Re: [boost] condition Win32 based on events

From: Roland Schwarz (roland.schwarz_at_[hidden])
Date: 2005-09-12 08:37:24

Anthony Williams schrieb:

>With your changes to use InterlockedDecrement, it looks good. Off hand, I
>can't think of a reason why it would break --- I don't believe it suffers from
>deadlocks, lost wakeups, spurious wakeups, or unfairness.
>
>  
>
Thanks for looking into it.

About spurious wakeups: They are possible, but AFAIK they are considered 
to be quite
normal in condition/mutex paradigm. As you will anyway use condition 
variables in
a loop to reevaluate the state variables, this is no harm.

About unfairness: Since I do not try to control the order of the wakeups 
this is left to
the scheduler, and thread priorities respectively. Of course I do hold 
off any waiters,
that entered after a broadcast or signal. But this is by design.
But then: In my small example I am seeing that the done1,done2,done3 are 
considerably
unbalanced. I am not sure yet if this is an indication of unfairness, or 
just an artifact
of how I am trying to measure it.
Any ideas?

>You just need to extend it to work with other mutex types, now.
>  
>
This is what currently puzzles me somewhat. The condition documentation 
is not explicit
on whether a try_mutex is different upon return of the wait. It simply 
says "and then
reaquires the lock". On the other hand it would not be meaningful to 
return without
holding the lock again.

Also I feel a little surprised, that try and timed are attributed to the 
mutex, I
would have expected them properties of locks?

Regards,
Roland