bug#33018: 26.1.50; thread starvation with async processes and accept-process-output

["Basil L. Contovounesios" <contovob@tcd.ie>]

Eli Zaretskii <eliz@gnu.org> writes:

>> From: "Basil L. Contovounesios" <contovob@tcd.ie>
>> Cc: <33018@debbugs.gnu.org>
>> Date: Sun, 14 Oct 2018 16:00:56 +0100
>> 
>> > When the hang happens, is there any wget process still alive, or did
>> > they all exit?  Please use OS tools to find that out, don't rely on
>> > what Emacs thinks.
>> 
>> When the hang happens, the wget process launched by the waiting thread
>> is still alive but asleep (idle), as reported by ps and top.
>
> Any idea why is that?

No idea, sorry.  I haven't yet found the time to look into this deeper.

> And if this is the situation, doesn't it explain why
> accept-process-output times out?

Possibly, but the question is why do we enter either of these situations
(process sleeping and accept-process-output timing out) exclusively when
using Emacs threads, no?

By the way, the problem isn't specific to wget; the same thing happens
with curl.  I also find it strange that the first thread doesn't suffer
from the same problem as the second thread, especially given each thread
is created only after the last thread exited.

-- 
Basil