[Dune] [Dune-Commit] dune-common r6227 - trunk/dune/common

[Christian Engwer]

On Thu, Nov 04, 2010 at 07:10:21PM +0100, Martin Nolte wrote:
> Hi Markus,
> 
> thanks for the hint (reading an MPI standard you never see when it
> was introduced). I actually only checked with OpenMPI.
> 
> The reason for introducing the was that I had third party software
> that brought a dummy implementation of MPI_Init. It took me about 2
> hours to find the problem. Now, the second assertion would fail. The
> reason for the other assertion is exactly opposite: Make sure that
> third party software does not initialize MPI before we do.
> 
> How should we deal with the situation? Simply reverting it will
> obviously solve the insecurity, loosing the debugging hours, again.
> Another option would be a configure check for MPI 2.0 (or
> specifically for MPI_Initialized). Since a few interesting features
> have been introduced in MPI 2.0, we might want to be aware of them
> anyway. The last possibility I can think of is waiting until someone
> complains and keeping this in mind.

I suggest adding an additional mpi2.0 check. Waiting for other codes
to break, if we already know that it might create problems, isn't a
good solution ;-)

Christian