[PATCH] Possible threads race condition in src/alarm.c

[Andy Dougherty]

On Tue, 28 Aug 2012, Andrew Whitworth wrote:

> Yes, that patch actually does look very reasonable. Does it change the
> test results you've seen, at all?

It's too early to say -- it'll probably take until tomorrow afternoon for 
me to have chance to run the full test suite on any "interesting" 
machines.  A quick check on OpenBSD shows that it no longer hangs in the 
t/pmc/timer.t test, so I think it's on the right track.

> On Tue, Aug 28, 2012 at 10:34 AM, Andy Dougherty <doughera at lafayette.edu> wrote:
> > I've never really done any threads programming, so I could be quite off
> > here, but it looks to me as if there's a race condition in src/alarm.c in
> > the threads branch.  Specifically, Parrot_alarm_init() creates a thread
> > that checks sleep_cond before it initializes sleep_cond.  (Similar remarks
> > hold for alarm_lock.)
> >
> > Does this patch look appropriate?
> >
> > diff --git a/src/alarm.c b/src/alarm.c
> > index 298387f..0ec6a1f 100644
> > --- a/src/alarm.c
> > +++ b/src/alarm.c
> > @@ -56,9 +56,9 @@ Parrot_alarm_init(void)
> >  {
> >      ASSERT_ARGS(Parrot_alarm_init)
> >      Parrot_thread thread;
> > -    THREAD_CREATE_JOINABLE(thread, Parrot_alarm_runloop, NULL);
> >      MUTEX_INIT(alarm_lock);
> >      COND_INIT(sleep_cond);
> > +    THREAD_CREATE_JOINABLE(thread, Parrot_alarm_runloop, NULL);
> >  }
> >
> >  /*
> >
> > --
> >     Andy Dougherty              doughera at lafayette.edu
> > _______________________________________________
> > http://lists.parrot.org/mailman/listinfo/parrot-dev
> 

-- 
    Andy Dougherty		doughera at lafayette.edu