Re: reference-transaction regression in 2.36.0-rc1

From: Bryan Turner <bturner@atlassian.com>
Date: Apr 13, 2022, 22:55 UTC
Message-ID: <CAGyf7-FGdciG+QF8NA_yi=o4B0RVPNiZM4LJZ610cDZzd3AjyQ@mail.gmail.com>
In-Reply-To: <xmqqilrc61te.fsf@gitster.g>

On Wed, Apr 13, 2022 at 3:48 PM Junio C Hamano <gitster@pobox.com> wrote:

Show 26 quoted lines

>
> Junio C Hamano <gitster@pobox.com> writes:
>
> > Ævar Arnfjörð Bjarmason <avarab@gmail.com> writes:
> >
> >> This does look lik a series regression.
> >
> > Yes.
> >
> >> I haven't had time to bisect this, but I suspect that it'll come down to
> >> something in the 991b4d47f0a (Merge branch
> >> 'ps/avoid-unnecessary-hook-invocation-with-packed-refs', 2022-02-18)
> >> series.
> >
> > With the issue that /var/tmp/.git can cause trouble to those who
> > work in /var/tmp/$USER/playpen being taken reasonably good care of
> > already, it seems this is the issue with the highest priority at
> > this point.
> >
> >> I happen to know that Patrick is OoO until after the final v2.36.0 is
> >> scheduled (but I don't know for sure that we won't spot this thread &
> >> act on it before then).
>
> Reverting the merge 991b4d47f0a as a whole is an option.  It might
> be the safest thing to do, if we do not to want to extend the cycle
> and add a few more -rc releases before the final.

I'm reviewing the changes in the patch series, but while I understand the visible behavior fairly well the internals of it aren't something I've spent a lot of time in. Compounding that, I'm down to my final 3 days with Atlassian before I start a new job, so I'm juggling trying to finish up all my work and/or hand things over to others.

Based on that, I definitely would not want work on 2.36 to hinge on me getting a patch up. I would _love_ to (finally) contribute something back to Git, and the community that has helped me so much over the last 10 years, but I can't say exactly when I'd be able to post something. There's another developer internally who is also looking at this, but it's been a number of years since he worked in C.

I guess all that's a long-winded way of saying a revert might be the more timely option, for decoupling resolving this issue from the 2.36 timeline. That would also mean Patrick would be able to review any fix that was posted, which seems valuable since he did the original work.

>
> Thanks.