Modularise and improve profiling

[MarionBWeinzierl]

As described in #81 , the profiling should be pulled out of the general testing, and in particular the profiling directory, and given its own module. It should then be extended to not only cover the pmodel. We should consider how we want to potentially automate things like creating the call graph etc.

Tasks:

• [x] Make profiling its own module.
• [x] Extend profiling beyond pmodel.
• [x] Automate call graph creation.
• [x] Consider additional profiling outputs and how to collect them (automatically).
• [x] Adapt and extend the documentation about profiling in the CONTRIBUTING.md file.

[tztsai]

I think there could be two solutions:

1. The first one would be to create a new folder profiling in the root directory and move tests/pmodel/test_profiling.py to profiling/test_pmodel_profiling.py. And we also create profiling scripts for the T Model and the Hygro model, as in the following picture:

Consequently, we need to update .github/workflows/pyrealm_ci.yaml to run the profiling separately (although this should be addressed later in issue #142), somewhat similar to this:

  test:
    ...

    steps:
    - ...

    - name: Profiling
      if: github.ref == 'refs/heads/main'
      run: poetry run pytest profiling/ --profile

2. The second one would be to add a conftest.py in the tests/ folder with the following content:

   This will make all classes or functions decorated by pytest.mark.profiling skipped during testing unless the argument --profiling is provided. An example of marking the tests in tests/pmodel/test_profiling.py would be

   

Consequently, we also need to update .github/workflows/pyrealm_ci.yaml later, somewhat similar to this:

  test:
    ...

    steps:
    - ...

    - name: Run tests with profiling
      if: github.ref == 'refs/heads/main'
      run: poetry run pytest --cov-report xml --profile

   - name: Run tests without profiling
      if: ${{github.ref != 'refs/heads/main'}}
      run: poetry run pytest --cov-report xml

This creates two different commands to run conditional on whether the branch is main.

---

The second solution seems more flexible, and we don't have to write the same code twice if we want to both test it and profile it. However, it profiles all tests when the argument --profile is provided. I don't know if this is the intended behaviour. If we only want to profile certain tests contained in files like test_profiling.py, the first solution may be better.