> The following is a proposal (probably for external) that is an
> extension of Mark Snir's proposal. It would also, allow MPI I/O to
> continue to be layerable on top of MPI-2.
I am only a passer-by in this issue, but
I recently learned with a pain that gcc has "long double", and its
length is 12 bytes on ix86 machines, and I wonder how this is handled
in your naming scheme for data representation constants.
Koichi Konishi firstname.lastname@example.org
NEC Research Institute Office: 609-951-2628