bug#49682: 27.2.50; accept-process-output within accept-process-output hangs emacs

[Dmitry Gutov <dgutov@yandex.ru>]

On 22.07.2021 08:47, Eli Zaretskii wrote:
>> From: Lars Ingebrigtsen <larsi@gnus.org>
>> Cc: Rajeev N <rajeev.jnk@sivalik.com>,  49682@debbugs.gnu.org
>> Date: Wed, 21 Jul 2021 23:32:16 +0200
>>
>> Eli Zaretskii <eliz@gnu.org> writes:
>>
>>>> (let () (run-with-timer 0 nil #'url-retrieve "https://www.gnu.org/"
>>>> #'ignore) (diary-mail-entries))
>>>
>>> Don't do that, because it will hurt.  Mixing timers with subprocesses
>>> is asking for trouble.
>>
>> That's not my experience, really -- url-queue.el is based around firing
>> off `url-retrieve' from timers, and I've never seen any problems.
> 
> Sheer luck, IMO.
> 
> I don't really understand the need to run subprocesses from timers,
> since subprocess support already includes async operation and built-in
> time-outs.

This is a fairly common configuration. Consider a minimal IDE setup:

1. Code completion popup which is shown when Emacs is idle for XX ms.
2. Eldoc hints which are shown at YY ms after the user stops typing.

When both are implemented using HTTP calls, you get "timers with 
subprocesses".

In fact, I have a configuration like that (based on url.el) and get an 
occasional strong freeze (one that requires 3 C-g's to get out of) about 
once every 1-2 weeks. Haven't found a solid repro so far, so if the 
reporter has something stable-ish, I'll be following this report intently.