On Sep 28, 2004, at 5:23 AM, hedman_at_[hidden] wrote:
> As far as I can understand when STL came on the scene very
> late in the C++ standardization process it caused a big
> shift in the perception of how an interface to a library
> could be structured. Clearly Generic Programming and the
> idea of "concepts" to define an interface could have a big
> impact on how a MPI C++ library could be defined. With more
> and more mature C++ compilers available and Generic
> Programming becoming widely used I don't think it is too
> farfetched to hope or wish for a kind of Technical
> Corrigendum for MPI2. To me this adds up to an opportunity
> for a new and improved interface for C++.
There's really nothing preventing you from implementing such a scheme
-- combined with several of the ideas on this thread, it could be quite
an interesting abstraction and potentially a useful package for others.
We originally implemented the C++ bindings as an
MPI-implementation-neutral package -- since it used the C bindings on
the bottom, it could be used with any MPI implementation. OOMPI was
developed the same way. If these kinds of issues interest you, I would
encourage you to explore them and make a portable package that others
can use.
This would probably be the best way to proliferate new C++ directions
into MPI.
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
