TECA prevents mpi4py from doing its default MPI_Init/MPI_Finalize during import/exit in order that TECA can be used on NERSC Cray's on the login nodes and more recently without running via srun on the compute nodes. We need to document that either
parallel codes import mpi4py before teca
explicitly call MPI_Init_threads/MPI_Finalize when importing mpi4py after teca.
TECA prevents mpi4py from doing its default MPI_Init/MPI_Finalize during import/exit in order that TECA can be used on NERSC Cray's on the login nodes and more recently without running via srun on the compute nodes. We need to document that either
MPI_Init_threads
/MPI_Finalize
when importing mpi4py after teca.