Configure a pkg config file only if the template file exists

See merge request !1501

Fix for DiagonalMatrix * OtherMatrix

Closes #400

See merge request !1479

Fix CMake warning CMP0177

See merge request !1510

Add MPI flags if ParMETIS is used

See merge request !1500

Ignore undefined policies if QUIET is passed to dune_policy(SET)

See merge request !1511

Remove add_dune_all_flags from dune_add_test

See merge request !1207

Add default include directories to headercheck targets

See merge request !1512

[dunecontrol] Skip directories without CMake config files for module_DIR

See merge request !1473

Markus Blatt authored 3 months ago and Christoph Grüninger committed 3 weeks ago

find_package(module) will fail if the directory set with module_DIR
does not contain CMake configuration file.

Hence the current approach requires that there are pkgconfig files
installed and found for each module to be used.

With this change we fall back to installed modules if the directory
does not contain any Cmake configuration files.

This fixes problems with using OPM as DUNE modules (in Debian) which
stopped shipping broken pkgconf file 2023-09

Add preprocessor macro DUNE_ASSUME

See merge request !1346

IsDenseMatrix for transposed matrix-view

See merge request !1496

[cleanup] Remove old code in Dune::Timer

See merge request !1504

Santiago Ospina De Los Ríos authored 1 month ago and Christoph Grüninger committed 1 month ago

Storing duration in doubles is sub-optimal as each cast may incur in truncation.
Accumulations may also truncate the value further.
Using the native value preserves the duration unit user request its value in seconds.

Use duration_cast instead of division:
Interestingly, dividing by 1.0s yields a different (slightly longer) assembly code than when making an explicit cast to seconds.
The difference seems to me is the cast manages to generate SEE vector registers/instructions.

Carsten Gräser authored 2 months ago and Christoph Grüninger committed 1 month ago

`Dune::Timer` includes two alternative implementations
for measuring time. Historically one was based on `getrusage()`
and the other one on `std::clock()`. The former was the default
and the latter could be enabled by defining a macro. Both in fact
measured the time spend computing by the process.

In `fa43f4bf` the default code path was changed from
`getrusage()` to `std::chrono::high_resolution_clock::now()`
while `std::clock()` could still be enabled by the macro.
This was in fact a breaking change because the new default
version measures the elapsed real time.

This patch removed the non-default version based on `std::clock` because:

* The breaking change happened almost 11 years so one can consider the
  new behavior the established 'correct' version.
* The macro switch was neither documented nor tested.
* Setting the macro manually in user code is error prone.
* Both versions do completely different things leading to
  different measurements. It's not even clear that one
  produces larger numbers in general:
  `std::clock` adds up time spend computing in all threads of the
  process. Thus time in concurrent threads is added up, while
  the time a thread is sleeping is not counted.

There a minor grain of salt: The documentation still documented
the old behaviour and instead of 'fixing' the code this patch
adjusts the documentation to established reality.

Fix wrong license in build system doc

See merge request !1508

Move build system doc license to another file

See merge request !1507

More constexpr!

See merge request !1459

Santiago Ospina De Los Ríos authored 4 months ago and Christoph Grüninger committed 1 month ago

Add constexpr qualifiers on many functions
Add constexpr backport versions of std::abs and std::sqrt

Install libraries in root build path instead on project build path

See merge request !1467

fix some issues with deprecations for pypi packaging

See merge request !1503

Ignore unused variables in debugaligntest

See merge request !1505

Add preprocessor macro DUNE_FORCE_INLINE

See merge request !1502