Closes #3100.

Usage: python3 scripts/test/gmldiff.py --help

Checks point coordinates only!

Allows for e.g. utils only tests with:

```
-DOGS_BUILD_UTILS=ON -DOGS_BUILD_CLI=OFF
```

EXECUTABLE argument is now required on AddTest().

WORKING_DIRECTORY.

Adapt utils tests to new working directory.

Default working directory for tests was changed from source to binary dir.

- PYTHON_PACKAGES allows for specification of additional Python module dependencies. Requires the poetry-tool and specifying the 'poetry'-wrapper.
- WORKING_DIRECTORY_SOURCE reverts to previous behavior of running the tests in the source tree. Also adapted MeshTest() for this.

Usage:

ctest --print-labels

ctest -L "Steady*|Liquid*" -E LARGE

This runs all benchmarks defined in subdirs starting with Steady OR Liquid
excluding LARGE benchmarks.

Unfortunately AND is not supported in the labels expression yet:

https://gitlab.kitware.com/cmake/cmake/-/issues/21087

Parallelism is controlled via CTEST_PARALLEL_LEVEL env var.

Ensures that the benchmark uses the same interpreter as the build
used for the bindings. Will fix ci mac-job.

Use env var CTEST_PARALLEL_LEVEL instead.

Store commands to shell script to be executed once.

Lars Bilke authored 5 years ago and Dmitri Naumov committed 5 years ago

When a benchmark uses Python (REQUIREMENTS OGS_USE_PYTHON) and
there is a requirements.txt file in benchmarks project file
directory a Python virtual environment is created and dependencies
are installed during benchmark execution via ctest (inside the
build directory).

The PYTHONPATH environment variable of the benchmark test is set
to the virtual environment to mimic virtual environment activation.

Normally you would use

source .venv/bin/activate

but this is not possible when executing the benchmarks via ctest.

Lars Bilke authored 5 years ago and Dmitri Naumov committed 5 years ago

mgis tests are now run along with our benchmarks.