bug#36609: 27.0.50; Possible race-condition in threading implementation

[dick . r . chiang]

I enter a plea of lacking the programming know-how for atomics.

I stand by everything I've said thus far: an obfuscated MRE led to a
misdiagnosis, led to an unnecessary and highly thread-unsafe fix, which can be
remedied by the vulgar "end-run" patch I submitted, and which I test every day
with parallel-thread Gnus.

Unfortunately in this and every urological contest of wills I seem to have with
project maintainers, I, by construction, must always lose.

>>>>> Eli Zaretskii <eliz@gnu.org> writes:

>> Date: Fri, 11 Jun 2021 09:00:07 +0300
>> From: Eli Zaretskii <eliz@gnu.org>
>> Cc: 36609@debbugs.gnu.org
>> 
>> I'm okay with replacing the simple increments and decrements of that
>> variable with atomic ones, using the likes of __atomic_fetch_add
>> intrinsics of GCC, if that solves the problem.  Did you try that?

> Ping!

> Did you try that, and if so, did using the atomics solve the original
> problem?

> Thanks.