bug#60220: 29.0.60; macOS 13.1 crash shortly after starting Emacs

[Eli Zaretskii]

> From: Gerd Möllmann <gerd.moellmann@gmail.com>
> Date: Thu, 22 Dec 2022 06:36:43 +0100
> Cc: Eli Zaretskii <eliz@gnu.org>, 60220@debbugs.gnu.org
> 
> I was more thinking of something like this:
> 
> A SIGALRM handler is installed in the original process. SIGALRM continues to 
> be delivered to the new process after execve but the signal handler is now 
> the default handler which terminates the process. 
> 
> The man pages I mentioned should say somewhere if that’s plausible. It looks 
> to me like that could be what’s happening. But it’s a guess. 
> 
> If it is that, one would need to arrange for SIGALRM to be ignored before 
> execve and reinitialize a timers in the new process. Or something like that. 

Yes, I think our implementation of restart-emacs might be too naïve.

Paul, could you perhaps audit the code which implements restart-emacs,
and see if we need to make it safer, in particular wrt signals and
*.eln files loaded via dynlib.  Note that on Posix platforms we
currently load *.eln files with RTLD_LAZY and without RTLD_GLOBAL --
is this of any significance for "restarting" Emacs that was built with
native-compilation enabled and has *.eln files loaded?  Maybe we need
to unload the *.eln before calling execvp?

Or maybe we should consider re-implementing restart-emacs in some
different way, to avoid these problems?

Thanks.