Re: Threads: better condition variables?

From: William Kempf (sirwillard_at_[hidden])
Date: 2000-08-10 14:39:43

--- In boost_at_[hidden], jsiek_at_l... wrote:
> 
> So one thing that bothers me about pthreads is the complexity of the
> usage scenario for condition variables. Have any of you seen ways to
> simplify this?
> 
> Just as a reminder, here's how the usage goes:
> 
> Thread 1:
> 
>   pthread_mutex_lock(&m);
>   while (!some_condition)
>     pthread_cond_wait(&cv, &m);
>   // now some_condition == true
>   // do something useful
>   pthread_mutex_unlock(&m);
> 
> 
> Thread 2:
> 
>   pthread_mutex_lock(&m);
>   some_condition = true;
>   pthread_mutex_unlock(&m);
>   pthread_cond_signal(&cv);
> 
> 
> Maybe there's a way to hide that while loop if a Predicate function
> object is attached to the condition variable?

What really bothers me is the need to obtain the lock before calling 
the wait but with out compile time requirements for this.  I'd really 
like a compile time requirement in our design (if we have mutexes and 
conditions... it sounds like this will be debated).

As for the predicate to remove the while...  I considered that a lot 
yesterday while thinking on this subject.  The only problem I see is 
that you don't always have a predicate or while loop.  Sometimes the 
signal is enough.  Other than that, the idea appeals to me.  It makes 
the code a little easier to read and understand (especially using 
lambda library), though it's not really any easier to write.  If we 
always require the predicate it would also insure that someone new to 
the concept is more likely to use the condition correctly, though it 
would require a "dummy predicate" in the corner cases where a 
predicate is not needed.

I wouldn't attach it to the condition, however.  I'd pass the 
predicate into the wait.  Thus the condition can be used to track 
different criteria.

   cond.wait((free1 > 10)(a));

William Kempf