 113 quoted lines
>
> On Thu, Sep 10, 2026 at 07:55:28AM -0700, Ariel Keselman wrote:
> > Deleting a root ref queues a packed-ref transaction in the files
> > backend, even though root refs cannot be packed. For example, holding
> > .git/packed-refs.lock makes "git update-ref --no-deref -d AUTO_MERGE"
> > fail, whether or not AUTO_MERGE exists.
> >
> > This also affects post-commit cleanup, which deletes AUTO_MERGE after
> > updating HEAD. In a linked worktree with read-only shared metadata,
> > commit succeeds but cleanup reports a packed-refs.lock error. Deleting
> > CHERRY_PICK_HEAD and REVERT_HEAD is affected as well.
> >
> > Skip the packed transaction for root-ref deletions. Keep loose-ref
> > locking and packed-ref deletion for other refs unchanged.
> >
> > Test deleting a root ref with packed-refs.lock held, and check that a
> > transaction deleting both a root ref and a packed branch still fails
> > without changing either ref.
>
> Nit: this last paragraph doesn't really add any value, as it's trivially
> visible from the patch that we add tests.
>
> > Signed-off-by: Ariel Keselman <skariel@gmail.com>
> > ---
> > Thanks for the review, Patrick.
> >
> > Changes since v1:
> > - Keep the packed-ref deletion comment focused on its original purpose.
> > - Use one existing root ref and include a packed-refs file in the test.
> > - Drop the timeout overrides and redundant individual-ref cases.
> > - Add a transaction deleting a root ref and a packed branch together;
> > check that a held packed-ref lock causes failure and preserves both refs.
> >
> > AI assistance was used to generate the patch, tests, and commit message,
> > including this revision.
> >
> > The root-ref deletion regression fails without the fix. With the fix,
> > 123 test scripts / 4078 tests pass, along with 249 unit tests and t0600
> > with SHA-256 (one platform skip in t0600).
>
> Huh? I hope that _all_ tests pass with this, not only 4078, and I would
> assume that you verified that this is the case at least on your machine.
>
> > diff --git a/t/t0600-reffiles-backend.sh b/t/t0600-reffiles-backend.sh
> > index 74bfa2e9ba..65ca19e84b 100755
> > --- a/t/t0600-reffiles-backend.sh
> > +++ b/t/t0600-reffiles-backend.sh
> > @@ -519,4 +519,47 @@ test_expect_success 'symref transaction supports false symlink config' '
> > test_cmp expect actual
> > '
> >
> > +test_expect_success 'deleting a root ref does not lock packed-refs' '
> > + test_when_finished "rm -rf root-ref" &&
> > + git init root-ref &&
> > + (
> > + cd root-ref &&
> > + test_commit initial &&
> > + git pack-refs --all &&
> > + cp .git/packed-refs expect &&
> > + git update-ref AUTO_MERGE HEAD &&
>
> For added benefit we could even execute git-pack-refs(1) after having
> created AUTO_MERGE and then execute `test_path_is_file` for it just to
> prove that it really doesn't get packed. But other than that the tests
> look good to me.
>
> > + : >.git/packed-refs.lock &&
> > + git update-ref --no-deref -d AUTO_MERGE &&
> > + test_path_is_missing .git/AUTO_MERGE &&
> > + test_path_is_file .git/packed-refs.lock &&
> > + test_cmp expect .git/packed-refs
> > + )
> > +'
> > +
> > +test_expect_success 'deleting root and packed refs in one transaction requires packed-refs lock' '
> > + test_when_finished "rm -rf root-ref" &&
> > + git init root-ref &&
> > + (
> > + cd root-ref &&
> > + test_commit initial &&
> > + git update-ref refs/heads/packed-branch HEAD &&
> > + git pack-refs --all &&
> > + test_path_is_missing .git/refs/heads/packed-branch &&
> > + git update-ref AUTO_MERGE HEAD &&
> > + git rev-parse AUTO_MERGE refs/heads/packed-branch >expect &&
>
> We could strengthen this a bit by listing the state of all refs:
>
> git refs list --include-root-refs >expect
>
> > + cat >stdin <<-EOF &&
> > + start
> > + delete AUTO_MERGE
> > + delete refs/heads/packed-branch
> > + prepare
> > + commit
>
> We can drop start/prepare/commit here, those are optional. We can also
> drop the extra file and just write the data into git-update-ref(1)
> directly via the heredoc.
>
> > + EOF
> > + : >.git/packed-refs.lock &&
> > + test_must_fail git update-ref --no-deref --stdin <stdin 2>err &&
> > + test_grep "Unable to create .*packed-refs.lock" err &&
> > + git rev-parse AUTO_MERGE refs/heads/packed-branch >actual &&
> > + test_cmp expect actual
> > + )
> > +'
>
> Thanks!
>
> Patrick