[petsc-users] openmpi/petsc 3.1/gfortran and multiply-defined symbols

[Satish Balay]

a couple of issues:

-  Looks like you've installed petsc at --prefix=/usr/local/petsc-3.1-p8-dbg - but
using from -I/usr/local/petsc-3.1-p8-opt

- Should avoid pointing to std locations as configure will always look there.
   --with-blas-lapack-dir=/usr/lib --with-mpi-dir=/usr 

- when reinstalling - make sure the prefix location does not have any old install files.

And to track the duplicate location of MPI_FILE_NULL - suggest
compiling the code with '-E' preprocessor directive - and then locate
MPI_FILE_NULL there..

i.e

mpif90 -c -g -ffree-line-length-none -fcray-pointer -DMOAB -fPIC -Wall
-Wno-unused-variable -O -I/usr/local/petsc-3.1-p8-opt/include
-I/usr/local/petsc-3.1-p8-opt/include -I/usr/include/scotch
-I/usr/lib/openmpi/include -I/usr/lib/openmpi/lib -I/usr/include/spooles
-I/usr/include/suitesparse -I/usr/local/petsc-3.1-p8-opt/include
-I/usr/include/scotch -I/usr/lib/openmpi/include -I/usr/lib/openmpi/lib
-I/usr/include/spooles -I/usr/include/suitesparse   -I.
-I./MeshLibrary/RTelements -I/usr/local/hdf5-1.8.8-par-openmpi-gcc/include
-I/home/tautges/code/MOABpar/include MeshLibrary/PNTmesh_Import.F90 -E > out.f

Satish

On Wed, 21 Mar 2012, Tim Tautges wrote:

> Hi all,
>   I'm almost embarrassed to be asking this, but haven't found anything that
> looks similar in the archives...
> 
> I'm compiling fortran code and get:
> 
> mpif90 -c -g -ffree-line-length-none -fcray-pointer -DMOAB -fPIC -Wall
> -Wno-unused-variable -O -I/usr/local/petsc-3.1-p8-opt/include
> -I/usr/local/petsc-3.1-p8-opt/include -I/usr/include/scotch
> -I/usr/lib/openmpi/include -I/usr/lib/openmpi/lib -I/usr/include/spooles
> -I/usr/include/suitesparse -I/usr/local/petsc-3.1-p8-opt/include
> -I/usr/include/scotch -I/usr/lib/openmpi/include -I/usr/lib/openmpi/lib
> -I/usr/include/spooles -I/usr/include/suitesparse   -I.
> -I./MeshLibrary/RTelements -I/usr/local/hdf5-1.8.8-par-openmpi-gcc/include
> -I/home/tautges/code/MOABpar/include   -o MeshLibrary/PNTmesh_Import.o
> MeshLibrary/PNTmesh_Import.F90
> mpif-mpi-io.h:64.27:
>     Included at mpif-config.h:65:
>     Included at mpif-common.h:70:
>     Included at /usr/lib/openmpi/include/mpif.h:59:
>     Included at /usr/local/petsc-3.1-p8-opt/include/finclude/petscsys.h:11:
>     Included at /usr/local/petsc-3.1-p8-opt/include/finclude/petsc.h:6:
>     Included at MeshLibrary/PNTmesh_Import.F90:16:
> 
>       integer MPI_FILE_NULL
>                            1
> Error: Symbol 'mpi_file_null' at (1) already has basic type of INTEGER
> 
> 
> My petsc configure options (as reported in arch/configure.log) are:
> 
> Configure Options: --configModules=PETSc.Configure
> --optionsModule=PETSc.compilerOptions --with-clanguage=C++
> --with-blas-lapack-dir=/usr/lib --with-mpi-dir=/usr --with-debugging=1
> --prefix=/usr/local/petsc-3.1-p8-dbg
> Working directory: /home/tautges/linux/petsc-3.1-p8
> 
> This is an Ubuntu 10.04 system, synaptic-installed openmpi, gfortran 4.4.3,
> all pretty standard stuff.  I think the problem started showing up when I
> inserted the "--with-clanguage=C++" in the configure line, expecting that
> would take care of finding the C++ std libs for me (which it did).  Clues?
> 
> Thanks.
> 
> - tim
> 
>