@pdiercks , Do you have any good ideas on how we can test the MPI functionality?
Ideally we should test if the solution fields and quadrature fields are identical for some Problem that has been solved with a different number of MPI processes.

Can we try to write a pytest case where we solve a distributed Problem with MPI.COMM_WORLD and on one of the ranks the same problem with MPI.COMM_SELF? That way we could produce an output within one test case without writing any results to the disk. But I don't know if we can mix the comms. For example would the following work?:

mesh_distr = df.mesh.create_unit_cube(MPI.COMM_WORLD,10,10,10)
mesh_local = df.mesh.create_unit_cube(MPI.COMM_SELF,10,10,10)

space_distr = df.fem.functionspace(mesh_distr, ("Lagrange",1))
space_local = df.fem.functionspace(mesh_local, ("Lagrange",1))

u_distr = df.fem.Function(space_distr)
u_local = df.fem.Function(space_local)

diff = u_distr - u_local

I tried it. Does not work with different communicators. I will try https://gitlab.com/dglaeser/fieldcompare

@pdiercks , Do you have any good ideas on how we can test the MPI functionality? Ideally we should test if the solution fields and quadrature fields are identical for some Problem that has been solved with a different number of MPI processes.

Can we try to write a pytest case where we solve a distributed Problem with MPI.COMM_WORLD and on one of the ranks the same problem with MPI.COMM_SELF? That way we could produce an output within one test case without writing any results to the disk. But I don't know if we can mix the comms. For example would the following work?:

mesh_distr = df.mesh.create_unit_cube(MPI.COMM_WORLD,10,10,10)
mesh_local = df.mesh.create_unit_cube(MPI.COMM_SELF,10,10,10)

space_distr = df.fem.functionspace(mesh_distr, ("Lagrange",1))
space_local = df.fem.functionspace(mesh_local, ("Lagrange",1))

u_distr = df.fem.Function(space_distr)
u_local = df.fem.Function(space_local)

diff = u_distr - u_local