Alexander Terekhov wrote:
[...]
> void release_strong() {
> int status = pthread_refcount_decrement(&strong_count);
> if (PTHREAD_REFCOUNT_DROPPED_TO_ZERO == status) {
> destruct_object();
> release_weak();
> }
> }
> bool acquire_strong_from_weak() {
> int status = pthread_refcount_enroll_one(&strong_count); // _many( &refcount, 1);
> if (PTHREAD_REFCOUNT_DROPPED_TO_ZERO == status) // Ouch, did not work [too late].
> return false;
> return true;
> }
Interestingly. pthread_refcount_enroll_one() in your implementation
really eliminates need in any explicit locking.
Your code showed also that I overlooked the more simple way to deal
with weak_count. To satisfy aesthetic feelings I place here the
corrected version of my previous code.
Pavel
-----
// If atomic_decrement() sets read memory barrier when returns 0,
// there is no more need to care about cache coherency
// (from http://www.boost.org/boost/detail/atomic_count.hpp ).
class weak_ref_control_block
{
...
weak_ref_control_block() : strong_count(1), weak_count(1) {}
void strong_refs_lost()
{
{
scope_lock lock(mutex_destruct);
if(atomic_query(&strong_count) != 0)
return;
// removed acquire_weak()
atomic_set(&strong_count, atomic_int_type_MIN);
}
destruct_allocated();
release_weak();
}
void release_weak()
{
if(atomic_decrement(&weak_count) == 0)
destruct_self();
}
}
_______________________________________________
Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost
