dolfin team mailing list archive

["Garth N. Wells" <gnw20@xxxxxxxxx>]

On Tue, 2010-08-17 at 19:51 +0200, Anders Logg wrote:
> On Tue, Aug 17, 2010 at 07:17:29PM +0200, Anders Logg wrote:
> > On Tue, Aug 17, 2010 at 07:06:43PM +0200, Anders Logg wrote:
> > > On Tue, Aug 17, 2010 at 06:01:46PM +0100, Garth N. Wells wrote:
> > > > On Tue, 2010-08-17 at 18:46 +0200, Anders Logg wrote:
> > > > > Why do we need to set PETSC_ARCH? We didn't before. Is there a way to
> > > > > figure it out?
> > > > >
> > > >
> > > > In general, no. It's possible (and common) to have a debug and an
> > > > optimised version installed side-by-side (there are more possibilities)
> > > > than just optimised and debug). The libs go in
> > > >
> > > >   ${PETSC_DIR}/${PETSC_ARCH}/lib
> > > >
> > > > If you do a regular
> > > >
> > > >    ./configure --prefix=/foo/bar
> > > >
> > > > then I don't think that it's required since there will be not
> > > > distinction between a debug and an optimised version.
> > >
> > > There seems to be some code for searching for PETSC_ARCH in the
> > > FindPETSc script:
> > >
> > > set (_petsc_arches
> > >     $ENV{PETSC_ARCH}                   # If set, use environment
> > >     variable first
> > >     ${DEBIAN_FLAVORS}  # Debian defaults
> > >     x86_64-unknown-linux-gnu i386-unknown-linux-gnu)
> > >
> > > But it doesn't include the following two options that we have in the
> > > old Scons script:
> > >
> > > elif os.path.exists(getPetscDir(sconsEnv=sconsEnv)+"/linux-gnu-c-opt"):
> > >         return "linux-gnu-c-opt"
> > > elif os.path.exists(getPetscDir(sconsEnv=sconsEnv)+"/lib/linux-gnu-c-opt"):
> > >         return "linux-gnu-c-opt"
> >
> > I see now these two are included in debian_arches.
> 
> I don't have a regular PETSC_DIR but rather an installed version of
> PETSc (installed by Dorsal). It looks like it should work anyway since
> then PETSC_ARCH can be empty and the following test will still find
> petsconf.h in $prefix/include:
> 

There is a lot to check - there were changes from PETSc 3.0 to PETSc
3.1.

Garth

> if (EXISTS ${PETSC_DIR}/${PETSC_ARCH}/include/petscconf.h)   # > 2.3.3
> 
> However, the PETSc test is still failing for me because of problems
> with a missing mpi.h:
> 
> /usr/bin/c++    -DBOOST_UBLAS_NDEBUG
> -DMULTIPASS_TEST_1_petsc_works_minimal
> -I/home/logg/scratch/lib/fenics-dev/include   -o
> CMakeFiles/cmTryCompileExec.dir/src.cxx.o -c
> /home/logg/scratch/src/dolfin-dev/build/CMakeFiles/CMakeTmp/src.cxx
> 
> In file included from /home/logg/scratch/src/dolfin-dev/build/CMakeFiles/CMakeTmp/src.cxx:3:
> /home/logg/scratch/lib/fenics-dev/include/petsc.h:139:17: error:
> mpi.h: Filen eller katalogen finns inte
> 
> My mpi.h is in /usr/include/openmpi (symlinked to
> /usr/lib/openmpi/include).
> 
> I don't see any settings for MPI include path in the PETSc conf
> script.
> 
> --
> Anders