On Wed, 13 Jan 2010, Kevin.Buckley at ecs.vuw.ac.nz wrote: > Hi there, > > I am tying to help a researcher here build his application against > a PETSc (3.0.0 p9) installtion at another site that seems to have > undefined references to: > > PETSC_VIEWER_SOCKET_
Is this unresolved symbol from user code? complete compile commands/ error messages are useful. > > in it. > > Looking at my local installtion, the undefined refs are seemingly > fixed up in libpetsc.a > > ... > send.o: > U MPI_Attr_get > U MPI_Attr_put > U MPI_Comm_rank > U MPI_Keyval_create > U OMPI_C_MPI_NULL_COPY_FN > U OMPI_C_MPI_NULL_DELETE_FN > 00001a4a T PETSC_VIEWER_SOCKET_ > ... > > however on the implementation that the researcher is looking to build > against, that "T PETSC_VIEWER_SOCKET_" is not there. > > > I noticed in the example config script for the Cray that there's a > notification that: > > ... > # After running configure, remove the following flags from petscconf.h > # > # PETSC_HAVE_SYS_PROCFS_H > # PETSC_HAVE_DLFCN_H > ... > ###### On Cray XT4 the following additional flags need removal ######## > # > # PETSC_USE_SOCKET_VIEWER > # PETSC_HAVE_GETPWUID > ... > > and think that this might be the problem here. > > Am I right in thinking that PETSC_USE_SOCKET_VIEWER, the macro > that determines whether PETSc compiles in any socket viewer code > is not something that gets turned /onoff automatically by any > option to configure ? There is a configure test for this - but cray supplies dummy socket stubs - so configure succeeds - but the functionality doesn't exist. [or it could be our configure test isn't as robust as it should be]. The result is - some manual tweaking might be needed on cray machines. But I'm not sure if this problem still persists on the newer software stacks on these machines. I don't remember having to do these edits on the NERSC XT4. Satish > > Kevin > >
