From: Richard Treumann <treumann@XXXXXXXXXX>
To: <mpi-core@XXXXXXXXXXX>,<mpi-comments@XXXXXXXXXXXXX>
Subject: MPI_ALLOC_MEM & no mem
Date: Fri, 31 Jul 1998 14:42:46 -0400
MIME-Version: 1.0
Content-Type: text/plain
Sender: owner-mpi-core@XXXXXXXXXXX
Precedence: bulk
X-UIDL: 48fa7071bfdf9c1798782e0129848b12

The standard says that MPI_ALLOC_MEM **may** return an error code if the
requested mem is not available.  If it returns an error and the user has left
MPI_ERRORS_ARE_FATAL in effect, the job gets killed.

Does anyone think this is too drastic?  Should there be a way for a user to
TRY_ALLOC?  After all, malloc is designed to allow failure and users may wish
to try for a big hunk of mem but settle for a smaller one if they need to.
Using MPI_ERRORS_RETURN will probably work but cannot be relied upon because
the standard still says that the internal state of MPI is no longer guarenteed
after an error.

                 Dick

Dick Treumann                IBM RS/6000 SP Development
Tele (914) 433-7846         Fax (914) 433-8363