[cpp-threads] Memory model counterproposal: synchronizedoperations

[Nelson, Clark]

> I'm also not sure that it's rare in regular C/C++.
> 
> Would the compiler benefit much from such a rule?  My mental
> model is that an auto variable which has had it's address
> taken (and the address has escaped sufficiently that the compiler
> can't track it) pretty much needs to be treated like a global anyway?

Yes of course that's true. The problems are: (1) How do we describe
"escaped sufficiently" in standardese? (2) How do we teach that rule to
programmers?

Clark