On Jan 23, 2009, at 1:20 PM, N.M. Maclaren wrote:
FWIW, ABI is not necessarily a bad thing; it has its benefits and
drawbacks (and enablers and limitations). Some people want it and
some people don't (most don't care, I think). We'll see where
that effort goes in the Forum and elsewhere.
Right. But, as someone with experience of trying to design portable
ABIs,
it requires more knowledge and skill than the typical person
tackling the
job knows even exists ....
Indeed. This is at least one of the reasons for the current deadlock
in the ABI discussions on the Forum (of which I am a part).
MPI did the Right Thing back in the mid-90's by just designing source-
level compatibility. Whether it's the right time to move to ABI or
not is a very politically- and religiously-charged discussion. :-)
FWIW, the F03 bindings for MPI may allow address-sized integers to
be handles in Fortran. In this case, MPI handles will likely take
on exactly the same value that they are in C. In OMPI's case,
that's a C pointer, so the F03 value for MPI_COMM_WORLD will be
some very large non-zero integer value. (standard disclaimers
about future features/ functionality -- time will tell if this
stuff plays out as expected)
That would solve this particular problem, if it is what I think it is.
Good.
A private Email made me realise that he was probably passing the
Fortran
MPI_COMM_WORLD to NetCDF4 for use as a communicator - and, unless
NetCDF
is much better quality than when I last looked at it, I will bet that
its Fortran interface is just a thin wrapper. You can guess the
rest :-)
FWIW, it probably works with MPICH and friends because they use
integer handles in both Fortran and C, and therefore the values are
exactly the same. Specifically, if NetCDF is just passing the value
of Fortran MPI_COMM_WORLD back to a C MPI API function, it'll likely
work in MPICH. But it won't in Open MPI because our handles are
different between Fortran and C. The Right solution is to use the
various MPI_*_f2c and MPI_*_c2f conversion routines (these are in the
MPI spec -- we didn't make them up for OMPI). See
http://www.mpi-forum.org/docs/mpi21-report-bw/node355.htm#Node355
I don't know anything about NetCDF4, so I don't know if it's
neglecting to do that or not.
...but it sounds probable. :-)
--
Jeff Squyres
Cisco Systems
