>> I think that, in this case, MPI should mandate a complete syntax and
semantics for thread calls, which can be wrappers, or special
implementations, but again provide complete portability. Currently,
we will not have true portability of threaded programs. This will
limit MPI's usefulness in systems as threads become a dominant form
of local programming expression.
Or, we think that the current half-measures should be removed completely
so that the Forum cannot claim to have addressed MPI and threads properly.
Again, it will be presumptuous of the MPI forum to try to standardize
threads. Other bodies are doing that: POSIX is a standard for "portable
operating system" interfaces, that is endorsed by ANSI and IEEE. It is a
mandatory requirement on many bids. I don't think they need the
endorsement of the MPI forum. The MPI forum specifies the interaction of
threads with MPI. We can say that it specifies the interaction of pthreads
with MPI, but this does seem to be a useful restriction.