Newsgroups: comp.parallel.mpi
From: lederman@reggiano.cs.wisc.edu (Steve Huss-Lederman)
Subject: Re: MPI_PROC_NULL and invalid message tags
Organization: CS Department, University of Wisconsin
Date: 05 Mar 1996 16:43:15 GMT
Message-ID: <LEDERMAN.96Mar5104315@reggiano.cs.wisc.edu>

In article <4hb1un$9t3@murrow.corp.sgi.com> salo@mrjones.engr.sgi.com (Eric Salo) writes:

 > 
 > What about an invalid pointer for a zero-length message?

I ran into that problem.  A code had NULL pointers when there was no
data in an unusual case.  The Bcast did not generate an error.  I have
my own Bcast routine and it did.  The person who asked me to look at
the code could not figure out what was wrong until we used my Bcast
routine.  They wanted MPI to generate the error in this case.  Others
have told me they don't.  I think there is not a clear choice on
this.

My recollection of the discussion during the MPI Forum is that we
never addressed this issue directly.  My reading of the standard would
be that you can choose to generate an error or not.  If people think
this is a bad choice, we could discuss it as an MPI-1 clarification in
the current MPI Forum discussions.  Do people think we should clarify
to allow only one type of behavior?

Steve

