[cpp-threads] Weak vs Strong Atomics

[Peter Dimov]

Raul Silvera wrote in message c++std-ext-9553:

> The new proposal using a derived type allows reasonable C++ access to
> weak atomics while still satisfying Herb's requirements, so I think
> it is a reasonable compromise.

The problem is that N2145 has

    atomic<int> IS-A atomic_int

and the C API on atomic_int is not SC-only. Adding a weak_atomic<int> that 
IS-A atomic<int> does not change the fact that the C API will work on 
atomic<int> because of the inheritance (which is a fundamental part of the 
current N2145 design). Something has to give.