Re: [mpi-21] ABI - call for working group

[Greg Lindahl <lindahl@xxxxxxx>]

On Sun, Jan 13, 2008 at 03:02:32PM -0800, Greg Bronevetsky wrote:

> >It means the combination was never tested. Commercial vendors of
> >MPI-using programs generally don't support something they haven't
> >tested.

> If we're talking about ABI-related issues then it would be the morph 
> layer's responsibility to test their interactions with any particular 
> MPI. The software vendor would need to test functionality-related 
> stuff but that problem remains regardless of what we do with the ABI.

We aren't talking about ABI-related bugs or functionality issues,
we're talking about MPI implementation bugs combined with
non-standard-conforming programs ("my program works with some other
MPIs, so it must be correct!"). And yes, an ABI doesn't make either
issue go away, but it can make testing with many MPIs much easier.

BTW, in the MorphMPI scenario, who is going to support MorphMPI? MPI
vendors are going to revolt at the thought of being on the hook for
bugs caused by software layers they don't control. And ISVs aren't
going to like that situation, either. I think the most likely outcome
is an even more complicated compatibility matrix, where MPI vendors
and ISVs will require particular versions of the MorphMPI. Yuck.

-- greg