[MPIWG Fortran] Reading of #408: deprecating mpif.h/freezing mpi module

[Bill Long]

Hi Jeff,

Thanks for the update.  I am encouraged about the vote on mpif.h - that is good news.

The vote on the mpi module is less fortunate, but not unexpected.  Is there sentiment for a Plan B?  From a broad view, there are three areas of difference between the modules mpi and mpi_f08:

1)  INTEGER handles (MPI) versus derived type handles (mpi_f08).

2)  Choice buffer declarations that are TYPE(*),DIMENSION(*)  [aka void *]  (mpi) versus TYPE(*),DIMENSION(..) [aka C descriptor] and use of ASYNCHRONOUS attribute (mpi_f08).

3) No OPTIONAL arguments (mpi) versus allowing OPTIONAL arguments (mpi_f08).

I believe our current Plan B is to “upgrade” the mpi module by keeping (1) as is but migrating (2) and (3) to the mpi_f08 conventions.  

Advantages:

 a) Solves the problems that (2) is intended to solve 
 b) The convenience of (3). 
 c) No required change to the user sources. (Particularly, the handle declarations).
 d) Potentially better performance (by avoiding the possible copy-in/copy-out of choice buffers at call sites).

Disadvantages:

 a) Requires TS compiler support, because of (2).


If that is even too much for the voters, then a watered down Plan C could involve keeping the mpi conventions for (1) and (2), but at least allowing the OPTIONAL arguments (3).

Advantages:

 a) The convenience of (3)
 b) Does not require any new compiler technology. OPTIONAL arguments have been around for 20 years.
 c) No required change to user sources.


Disadvantages:

 a) No access to the feature improvements from (2)
 b) Reduces momentum towards removing the mpi module, requiring longer period of dual maintenance by vendors. 


Personally, I prefer Plan B, but if that does not have enough support, Plan C is better than nothing.  At least Plan C will reduce the number of times support people have to explain why the call to MPI_Init failed because the user forgot the error argument. 

Cheers,
Bill



On Mar 4, 2014, at 7:37 PM, Jeff Squyres (jsquyres) <jsquyres at cisco.com> wrote:

> Due to scheduling, I ended up reading #408 today (instead of tomorrow):
> 
> Please see the results:
> 
>    https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/408#comment:5
> 
> -- 
> Jeff Squyres
> jsquyres at cisco.com
> For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
> 
> _______________________________________________
> mpiwg-fortran mailing list
> mpiwg-fortran at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpiwg-fortran

Bill Long                                                                       longb at cray.com
Fortran Technical Suport  &                                  voice:  651-605-9024
Bioinformatics Software Development                     fax:  651-605-9142
Cray Inc./ Cray Plaza, Suite 210/ 380 Jackson St./ St. Paul, MN 55101