this is a problem due to the transition from EntityPointer to Entity in dune-grid

this is a problem due to the transition from EntityPointer to Entity in dune-grid

The method postAssembly was apparently never used, as it had
a wrong signature.

Those EntityPointers are Entities now, and don't have to be dereferenced.
This fixes a few warnings.

The test optionally takes a command line parameter that defines the
chunk size. If none is given, it defaults to 5.

This patch adds a new decorator ordering::Chunked that makes it possible
to introduce a "stupid" blocking scheme at any given function space
node. It must be given a chunk size and simply requires the block count
of the underlying ordering node to be a multiple of that chunk size.

This patch adds a special hook
deactivate_standard_blocking_for_ordering() that can be used to turn off
blocking. It gets passed an OrderingTag and can return false to turn off
blocking if it is not desired.

This functionality is needed for the new ChunkedBlockOrdering decorator.

transforming pointers

Nobody ever seemed to have used boundary pattern assembly.
The assemble method was given caches instead of actual LFSs.

There was a spurios typedef there...

BlockMatrixDiagonal is a helper class for the parallel nonoverlapping
Jacobi implementation that provides optimized storage for the diagonal
matrix blocks, as those need to be communicated to create a consistent
preconditioner.

In order to create a generic infrastructure for setting up this data
structure, the setup procedure doesn't directly use the native iterators
of the block rows to copy the per-DOF data, but replaces them with
simple pointers to the dense, row-major data.

Unfortunately, the construction of the end iterators in this scheme
simply dereferenced the original end() iterator and then took the
address of that (non-existing) element. This is not legal C++, however,
and rightfully caused breakage when compiling with assertions.

This patch works around the problem by instead decreasing the end()
iterator by one, dereferencing that (valid) iterator, and incrementing
the address of the resulting element.

Kudos to Nagaiah Chamakuri for reporting the problem and to Dominic
Kempf for debugging it!

This patch fixes a problem related to FS#1549.

The test is unrelated to ALUGrid.

With the move to dune-alugrid, <dune/grid/alugrid.hh> needs to become
<dune/alugrid/grid.hh>. Note that with this change, the test doesn't
work with the old ALUGrid anymore.

We now require a sufficiently recent compiler to be able to always rely
on std::initializer_list, so drop the preprocessor checks for that
feature.

This just follows the upstream change in dune-grid.

The header for the finite element has been renamed to monomial.hh and
the class to MonomialLocalFiniteElement.

Make sure we compile again.

dune-multidomain needs direct access to the undecorated base ordering in
a stack of decorated orderings for its custom DataHandleProvider.

This patch adds a little function undecorated_ordering() for that purpose.

SGrid is deprecated and not used for anything anyway, so shut up the
deprecation warning.

This patch fixes the creation of nested patterns (for nested BCRS
matrices).

When using nested BCRS matrices in multidomain to implemented coupling
schemes like Neumann-Dirichlet, applying Dirichlet constraints to the
matrix fails if the associated diagonal block is missing (because it
belongs to the subproblem we are not interested in right now). In that
case, attempting to write a 1 on the diagonal raises an exception.

This patch fixes the problem by adding a new helper method
istl::write_matrix_element_if_exists(), which makes sure that the block
exists before traversing into it. This new method is used by the matrix
wrapper when setting Dirichlet constraints.

When repeatedly solving the same problem for different right hand sides,
StationaryLinearProblemSolver shouldn't reassemble the matrix each
time.

This patch adds a flag to the class that makes it possible to switch to
this behavior.