bug#554: OSX: with-temp-buffer kills unrelated processes

[YAMAMOTO Mitsuharu <mituharu@math.s.chiba-u.ac.jp>]

>>>>> On Sun, 13 Jul 2008 19:39:57 +0200 (CEST), Markus Triska <markus.triska@gmx.at> said:
--text follows this line--
> Let w.el consist of the following forms:
>  (start-process "bc1" (current-buffer) "bc")

>  (with-temp-buffer (start-process "bc2" (current-buffer) "bc"))

> When I do "emacs -Q w.el", and press C-M-x on the first form, "bc"
> (the GNU arbitrary precision calculator) is started and puts its
> start-up message at the end of the buffer (you can also use any other
> program that waits for further input). When I then press C-M-x on the
> second form, the first process receives SIGHUP, and the message reads:

>  "Process bc1 hangup"

> I can reproduce this with Emacs 22.2 and latest CVS on OSX. I tried to
> trace it in gdb by putting a breakpoint at process_send_signal, but
> couldn't reproduce it in the debugger. Otherwise it works every time.

Just some analysis: not a solution I'm afraid.

It seems that the child process receives SIGHUP before it calls
execvp.  As the child process is still running the Emacs executable,
it sends SIGHUP to all the subprocesses when it receives SIGHUP
(fatal_error_signal -> shut_down_emacs -> kill_buffer_processes).

A similar problem can be found in following example:

  (process-running-child-p (start-process "bc1" (current-buffer) "bc"))
  => t

  (process-running-child-p (prog1 (start-process "bc1" (current-buffer) "bc")
			     (sleep-for 1)))
  => nil

I think these behaviors are not observable in other platforms, because
they use vfork, and the parent side effectively blocks until the child
side calls execvp or others in its typical implementations.  Emacs on
Darwin is using fork instead of vfork because the latter had (or
have?) some problems I'm not familiar with.

				     YAMAMOTO Mitsuharu
				mituharu@math.s.chiba-u.ac.jp