On Wed, Aug 29, 2012 at 5:20 AM, Jan-Philip Gehrcke <jgehrcke.googlemail.com
> wrote:
> Dear Amber developers, mainly Jason,
>
> the error message printed by MMPBSA.py.MPI in case of an ImportError of
> mpi4py is improvable. Consider:
>
> $ MMPBSA.py.MPI
> Fatal Error: I could not find the Python MPI package (mpi4py)! This
> means that
> it could not be built during the standard AmberTools
> installation.
> Check $AMBERHOME/src/AmberTools/mpi4py_install.log for
> errors.
>
> The actual error in this case was not that mpi4py was not built, but
> there was a binary incompatibility:
>
> from mpi4py import MPI
> ImportError:
>
> /apps11/bioinfp/amber12_centos58_intel1213_openmpi16/amber12/bin/mpi4py/MPI.so:
> undefined symbol: PyExc_BaseException
>
This can be clarified easily enough. I admit I hadn't considered causes of
ImportErrors *besides* a problem in the mpi4py build process (and there are
numerous platforms it fails to build on).
The original error message was quite misleading. Hence, I think we
> should not swallow the actual error message part of the ImportError
> exception. Btw, in the path in "Check
> $AMBERHOME/src/AmberTools/mpi4py_install.log for errors." 'src' and
> 'AmberTools' are mixed up.
>
True. I will fix the typo.
Thanks!
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 29 2012 - 07:30:05 PDT
