Sébastien Villemot <sebastien.villemot@ens.fr> writes: > Using the MPI version of HDF5 with liboctave-dev was not working (see > #598227). Then the HDF5 maintainers encouraged us to enforce the serial > version. I have just tested the solution that Thomas had implemented for Squeeze, which was: - make liboctave-dev depend on libhdf5-dev | libhdf5-openmpi-dev | libhdf5-mpich2-dev - add -I/usr/include/mpi to XTRA_CXXFLAGS in mkoctfile This change should not introduce new bugs. Then I have been able to compile octave-optim (which compiles .oct files) with libhdf5-openmpi-dev installed, and also with libhdf5-mpich2-dev. So it looks that this simple change solves the co-instability issue of HDF5/MPI with liboctave-dev. As an additional benefit, that would allow us to reintroduce back octave-bim and octave-msh, and to upgrade to the latest version of octave-secs1d. So my impression is that we should implement it, given its clear benefits, and despite the fact that explicitley adding a -I flag is a bit ugly. Thomas, Rafael: what do you think? Am I missing something? Best, -- .''`. Sébastien Villemot : :' : Debian Maintainer `. `' http://www.dynare.org/sebastien `- GPG Key: 4096R/381A7594
Attachment:
pgpIe5ICc0pzl.pgp
Description: PGP signature