Zoltan: added MPI-only test (no Trilinos) to verify correct MPI behavior
Created by: kddevin
This PR adds a non-Trilinos test to verify that self-messages work in MPI.
Xyce users reported errors coming from Zoltan using some Ubuntu and FreeBSD platforms. The errors were reported by Zoltan's hypergraph construction code in PHG; they included not-found GIDs and incorrect message lengths in the directory.
The real problem was that the MPI installations on these systems were faulty; they did not handle self-messages correctly. The issue has been reported; see, for example, https://bugs.launchpad.net/ubuntu/+source/openmpi/+bug/1731938
The issue arose with OpenMPI2, but comments at the site above give little confidence that it is fixed in OpenMPI3.
This new test is a canary-in-a-coalmine. If it fails, no one should expect Zoltan to work.
Is there a better place in Trilinos for such tests, rather than in the packages? Do we need a "stableSystemPackage" that includes tests like these, which have no dependence on Trilinos yet flag conditions that will cause Trilinos to fail? For a similar MPI diagnosis test, see #3500.
@trilinos/zoltan
Tested with: mac os x with OpenMPI 1.10