@davydden: is this ready to merge?

@tgamblin should be good now.

This breaks petsc. The test is a mpi code run in sequential on our site we have mvapich2+slurm if you do not run with srun it fails.
And no wrapper mpirun is installed when using +slurm so just adding mpirun will not solve it.

@nrichart i will check the example file and if needed switch to another one.

However, the fact that you was not able to run it with MPI in my opinion indicates a bigger issue, not related to petsc per-se.

No don't get me wrong I am able to run it.
srun ./ex50 <parameters> works fine, but it mean you cannot hardcode mpirun in the package.py of petsc or other packages.
Since depending on the variant of mpi the mpirun wrapper might not be installed

@nrichart how about we introduce another variable to mpi wrappers spec['mpi'].run which for OpenMPI would be mpirun and for mvapich2+slurm -- srun, whereas without slurm it should probably also be mpirun. I am not sure about the parameter for the number of processes, i think this is also not the same among MPI implementations -np, -n, etc.

I would then switch these tests to mpi-only and use spec['mpi'].run to run them.

@nrichart can you check if the following test works with your installation

        if 'mpi' in self.spec:
            with working_dir('src/ksp/ksp/examples/tutorials'):
                env['PETSC_DIR'] = self.prefix
                make("ex50", parallel=False)
                make("runex50", parallel=False)
                make("ex52", parallel=False)
                make("runex52", parallel=False)
                make("runex52_mumps", parallel=False)
                make("runex52_superlu_dist", parallel=False)
                make("ex55", parallel=False)
                make("runex55_hypre", parallel=False)

It uses PETSc's make file and targets to run. Those are supposed to pick up mpiexec and alike.