[MPI3 Fortran] MPI Fortran bindings

[Jeff Squyres]

On Jun 4, 2009, at 2:06 PM, Lionel, Steve wrote:

> It could be documented as <any type and shape> (without the  
> dimension(*)) with a note that the actual mechanics of this are  
> implementation-specific.
>

I think that this is fine for 2 reasons:

1. We did something similar in the C++ bindings back in '96 when  
"namespace" wasn't fully implemented by all compilers.

2. More importantly, the current F77 bindings in MPI 1 through 2.2  
similarly punt on the type argument in all Fortran bindings:

MPI_SEND(BUF, COUNT, DATATYPE, DEST, TAG, COMM, IERROR)
    <type> BUF(*)
    INTEGER COUNT, DATATYPE, DEST, TAG, COMM, IERROR

So having some kind of neutral placeholder in the MPI spec already has  
a precedent -- all the way back to MPI-1.  We might need to add some  
verbiage about how it's implementation specific (I'll have to check;  
such language might already exist).

-- 
Jeff Squyres
Cisco Systems