From: Andy Glew (glew_at_[hidden])
Date: 2000-01-07 15:46:27
> No matter how wonderful such a method is and how useful
> it could be to implement, say, synchronized containers,
> my point was that using mutexes to synchronize shared_ptr
> reference counting is a huge overkill - instead of changing
> one instruction from simple increment to atomic one,
> you would _add_ new data members and mutex syncronization
> which is much more that atomic increment. Also, shared_ptr
> does not neeed to be fully synchronized, only assignment does.
So can we not specialize, then:
template<class T>
class mutex< shared_ptr<T> > : public shared_ptr<T> {
public:
....
};
No, you're right... G++ fails this.
You can specialize on a case by case basis:
template<>
class mutex< shared_ptr<int> > : public shared_ptr<int> {
public:
type mutex< shared_ptr<int> > m;
m& operator=( const m& that ) {
atomic_increment( ... )
...
}
};
but that leaves a lot to be desired.