On Feb 18, 2008 1:03 AM, Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
> You recently changed gdb.threads/sigthread.c to use barriers.
> Unfortunately the pthread_barrier_xxx() interfaces are not found in
> the standard POSIX threads standard, but part of the Advanced Realtime
> Threads POSIX extension.  Therefore they are not widely available.

Yeah, I mentioned this concern when I posted the patch.  I guess we know now.

> Could you change the test to use a more standard synchronisation
> mechanism?  I guess having a global variable of type volatile
> sig_atomic_t, setting it to a non-zero value in main() (after the
> threads have been created) and checking the value in child_two() and
> thread_function() (perhaps calling sched_yield() if the variables
> isn't set yet) would do the trick.

I'll bet this isn't the only test in the test suite that would like
some barriers.  How does the attached look?

> I can roll you a diff to do this if you want, but I don't think I have
> access to a system that shows the problem you tried to fix.

I developed the fix on a Core 2 Duo laptop running Linux, running
'while true; do :;done' in a few windows while running the test over
and over again as 'while runtest sigthread.exp; do :;done'.  Or
something like that.  It sometimes took a half-dozen iterations or so,
but it would fail eventually.