X-Mailer: QUALCOMM Windows Eudora Pro Version 4.2.2 
Date: Mon, 09 Jul 2001 08:44:37 -0500
To: mpi-21@XXXXXXXXXXXXX
From: William Gropp <gropp@XXXXXXXXXXX>
Subject: Re: redefining MPI_LONG_LONG_INT
Cc: mpi-21@XXXXXXXXXXXXX,hans-christian.hoppe@XXXXXXXXXX,In-Reply-To: <200107090914.LAA00843@XXXXXXXXXX>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed
Sender: owner-mpi-21@XXXXXXXXXXXXX
Precedence: bulk
Reply-To: mpi-21@XXXXXXXXXXXXX

At 11:14 AM 7/9/2001 +0200, Alexander Supalov wrote:
Hi everybody,

It's just stricken my eye that the MPI_LONG_LONG_INT as voted in last time
conflicts with the usual meaning of the MPI_type_INT constants. Those are
normally used to indicate a <type,MPI_INT> pair suitable for the MPI_MINLOC and
MPI_MAXLOC reduction operations.

Note that we didn't vote *in* MPI_LONG_LONG_INT; it is part of the MPI 1.0/1.1/1.2 standard (see page 19, line 1 of the 1.1 standard).  The vote was only to correct the definition of the length of MPI_LONG_LONG_INT to use the datatype that was defined as part of the MPI Standard.

From this point of view, the now affirmed equivalence of the MPI_LONG_LONG_INT
and MPI_LONG_LONG (item 16 of the 1st bunch) looks fairly unfortunate, doesn't
it?

I agree that the name chosen in MPI 1 was unfortunate, but rather than change part of MPI 1.1, we'll need to define a different name for the MAXLOC and MINLOC datatypes.

Bill