[cpp-threads] Visibility question

[Nelson, Clark]

> Is my understanding correct?

I'm sure that you have produced a scholarly summary of what the
semantics of fences would/should be. However, I'm obliged to point out
that <fence> was not actually among the atomic ordering constraints
proposed -- at least not formally to the C++ committee in N2047.

Full disclosure: there are a couple of fence functions proposed therein
for the low (i.e. C) level API.

Clark