[bugfix] avoid subscript operator on JacobianTransposed

See merge request !154

A geometry's JacobianTransposed implementation may be any type modelling a
linear mapping, i.e., implementing the mv-type methods. Some grids
(e.g., SPGrid) only implement this interface for performance reasons.
However, we require that a JacobianTransposed can be assigned to a
FieldMatrix.

This patch fixes "checkCodim1Mapping" such that it will cast the
JacobianTranposed down to a FieldMatrix before accessing the subscript
operator (although the same result could be achieved using "mv").

UGGrid: append to-be-visited sons at the end of the list

Closes #53

See merge request !150

Handle unused boundary nodes correctly

Closes #55

See merge request !152

For "unused" boundary nodes the special value `-1` is used.  Mapping
this through `isBoundaryNode[-1]` would result in an out-of-bounds
memory access and invalid values in the `vertices_c_style` array.

Closes: #55

Based on the test case provided by Lasse Hinrichsen.

Prepending them to the beginning of the list would mean they are never
visited as the iterator only moves forward.  This required switching to
a `std::list` as the `std::forward_list` does not provide methods to add
items at the end.

This regression was introduced in 20b821be
which replaced `Dune::SLList` with `std::forward_list`, but changed
`push_back` to `push_front` as well.

Closes: #53

If the code path should never be reached, having an exception thrown
makes it easier to nitce it was reached after all.

Pass `OBJECT` option to `add_dune_mpi_flags`

See merge request !140

Apply vertex reordering to boundary segment vertices

Closes staging/dune-uggrid#18

See merge request !145

Adapt to changes of the UG CreateDomain method

See merge request !146

Said method lost three arguments in

  staging/dune-uggrid!45

This patch contains the corresponding changes in the
UGGrid implementation.

UG wants all boundary vertices to be listed before the other ones.
Therefore, the UGGrid GridFactory reorders the vertices before
they get inserted into the actual UGGrid object.

Unfortunately, vertices of explicitly given boundary segments
used the wrong numbering.  Therefore, constructing grids with
explicitly given boundary segments would fail, unless the
vertices happened to be in the correct order.

This fixes staging/dune-uggrid#18

Add missing renumbering in UGGridEntity specialization

See merge request !136

[utility.parmetisgridpartitioner] fix "temporary cannot bind to non-const ref" error

See merge request !142

Use std::array instead of std::vector for offset

See merge request !141

* The size is fixed and known at compile time.
* No more maybe-uninitialized warnings

The `OBJECT` option has to be passed on to `add_dune_mpi_flags`.
Otherwise one will get an error when using the parallel version of UG
with object libraries:

```
CMake Error at [...]/dune-common/cmake/modules/DuneMPI.cmake:62 (target_link_libraries):
  Object library target "[...]" may not link to anything.
Call Stack (most recent call first):
  [...]/dune-grid/cmake/modules/UseUG.cmake:206 (add_dune_mpi_flags)
  [...]/CMakeLists.txt:62 (add_dune_ug_flags)
```

Change module version number to 2.6-git

See merge request !139

Allow using non-double ctype for YaspGrid in dgf parser

Yasp allows this feature, the DGF parser should not shadow it.

See merge request !137

add conversion of PartitionIteratorType to PartitionSet

This patch adds a function "partitionSet<PartitionIteratorType>()" to convert
a given PartitionIteratorType into a PartitionSet. This helps in the transition
from PartitionIteratorTypes to PartitionSets.

See merge request !45

Do not check non existent geometries

This MR introduces hasGeometry<Grid,codim> and makes sure that the grid check does not test these geometries for subentities.

It turns out that all the grids I use do implement all geometries (my previous claim was an oversight).

Unfortunately I had to base this on !114 which needs to get merged before. I will mark this as WIP until that time.
Closes #14 
Closes #40

See merge request !117

Yasp allows this feature, the DGF parser should not shadow it.

The VTK tests are close to the timeout on the CI system and sometimes
fail.

In addition the parallel vtktest has one process busy with polling in a
MPI function, taking 100% CPU for doing nothing, while the main process
is running the Python script. In case of parallel builds, this
unexpected resource usage slows down the test even more.

Correct spelling errors [ci skip]

connot    -> cannot
  verticies -> vertices

See merge request !133

Ansgar Burchardt authored 8 years ago and Christoph Grüninger committed 8 years ago

  connot    -> cannot
  verticies -> vertices

[bugfix] add missing header guards and headers to make headercheck pass

See merge request !127