Re: a less-invasive racy-leak fix, was Re: What's cooking in git.git (Dec 2024, #11; Mon, 30)

From: Jeff King <peff@peff.net>
Date: Jan 2, 2025, 02:32 UTC
Message-ID: <20250102023224.GA3853144@coredump.intra.peff.net>
In-Reply-To: <xmqqa5cavz8h.fsf@gitster.g>

On Wed, Jan 01, 2025 at 04:25:02PM -0800, Junio C Hamano wrote:

Show 9 quoted lines

> Jeff King <peff@peff.net> writes:
> 
> > On Mon, Dec 30, 2024 at 09:33:20AM -0800, Junio C Hamano wrote:
> >
> >> * jk/lsan-race-with-barrier (2024-12-30) 5 commits
> > ...
> > This graduated faster than I expected. :)
> 
> Heh, it is before -rc2 and the change is only about tests, so ...

Yeah, I figured as much. I also considered it of relatively low importance during -rc, but I guess CI false positives do tend to annoy everybody and waste their time. :)

It looks like you pushed out the version of 'master' with it merged. I had figured you'd revert jk/lsan-race-with-barrier out of next, so I wondered how we would proceed (revert the whole merge from master to rebuild, or do a moral revert of the final three).

Looking at jk/lsan-race-ignore-false-positive, it looks like you did the moral revert via fc89d14c63 (Revert barrier-based LSan threading race workaround, 2025-01-01). That commit's tree matches what I'd expect (I guess you probably used "revert -n HEAD~3..HEAD" just like I did).

It would be nice if the 3-commit revert mentioned the specific commits it was reverting. If you revert the whole merge, you get the merge commit's id and you can find the original commits by inspecting the history (but of course here we were reverting only part of it, so we couldn't do that here). If you revert the sequence without "-n" you get three individual reverts. Which is informative, but a little clunky in the history.

I wonder if revert should have a "squash" mode that reverts all of the commits (perhaps in reverse order of application in case they depend on each other textually), and then gives you a commit message template similar to git-fmt-merge-msg, where we list all of the commits, one per line (though probably with their commit ids in this case).

Probably not a big deal either way, and certainly not a blocker for the series. Thanks (as usual) for doing all the maintainer juggling.

-Peff