[MPIWG Fortran] ticket 351

Jeff Hammond jeff.science at gmail.com
Tue Oct 22 12:48:11 CDT 2013

>>> And, lastly, MPI 3.0 already defines a way of getting this that is
>>> the same in both Fortran and C - using MPI_INFO_ENV.
>> If only that were true.
> Point taken.  I think that's a ghastly specification, but I can see why
> you want it.
> But I still don't think that adding yet another hack on top of all of
> the existing ones is going to help.  My remarks that the current mechanism
> isn't reliable for C weren't theoretical, incidentally, though how many
> of the implementations that caused trouble are still extant and doing so
> I can't say.
> Realistically, the simplest solution is to say that implementations
> should take steps not to pollute the programmer's argument list, which
> would force those without mpiexec equivalents to use environment variables
> or equivalent.  I don't think that would fly because of opposition, but
> it's the only simple one I can see that would actually clean up the
> current specification.

I would not oppose an attempt to specify that MPI cannot mess with
argc+argv but I'm not sure that will pass.  We'd need input from a
historian like Bill Gropp to determine the odds.

It would seem that the WG is far from universally approving of the
ticket and is unlikely to reach a consensus (other than a negative
one) any time soon.

Craig, JeffS, Rolf: Do you have any thoughts?  If you don't have any
counters to Bill and Nick, I think we should withdraw the ticket.


Jeff Hammond
jeff.science at gmail.com

More information about the mpiwg-fortran mailing list