[cpp-threads] High-level vs. low-level

[Lawrence Crowl]

I was working on cleaning up the high-level approach.  I now think
that would be counter-productive, due to the problems Peter and Herb
have outlined in communicating and reasoning about such code.

The low-level approach also has the advantage that the specification
should work equally well for C and C++, which will also save much
hand wringing.

As an aside, I mentioned to Hans that perhaps the low-level operations
should act on system-defined structures rather than basic types.  this
approach can deal with additional alignment constraints easily and
avoids inappropriately accessing an atomic variable with regular
operations.  The disadvantage is that a single variable cannot change
its atomicity over its lifetime.

-- 
Lawrence Crowl