On Nov 30, 2007, at 7:29 PM, William Yu wrote:
3.0
- better Java support/bindings beyond just one-to-one mapping of
primitives and functions?
FWIW, there was strong resistance to similar things for the C++
bindings for MPI 2.0. This was after considerable debate within the
Forum. FWIW, a) I was the one who originally pitched a C++ class
library for the MPI bindings, and b) I agree with the decision *not* to
have a class library but rather to have relatively simplistic C++
bindings that are more-or-less 1:1 mapped to their C equivalents.
The rationale was:
- For a standard, it is best to give the building blocks.
- The bindings the different languages should be more-or-less
equivalent. If they're not, then you've effectively got different
standards.
- Trying to define a class library (e.g., in C++) will necessarily
impart some interpretation and semantics which will, by definition, be
different than the bindings for the other languages.
- If a standard has building blocks, others can build class libraries
on top of it (don't forget that one of MPI's primary targets was to be
the underpinnings of higher-layer abstraction libraries). E.g.: OOMPI
(long-since dead), the Boost C++ MPI package, ...and I think there were
others.
So if I may be so bold -- I think the real proposal you're pitching is:
official Java bindings.
--
Jeff Squyres
Cisco Systems