On 29 Dec 2000, Alexandre Julliard wrote:
> Ove Kaaven <[EMAIL PROTECTED]> writes:
>
> > Seems Alexandre applied my first signal_i386 patch. So here's something
> > slightly less likely to attract cheers... blocking of all signals until
> > the %fs is set in the signal handler. If an async signal occurs inside a
> > previous signal, before the previous has set its fs, then the new signal
> > won't be able to figure out which %fs to use... crash. So what could I do?
>
> I'd find it cleaner to only block the signals that can actually cause
> trouble (SIGALRM, SIGINT, SIGIO), and to not unblock them at all
> during signal handling.
Not unblock them? But wouldn't that cause problems when some signal
handlers throw exceptions, and the exception handler decides to take the
exception and thus force a stack unwind? I assumed that if that happened,
then the normal sigaction mechanism wouldn't get its own chance to
unblock. Or how does this work?
