Re: [Libmesh-users] Libmesh0.4.2 compiling problem in Solorias

[Benjamin S. K. <be...@cf...>]

I have found a Solaris 2.9 machine (thanks Sourceforge!) that should be 
similar to yours.  I was able to get the default ./configure to work... 
  You will need to change two lines in the Make.common file after 
running configure:

change

RPATHFLAG            = -Wl,-rpath,

to

RPATHFLAG            = -Wl,-R,

and

LIBS =

to

LIBS = -lrpcsvc

This is due to differences in the Linux and Sun linkers.   I have 
checked these changes in to the CVS repository.  If you follow the 
anonymous CVS access instructions 
(http://libmesh.sourceforge.net/installation.php) and download what is 
currently in the CVS repository you should be able to just ./configure 
and go.

It does seem to take a long time to link the libraries, but all the 
tests seem to run fine.


                                                                                

A few notes on the output from configure that you sent me:

--------------

- configure is a GNU tool, and as such looks for GNU compilers first.  
In your case it found gcc-3.3 and is using it to build the library 
instead of the Sun Forte compiler.  This shouldn't be a problem (indeed, 
I have never built libMesh with the Forte compiler, so this might be 
good).  However, problems *may* arise if PETSc is built with the Forte 
compiler and libMesh is not.  You should be able to figure out which 
compiler PETSc was built with by looking in the file 
$PETSC_DIR/bmake/$PETSC_ARCH/variables

To specify which compiler you want to use instead of letting configure 
detect it automatically you have to use environment variables.  For 
instance, under bash or sh this should work:

CXX=CC CC=cc F77=f77 ./configure ....

(For other shells you might need to export the environment variables 
before running configure).

After you have configured the library you don't need to worry about the 
compiler, the right one will be specified in the Make.common file.

--------------

- As for the MPI stuff:  libMesh will use MPI if it is available, 
although it is not required.  If PETSc is found then MPI is definintely 
available, and the file $PETSC_DIR/bmake/$PETSC_ARCH/packages is queried 
to find it.  This could pose a problem if your PETSc has been provided 
as binaries and not built from source.  In this case I would be 
interested in hearing about it and improving the configuration.

If PETSc is not there, then a number of other tests are conducted to 
find a valid MPI implementation.  First mpi libraries and header files 
are looked for in /usr/lib & /usr/include (you can specify a directory 
with --with-mpi=/path/to/mpi).  If nothing is found there then we do a 
final test to see if the compiler supports MPI directly.  This is the 
case with the AIX compilers or the mpiCC provided by MPICH, for example.

There was a bug in the Parmetis configuration that tried to build 
parmetis even if no valid MPI installation was found.  I have fixed that 
in the CVS branch.