Subj : Re: CondVar on Windows : what to do when a consumer thread is killed
To   : comp.programming.threads
From : Alexander Terekhov
Date : Fri Jul 01 2005 04:58 pm


adebaene@club-internet.fr wrote:
[...]
> During the singleton destruction, a special message is posted to this
> queue to signal the worker thread to stop, and posting to the queue
> implies calling pthread_cond_broadcast.
> The problem is that, since the singelton is a static object, it's
> destructor runs during CRT termination cleanup, *after* the system has
> forcibly killed all worker threads --> deadlock  :-(

So what's the point in calling pthread_cond_broadcast() at process 
termination time in the last thread ("uncontrolled" evaporation of 
other threads resulting in broken shared state aside for a moment)? 

"Doctor, it hurts when I do this. Doctor: Don't do that." ;-)

> 
> > A signal isn't guaranteed to generate
> > a spurious wakeup.
> What do you mean exactly? There is nothing left to wakeup since the
> consumer thread has been killed...
> 
> >  If you  used pthread cancel then your cleanup
> > handlers should have taken care of the problem.
> I am working on Windows and do not use pthreads, but Win32 threading
> API (in fact, my condvar hasn't got exactly the pthread_condvar
> interface - although it has the  same semantic).

All Schmidt's "condvars" are busted. See pthreads-win32.

regards,
alexander.

.