test assignment of matrices with cast operator

Some matrix implementations, e.g., the Jacobian implementations of SPGrid have
cast operators into the corresponding FieldMatrix type. This patch adds a
simplified version of such a cast operator to the densematrixassignment test.

See merge request !139

Update test environments

- No longer test with clang 3.5.
- Add clang 3.8 from Debian 8 + backports
- Add gcc 5.4 + clang 3.8 from Ubuntu 16.04 LTS.
- Use `duneci-standard-test`

See merge request !146

gcc 5.4 and clang 3.8.

Reference: core/dune-common#37

[Singleton] Make creation of the instance thread-safe.

Closes: core/dune-common#46.

See merge request !141

It should compile for the FieldMatrix class but not DynamicMatrix

[CMake] always use check for C++14

This prevents users to run into compiler errors for unsupported
compilers like GCC 4.8 or C++ std libraries with insufficient
C++14 support.

See merge request !120

This prevents users to run into compiler errors for unsupported
compilers like GCC 4.8 or C++ std libraries with insufficient
C++14 support.

Implement workarounds for broken sizeof...

This adds `Dune::SizeOf`  that can be use as `Dune::SizeOf<T...>::value` as a drop in replacement for `sizeof...(T)`. This is necessary because the latter is broken in clang<3.8.
Because the bug also effects `std::index_sequence_for` we now __always__ provide our manually written fallback implementation `Std::index_sequence_for` that works around the bug by using `SizeOf`.

As a rule of thumb you should always use `SizeOf` instead instead of `sizeof...` in variadic template aliases if you want to avoid compiling code that produces wrong results.

See merge request !127

Make FieldVector::size() static

Being static allows to use size() in constexpr contexts. This implements #2.

See merge request !144

Being static allows to use size() in constexpr contexts.

[cmake] Fix empty unset

That was to much empty paratheses in 4b0edfc2…

See merge request !143

That was to much empty paratheses in 4b0edfc2…

Closes: core/dune-common#46.

Add `dune-ctest`, a wrapper around `ctest`

CTest returns with an error status not only when tests failed, but also
when tests were only skipped.  This wrapper checks the log and returns
successfully if no tests failed; skipped tests do not result in an error.
This behaviour is needed in a continuous integration environment, when
building binary packages or in other cases where the testsuite should be
run automatically.

Closes issue #12

See merge request !132

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

CTest returns with an error status not only when tests failed, but also
when tests were only skipped.  This wrapper checks the log and returns
successfully if no tests failed; skipped tests do not result in an error.
This behaviour is needed in a continuous integration environment, when
building binary packages or in other cases where the testsuite should be
run automatically.

Closes issue #12

Cleanup/cmake formatting



See merge request !138

This patch disables the implicit cast operator to FieldMatrix or
DynamicMatrix, if there is no suitable DenseMatrixAssigner. This way,
the compiler will start looking for cast operators on the right hand
side, again.

* mainly emtpy if(), else() and so on
* macro lowercase
* no space after if or macro

[cmake] Rename Doxyfile target to prevent name clashes

The output is called Doxyfile and the target should have a
different name, otherwise Ninja warns about it.

See merge request !137

The output is called Doxyfile and the target should have a
different name, otherwise Ninja warns about it.

Some matrix implementations, e.g., the Jacobian implementations of SPGrid have
cast operators into the corresponding FieldMatrix type. This patch adds a
simplified version of such a cast operator to the densematrixassignment test.

FieldMatrix: Allow copy construction from scalar

See also !52.

See merge request !140

See also !52

Rework DenseMatrix assignment

This is meant as a fix for bug #5.

It
* Enables assignment across different `DenseMatrix` implementations, e.g. assignment from a `DynamicMatrix` to a `FieldMatrix` (this previously either compiled and segfaulted or did not compile)
* Catches invalid `FieldMatrix` assignments at compile-time
* Retaines the `DenseMatrixAssigner` class in order to remain compatible with user-defined matrix classes

~~To me it looks 99% good to go. There is only remaining question that I cannot answer on my own:~~
* ~~Should `FieldMatrix<double, 3, 3> x = 27` compile? In other words, should the constructor be marked as `explicit` or not? Of course, one can always use `FieldMatrix<double, 3, 3> x(27)` instead.~~

I've tested this fix with multiple modules and it doesn't break anything for me.

See merge request !52

This addresses issue #5

Deprecate the identitymatrix.hh file

This patch makes the preprocessor emit a #warning whenever the file
identitymatrix.hh is included.

Reasons:
- It is not used anywhere in Dune core or staging or any other module I know.
- I have never heard a compelling reason to have it.
- It produces maintenance costs
  (see core/dune-common!52)
- IIRC it entered dune-common without discussion.

See merge request !136

This patch makes the preprocessor emit a #warning whenever the file
identitymatrix.hh is included.

Reasons:
- It is not used anywhere in Dune core or staging or any other module I know.
- I have never heard a compelling reason to have it.
- It produces maintenance costs
  (see !52)
- IIRC it entered dune-common without discussion.

This reverts merge request !135

[cmake][bugfix] Correctly determine scripts directory with dune_module_path

The previous implementation assumed that dune-common exports the path,
but it does not.

See merge request !135

The previous implementation assumed that dune-common exports the path,
but it does not.

[parametertree] Fix test for non-existent subtree

With the recent change to return an empty subtree for non-existent
subtrees (cf. !101), the test for the `ParameterTree got broken.

This patch fixes the issue by testing for both the old
behavior (explicitly activated) and the new behavior.

This closes #33 

See merge request !104