Subj : Re: Consistency of a byte shared variable
To   : comp.programming.threads
From : Joseph Seigh
Date : Tue Feb 01 2005 07:32 am

On Tue, 1 Feb 2005 01:32:27 -0800, SenderX <xxx@xxx.com> wrote:

>> Could the example "ever" abort in B3 under the POSIX memory visibility
>> rules?
>
> I currently assume POSIX lays out a memory model based on stringent
> acquire/release semantics:
>
>
> sem_post( ... ) == release barrier "before" sema inc is made visible
>
>
> sem_wait( ... ) == acquire barrier "after" sema dec is made visible
>
>
> Does POSIX standard "explicitly" guarantee this type of straight forward
> memory model? If so, where is it documented?
>
Posix doesn't explicity guarantee anything.  Posix decided not to formally
document any memory visibility semantics.  Remember?

Semaphores probably do have acquire/release semantics.  If you subscribe to
the notion that you need to use a mutex anyway, sem_post becomes pretty
useless in a signal handler.  And somebody will have to write a "Single
Producer - Single Consumer with Semaphores is Broken" declaration.


-- 
Joe Seigh

.