[MPI3 Fortran] Deprecate mpif.h?

[N.M. Maclaren]

On Mar 4 2010, Supalov, Alexander wrote:
>
> We can argue for ages here. I suggest that a more constructive way is to 
> identify what parts of the MPI-3 may be of interest to the long time 
> Fortran users who may not want or be able to move to Fortran 2008. NB 
> collectives is a good starter. What else?

Most of the people I speak to are more interested in getting the advantages 
of Fortran 2008 + TR than in using non-blocking collectives. Our experience 
is that fully-synchronised collectives (i.e. preceded by barriers) are 
rarely much slower, quite often significantly faster, and VASTLY easier to 
debug.

> If we identify enough points of interest, and their addition to the 
> mpif.h and the respective mpi module will be trivial, I reckon we'll do 
> the standard and the user community a better service if we just extend 
> both accordingly.

My guess is that will be limited to a few constants and similar.

> Those in the new app camp will "use mpi3". Those in the old camp (99%?) 
> will include "mpif.h" or, if they please, "use mpi". And there will be 
> peace.

I doubt it!

Regards,
Nick Maclaren.