Roy Stogner <> wrote on 09/23/2011 01:53:58 PM:

> On Tue, 20 Sep 2011, Roy Stogner wrote:
> > There's a bug here: you forgot to change PetscNonlinearSolver to
> > NoxNonlinearSolver in a couple START_LOG lines.
> One more serious bug, too: there's a use of libmesh_cast_ref which
> breaks for me in optimized modes.  Apparently (at least in Trilinos
> 9.0.3) the Epetra_Operator is a virtual base class of
> Epetra_FECrsMatrix, so we're not allowed to optimize that into a
> static_cast.  Replacing libmesh_cast_ref with dynamic_cast fixes
> compilation for me.

Fixed both issues.
Also found another one: returning total number of _lin._ iterations instead of _nonlinear_ ones.
While doing that, I implemented get_total_linear_iterations for Nox solver as mentioned in another thread here.

All is checked in.