Re: [PATCH v2 00/21] Various memory leak fixes

From: Patrick Steinhardt <ps@pks.im>
Date: May 28, 2024, 05:09 UTC
Message-ID: <ZlVnJm4OcPGHdC3P@tanuki>
In-Reply-To: <xmqqwmnf9mto.fsf@gitster.g>

On Mon, May 27, 2024 at 10:38:59AM -0700, Junio C Hamano wrote:

Show 13 quoted lines

> Patrick Steinhardt <ps@pks.im> writes:
> >> Also
> >> 
> >>  https://github.com/git/git/actions/runs/9231313414/job/25401102951
> >> 
> >> shows that t1460-refs-migrate fails on Windows.
> >
> > Hm, this one is curious. There are no leak logs at all, and the exit
> > code is 139. Might be SIGSEGV, indicating that something else is going
> > on here than a memory leak.
> 
> Sorry, I wasn't clear enough.  I do not suspect this is about leaks
> (and the failing job on Windows is not about leaks).

I figured this one out now: Windows of course refuses to remove some of the files because they are kept open. This shouldn't lead to a segfault itself. But we call `free_ref_cache()` on a partially initialized files ref store, nad that function didn't handle the case when it is being passed a `NULL` pointer.

I've addressed this by releasing the reftable ref store before trying to remove it from disk so that all files are being closed. But it also surfaced another bug: our worktree code creates a copy of the main ref store by accident because we set up `wt->is_current` _after_ we have called `get_worktree_ref_store()`.

I fixed all of this and will send a new version in a bit.

Patrick