On Mon, Apr 27, 2009 at 1:07 PM, John Peterson
> On Mon, Apr 27, 2009 at 12:43 PM, ZeljkaT <zeljkat@...> wrote:
>>>You should probably consult the PETSc documentation to see if -pc_type
>>>has disappeared in 3.0.
>> It hasn't disappeared - it works fine with libMesh-0.6.3.
>> With svn the preconditioner is not accepted - the default value (ilu) always appears.
> Looks like PCSetFromOptions was commented out of petsc_linear_solver.C
> (line 171) and labeled as unnecessary.
> // Not sure if this is necessary, or if it is already handled by
> // NOT NECESSARY!!!!
> //ierr = PCSetFromOptions (_pc);
> Are you using PETSc+LibMesh in a non-standard way? If so, we may need
> to take a look at the code (if it's simple enough) to get an idea of
> what's going on.
> Was PETSc definitely detected during your configure step? If yes, you
> could try uncommenting this part, recompiling, and rerunning, and see
> if it makes any difference...
Actually, I think Derek broke this one ;-) At the bottom of init()
there's a call to the new PetscPreconditioner class...
I think there needs to be one final call to PCSetFromOptions after any
and all other calls to set_preconditioner_type() to ensure that the
command line has the final say...