I have no particular objection to MPI-2 providing something like
MPI_Get_char_datatype().
My nervousness is just that I'm not at all clear how to make this
capture precisely the information that is actually required.
The problems here arise from two factors :-
1) an MPI datatype expresses both the type signature and the type
layout. You probably only want the type signature for your
applications, while other people may really want both parts of the
description.
2) an MPI datatype can only provide enough information to allow you to
recode data if you also know the source machine. (e.g. something
coded as an integer may occupy anywhere between 2 and 8 bytes
depending which machine and which compiler on that machine was used
when generating the data). Therefore an MPI datatype on its own is
not enough to allow you to unpack and re-format a message without
additional information.
-- Jim
James Cownie
BBN UK Ltd
Phone : +44 117 9071438
E-Mail: jcownie@bbn.com