Hi,

I've installed petsc-dev with *--download-hdf5* (along with other
*--download-xxx* to have petsc downloading and installing other packages
like scotch, parmetis, etc.)

I've downloaded fenics-dev and set it to use scotch, pastix parmetis,
umfpack, cholmod, amd, and hdf5 from the ones installed by pestc.

I've done it because my Fedora 20 already has two installations for hdf5:
one serial and another paralllel, and I've prefer to have petsc
installation for hdf5.

As fenics/dolfin scripts use *find_package(HDF5)*, I changed my Fedora's
*FindHDF5.cmake* in order to have the correct petsc's hdf5, and put it
directly into *dolfin/cmake/modules/* to be seen by dolfin cmake scripts.
This modification (basically forcing *FindHDF5.cmake* to set *HDF5_**
variables with values for petsc hdf5 installation). I check it by printing
these variables directly in *dolfin/CMakeList.txt* (put *message* command
after find_package(HDF5)).

Well, everything was good EXCEPT that along with the right hdf libraries (
*libhdf5.so* and *libhdf5_hl.so*), fenics installation script (cmake file)
brought together the Fedora installation version of these libraries.

An ldd in *libdolfin.so* reveals
*/opt/petsc/lib/libhdf5.so*
*...*
*/usr/lib64/libhdf5.so*

My question is: in which part of the fenics' cmake scripts you consolidate
the libraries (or **_EXTERNAL_OBJECTS* in some build.make files, for
instance)?

One way I adopted to circumvent it was to create a script to change all
*/usr/lib64/libhdf5* by */opt/petsc/lib/libhdf5* BEFORE call *make* and *make
install*.

Well, if you need some log file, feel free to ask me. I should emphasize
that the fenics and petsc are working just fine but I think there is a more
fine solution for this inconsistencies with hdf5 libraries.

Regards,
Marcelo Rosa.
_______________________________________________
fenics-support mailing list
[email protected]
http://fenicsproject.org/mailman/listinfo/fenics-support

Reply via email to