Wolfgang Bangerth

What should we do here in view of the release coming? Can this be fixed within the next week?

As a point of reference, I removed the default arguments in `grid_tools.h` and compiled the library with no problems. In other words, all places in the library itself that call...

It's places like these in `GridTools`: ``` template double volume(const Triangulation &tria, const Mapping & mapping = (ReferenceCells::get_hypercube() .template get_default_linear_mapping())); ``` The default value for the last argument will only...

@blaisb: Can you try what happens when you replace ``` columns[key].entries.push_back(internal::TableEntry(value)); ``` by ``` columns[key].entries.emplace_back(internal::TableEntry(value)); ``` ?

Yes please.

Thanks for your input already. We'll see that we move these variable declarations out of `vmult()` so that they have to be set up only once. I do still think...

For reference, we currently construct these range and domain maps in the following way: ``` template TrilinosPayload::TrilinosPayload( EpetraOpType & op, const bool supports_inverse_operations, const bool use_transpose, const MPI_Comm &mpi_communicator, const...

@juleoc02 has dug a bit deeper, and it turns out that this code takes 4.2 seconds: ``` { TimerOutput::Scope t(timer, "solve with inverse op"); SolverControl solver_control(100, 1e-6); SolverGMRES solver (solver_control);...

Yes, we should know better than to do an explicit inversion. Can you point me at the place where that happens?

That second hunk is in `FE_PolyTensor::shape_value_component()`, which means that we do a matrix-vector product there anytime we ask for shape values. That's kind of expensive, but I guess that happens...