Re: Annoyting type_size problem

Steve Huss-Lederman (lederman@cs.wisc.edu)
Thu, 19 Dec 1996 07:50:13 -0600 (CST)

This issue has been raised many times. It is explicitly addressed on
p. 25 (Section 3.2.5) of the current MPI-1.2 document. The bottom
line is that the Forum has decided not to change from int even though
MPI_Aint make more sense to some. Breaking current programs was one
of the main reasons.

Steve

> From: John M May <johnmay@coral.llnl.gov>
> Date: Wed, 18 Dec 1996 17:05:20 -0800 (PST)
> X-Mailer: ELM [version 2.4 PL25]
> MIME-Version: 1.0
> Content-Type: text/plain; charset=US-ASCII
> Content-Transfer-Encoding: 7bit
> Sender: owner-mpi-io@mcs.anl.gov
> Precedence: bulk
>
> The MPI-1 standard says that MPI_Type_size returns the size
> of a type in an int parameter:
>
> int MPI_Type_size( MPI_Datatype type, int * size );
>
> On the other hand, MPI_Type_extent uses an MPI_Aint. Others
> have already noticed this inconsistency; for example early
> versions of MPICH used an MPI_Aint in MPI_Type_size,
> while later ones follow the letter of the standard here.
>
> Now in the I/O proposal, there are two analogous functions,
> MPI_File_get_type_size and MPI_File_get_type_extent. Both
> return their results in an MPI_Aint.
>
> My question is, will the MPI_Type_size function be changed
> to use an MPI_Aint, or should the MPI_File_get_type_size
> function be changed to use an int? Or will the inconsistency
> remain so that MPI-1 can remain as it while the new functions
> "get it right"?
>
> John
>