From: cgf AT cygnus DOT com (Christopher Faylor)
Subject: RE: SIGKILL and block_sig_dispatch
15 Jun 1998 06:29:40 -0700
Message-ID: <199806151304.JAA13355.cygnus.cygwin32.developers@kramden.cygnus.com>
To: sos AT prospect DOT com DOT ru
Cc: cygwin32-developers AT cygnus DOT com

>From: Sergey Okhapkin <sos AT prospect DOT com DOT ru>
>Date: Mon, 15 Jun 1998 17:00:47 +0400
>
>Christopher Faylor wrote:
>> >> Are you proposing that do_exit *always* runs in the context of the signal
>> >> handler?
>> >> 
>> >
>> >Yes, if the signal must terminate a process.
>> 
>> void
>> killme(int sig)
>> {
>>     exit(sig);
>> }
>> 
>> int
>> main(int argc, char **argv)
>> {
>>     signal(SIGALRM, killme);
>>     alarm(5);
>>     pause();
>> }
>> 
>> How do you determine that the exit() above should run from the signal thread?
>
>
>Chris, once again - I'm talking about _uncatched_ signals!

Sergey, once again - I'm pointing out that it is still possible to hang
cygwin with something like the above.  You proposed a solution.  I have
been trying to point out that it is not foolproof.

Since you understand the problem better than anyone, I was hoping that you
might be able to come up with a foolproof solution.

cgf