Is this MUMPS issue resolved?
We cannot assume the MPI world is binary and have only MPICH or OPENMPI.
This specific commit 25aac85c is wrong; it should test for MPICH + specific version to prevent from using the distributed rhs feature. In general, such hardcoded disabling of features should target the specific version and distribution of the package.
Further, IMHO we should carefully think if hardcoding such "fixes" is really beneficial: we can easily forget we patched the library and prevent others to benefit from features. In addition, according to https://bitbucket.org/mpi4py/mpi4py/issues/162/mpi4py-initialization-breaks-fortran, this could have been fixed from user side. C code should not be a place to provide patches for python.
cc @dalcinl @jedbrown @BarrySmith @sbalay @knepley for their opinions