Hi parallel processers, I noticed there's one additional peculiarity about using MPI: Since it's a different environment, the mpi block apparently doesn't get any information about the currently active dataset. I have worked around that problem in my application by putting in a 'nulldata ... --preserve' at the mpi toplevel (after reading in the bundle with a contained list). This seems to work, but here are still some questions or suggestions. - We have a --send-functions option to mpi; wouldn't it make sense to also have something like --send-data, which would resemble what we already do for 'foreign'. - Slightly more general than purely MPI-related: I noticed in the gretl+MPI guide section 12 (ML estimation, illustration) it is explained that some dummy 'nulldata' commands must be given just because the 'mle' command formally requires it - although as explained there as well it really isn't using any dataset in that case. This feels very awkward and/or clumsy. Shouldn't the dataset requirement be restricted to those commands which really need it? thanks sven