Bug 1:
The changes to the MPI stubs on Jan 5th seem to have broken the python serial interface.
Compiled serial lammps-4Jan12 module. A small fix is required for lammps.py (uncomment `except` on line 44-45).
Using the python shell:
from lammps import *
lmp = lammps()
LAMMPS (4 Jan 2012)
Repeat for lammps-5Jan12 and you get: "OSError: Could not load LAMMPS dynamic library"
If I remove the try/except I see the error:
OSError: dlopen(_lammps_serial.so, 6): Symbol not found: _MPI_Abort
An error appears to persist to the most current version: lammps-17Feb12.
However if I replace the lammps-17Feb12/src/STUBS directory with the copy from lammps-4Jan12 and recompile, the python interface works correctly.
Possible Insignificant Bug 2:
I have also noticed that after lammps-1Dec11 a serially compiled version of lammps will print out:
1 by 1 by 1 MPI processor grid
Rather than:
1 by 1 by 1 processor grid
Axel, can you check if there is something missing in the changes
you made to STUBS/mpi.h and c ?
The latter issue is not a bug, the MPI was added
to distinguish between the full grid of MPI processors
and the "grid" of cores within a node. One proc
is still an MPI proc.