Merge branch 'feature/support-manually-set-ninja-in-fortran-test' into 'master'

ref:core/dune-common Prior to this change the build program (make, Ninja) was
not identical to the one used by CMake, but the one installed on the system.

I found this when testing Fortran support of upcoming CMake 3.17 with upcoming
Ninja 1.10 - which works!

See merge request [!761]

  [!761]: gitlab.dune-project.org/core/dune-common/merge_requests/761

Prior to this change the build program (make, Ninja) was
not identical to the one used by CMake, but the one
installed on the system.

Merge branch 'fix-python-venv-debian-testing' into 'master'

ref:core/dune-common cmake parameters are passed as a list. The "-p
${PYTHON_EXECUTABLE}" option was passed as string, which lead to problems
setting up the venv on my Debian testing system.

We now pass all parameters as lists, just like it should be\...

1.  @core: perhaps someone from the cmake and/or python crowd should have a
    look\...
2.  I guess this wuld be god in 2.7?!

See merge request [core/dune-common!747]

  [core/dune-common!747]: gitlab.dune-project.org/core/dune-common/merge_requests/747

Merge branch 'add_missing_install_header' into 'master'

See merge request [core/dune-common!754]

  [core/dune-common!754]: Nonecore/dune-common/merge_requests/754

Merge branch 'issue/add_missing_install_file_quadmath' into 'master'

ref:core/dune-common Maybe this MR could also be cherry-picked into the
releases/2.7 branch

See merge request [!759]

  [!759]: gitlab.dune-project.org/core/dune-common/merge_requests/759

Merge branch 'gcc10' into 'master'

See merge request [!758]

  [!758]: Nonecore/dune-common/merge_requests/758

GCC 10's libstdc++ deleted some `operator<<` overloads to write wide
characters to non-wide streams as these would be treated as integers
instead of characters. One could use wide variants `wstringstream`,
`wstring` instead of the usual `stringstream`, `string`, but at some
places would have to convert between different character
representations.

As wide characters really should only be used for text, DUNE doesn't
really need to support them as numerical types here. So we just drop
the tests using wide character types.

This addresses compiler errors like
```
../dune/common/simd/test.hh:1732:20: error: use of deleted function ‘std::basic_ostream<char, _Traits>& std::operator<<(std::basic_ostream<char, _Traits>&, wchar_t) [with _Traits = std::char_traits<char>]’
In file included from /usr/include/c++/10/iostream:39,
                 from ../dune/common/simd/test.hh:12,
                 from dune/common/simd/test/standardtest_vector_wchar_t.cc:5:
/usr/include/c++/10/ostream:538:5: note: declared here
  538 |     operator<<(basic_ostream<char, _Traits>&, wchar_t) = delete;
```

Several members of `std::allocator` were already deprecated in C++17
and have now been removed in C++20. This patch uses the appropriate
members of `std::allocator_traits` instead.

Merge branch 'fix/issue-183-skipped-tests' into 'master'

ref:core/dune-common Fixes [#183].

See merge request [!755]

  [#183]: gitlab.dune-project.org/NoneNone/issues/183
  [!755]: gitlab.dune-project.org/core/dune-common/merge_requests/755


Closes #183

This fixes the bug (#183) introduced with a225b2da.
The command can be a complex path while for skipped tests
we actually want the test command to just be equal to
the target name (executable). This name is
expanded to a full path using the TARGET_FILE generator
expression so it should work on all platforms that CMake
supports.

Merge branch 'fix/install-latex-doc-optional' into 'master'

ref:core/dune-common Otherwise target "install" fails when target "doc" was
not run.

Fixes [#182]

See merge request [!751]

  [#182]: gitlab.dune-project.org/NoneNone/issues/182
  [!751]: gitlab.dune-project.org/core/dune-common/merge_requests/751


Closes #182

Otherwise target "install" fails when target "doc" was
not run.

Merge branch 'bugfix/headercheck-mpifuture-mpipack' into 'master'

ref:core/dune-common mpifuture.hh and mpipack.hh missed includes for
Communication.

See merge request [!752]

  [!752]: gitlab.dune-project.org/core/dune-common/merge_requests/752

Merge branch 'install-FindQuadMath.cmake' into 'master'

ref:core/dune-common In is used downstream and needs to be installed.
Otherwise cmake for e.g. dune-geometry fails.

See merge request [!749]

  [!749]: gitlab.dune-project.org/core/dune-common/merge_requests/749

In is used downstream and needs to be installed. Otherwise cmake for
e.g. dune-geometry fails.

cmake parameters are passed as a list. The "-p ${PYTHON_EXECUTABLE}"
ption was passed as string, which lead to problems setting up the venv
on my Debian testing system.

We now pass all parameters as lists, just like it should be....

Merge branch 'fixes-for-mingw' into 'master'

ref:core/dune-common We fix two general issues f the build-system regarding
mingw/windows support.

@gruenich and again\... can we get this into 2.7?

See merge request [!744]

  [!744]: gitlab.dune-project.org/core/dune-common/merge_requests/744

Merge branch 'run-mpipacktest-only-with-mpi' into 'master'

ref:core/dune-common @gruenich can we merge this to 2.7?

See merge request [!743]

  [!743]: gitlab.dune-project.org/core/dune-common/merge_requests/743

skipped test targets are manually overwritten. We have use the TARGET_FILE
and can not create the name manually, as this will fail when the target name
differs from the executable name.

the fallback test used an invocation via sh to capture non-zero
exit codes as non-fatal. This didn't work on win32 as the hard-coded
command name was wrong (missing exe extension).

We now explicitly build two different executables with slightly different
error handling.

Merge branch 'fix-cmake-latexmk' into 'master'

ref:core/dune-common When cross compiling for windows it is necessary to tweak
CMAKE_FIND_ROOT_PATH_MODE_* to modify default behavior of FIND_XXX() commands
to search for headers/libs in the target environment and search for programs
in the build host environment.

This lead to a problem when searching for latexmkrc.cmake.

We now explicitly change the UseLatexMk find_file call to ignore the
CMAKE_FIND_ROOT_PATH and thus beeing able to find the file in the source
directory.

@gruenich can we backport this to 2.7?

See merge request [core/dune-common!740]

  [core/dune-common!740]: gitlab.dune-project.org/core/dune-common/merge_requests/740

When cross compiling for windows it is necessary to tweak
CMAKE_FIND_ROOT_PATH_MODE_* to modify default behavior of FIND_XXX()
commands to search for headers/libs in the target environment and
search for programs in the build host environment.

This lead to a problem when searching for latexmkrc.cmake.

We now explicitly change the UseLatexMk find_file call to ignore the
CMAKE_FIND_ROOT_PATH and thus beeing able to find the file in the
source directory.

Merge branch 'feature/bump-to-2.8' into 'master'

See merge request [!739]

  [!739]: Nonecore/dune-common/merge_requests/739

Merge branch 'feature/deprecate-collective-headers' into 'master'

ref:core/dune-common They will be removed after Dune 2.7.

See merge request [!738]

  [!738]: gitlab.dune-project.org/core/dune-common/merge_requests/738

They will be removed after Dune 2.7.

Merge branch 'feature/variablesizecomm-copy' into 'master'

ref:core/dune-common This basically implements the rule of 3, as we have a
custom destructor. Without this patch, copying a VariableSizeCommunicator will
lead to double frees.

Supersedes [!690]

See merge request [!737]

  [!690]: gitlab.dune-project.org/NoneNone/merge_requests/690
  [!737]: gitlab.dune-project.org/core/dune-common/merge_requests/737

Lasse Hinrichsen-Bischoff authored 5 years ago and Christoph Grüninger committed 5 years ago

This basically implements the rule of 3, as we have a custom destructor.
Without this patch, copying a VariableSizeCommunicator will lead to double frees.

Merge branch 'fix/mpi-cmake' into 'master'

ref:core/dune-common Providing space-separated MPI_DUNE_COMPILE_FLAGS would
previously add them as a single list entry to the cmake list COMPILE_OPTIONS
and thus ruin the command-line having something like

    /usr/bin/c++ "-arg1 -arg2" -other -args

Supersedes [!709]

See merge request [!736]

  [!709]: gitlab.dune-project.org/NoneNone/merge_requests/709
  [!736]: gitlab.dune-project.org/core/dune-common/merge_requests/736

Stephan Hilb authored 5 years ago and Christoph Grüninger committed 5 years ago

Providing space-separated MPI_DUNE_COMPILE_FLAGS would previously add
them as a single list entry to the cmake list COMPILE_OPTIONS and thus
ruin the command-line having something like

    /usr/bin/c++ "-arg1 -arg2" -other -args

Merge branch 'fix/fix-inconsistent-1d-dense-vector-conversions' into 'master'

ref:core/dune-common The commit also adds a test which fails without the other
changes in the commit.

Problem is that DenseVector and FieldVector\<\..., 1\> implement catch-all
assignment constructors and ctors. I.e., the 1d FieldVector provides CTOR
call-signatures and the DenseVector<C> provides and assignment operator

std::is_assignable<To, From\> std::is_convertible<From, To\>

This poses a problem when nesting vectors and trying to assign to them from
another DenseVector implementation.

The problem is that the implementation of the CTOR and the assignment
operation fails, but OTOH the FieldVector implementation uses
std::is_convertible. Now std::convertible will report "yes" but the body of
the assignment operation will fail nevertheless.

The commit fixes this by restricting the call-signature to the cases where the
implementation of the assignment and the CTOR will succeed but testing the
assignability of DV::value_type.

Supersedes [!562] as this is a rebased and squased version.

See merge request [!735]

  [!562]: gitlab.dune-project.org/NoneNone/merge_requests/562
  [!735]: gitlab.dune-project.org/core/dune-common/merge_requests/735

Claus-Justus Heine authored 6 years ago and Christoph Grüninger committed 5 years ago

The commit also adds a test which fails without the other changes in the
commit.

Problem is that DenseVector and FieldVector<..., 1> implement catch-all
assignment constructors and ctors. I.e., the 1d FieldVector provides
CTOR call-signatures and the DenseVector<C> provides and assignment
operator

std::is_assignable<To, From>
std::is_convertible<From, To>

This poses a problem when nesting vectors and trying to assign to them
from another DenseVector implementation.

The problem is that the implementation of the CTOR and the assignment
operation fails, but OTOH the FieldVector implementation uses
std::is_convertible. Now std::convertible will report "yes" but the body
of the assignment operation will fail nevertheless.

The commit fixes this by restricting the call-signature to the cases
where the implementation of the assignment and the CTOR will succeed but
testing the assignability of DV::value_type.

Merge branch 'fix_future_of_void' into 'master'

ref:core/dune-common There was a problem with a ctor of Future<void>. This MR
disables the ctor if T is void. Furthermore i changed the autos in the
mpifuturetest.cc to Futures to check for this.

See merge request [!721]

  [!721]: gitlab.dune-project.org/core/dune-common/merge_requests/721

Merge branch 'update-build-system-doc' into 'master'

See merge request [!732]

  [!732]: Nonecore/dune-common/merge_requests/732

Merge branch 'update-generic-mpi-op' into 'master'

See merge request [core/dune-common!733]

  [core/dune-common!733]: Nonecore/dune-common/merge_requests/733