bug#13546: 24.2.92; Error(s) when sending emails

[Eli Zaretskii]

> From: Thierry Volpiatto <thierry.volpiatto@gmail.com>
> Date: Thu, 14 Feb 2013 12:01:54 +0100
> 
> Helm is starting updating (create the first process) after the
> `requires-pattern' value of source, for locate it is 3, so when entering
> the third char in pattern the process start, return the data through a
> process-filter, then stop the process-filter (setq *filter t) and kill
> the process. When you enter the 4th char, a new process is created,
> killed and so on.

The "kill process" part was an important clue, thanks.  Using that, I
was able to re-create a situation similar to what happens for
Sebastien.  I installed a few changes on the emacs-24 branch to
prevent these problems, so I hope the next pretest will behave better.

FWIW, the improved handling of subprocesses on the trunk avoids these
problems altogether, at least in the scenario I used to re-create
them.

(There's still some strange issue with gnutls calls, as reported by
Sebastien, which I'm still looking in.  Not sure if this is an
additional problem or just "how things should work".)

> This was just for info in case that help for debugging.

Thanks, it helped.

Btw, I'm surprised that helm starts an async subprocess, just to kill
it after its output was collected.  Isn't this exactly what
synchronous subprocesses are for?  What is the purpose of such a
strange design? (except if you wanted to help find bugs in Emacs when
async subprocesses are launched and killed at high frequency ;-)