Testing¶

Running the suite¶

pip install -e ".[dev]"
pytest

pytest.ini points at tests/ and enables live INFO logging, so you see the same reconstructor log lines the library emits at runtime.

Useful invocations:

pytest tests/test_tomographicReconstructor.py    # one module
pytest -k "atmosphere"                           # by name
pytest -x -vv                                    # stop at the first failure, verbose
pytest --cov=pyTomoAO --cov-report=term-missing  # coverage with uncovered lines

Version-specific behaviour is easiest to reproduce in a throwaway environment:

conda create -y -n pytomoao310 python=3.10
conda run -n pytomoao310 pip install ".[dev]"
conda run -n pytomoao310 pytest

What is covered¶

Test module	Covers
`test_atmosphereParametersClass.py`	Layer validation, airmass and `r0` scaling
`test_lgsAsterismParametersClass.py`	Asterism geometry and direction vectors
`test_lgsWfsParametersClass.py`	Lenslet maps, `nValidSubap`, support padding
`test_dmParametersClass.py`	Actuator maps and geometry validation
`test_fitting.py`	Influence functions and the fitting matrix
`test_tomographicReconstructor.py`	End-to-end reconstructor construction

The parameter classes are the cheapest place to add tests, and the place where a bad configuration does the most damage — new validation logic should always come with a test for both the accepted and rejected cases.

The CI gate¶

.github/workflows/test.yml runs on pull requests and on pushes to main and dev. It installs the package itself — pip install ".[dev]" — and runs pytest against a matrix of Python 3.9 through 3.13, so both the code and the dependency metadata are checked on every supported version.

On the 3.12 leg it additionally runs a wrapper script rather than pytest directly:

python .github/github_pytest_workflow.py --fail-on-problems --verbose --coverage-threshold 50

The wrapper runs pytest with coverage and then applies two additional checks:

Coverage threshold — the run fails if total coverage falls below the given percentage (50% in CI; the script’s own default is 75%).
Untested-file heuristic — source files above --line-count-threshold lines (default 150) with no corresponding test file are reported as problems.

With --fail-on-problems set, either condition fails the job. Run the same command locally before opening a pull request if you want to see exactly what CI will say.

Writing tests¶

Use the configurations under examples/benchmark/ rather than inventing new YAML where possible — they exercise realistic geometries.
Reconstructor construction for the full KAPA configuration is not fast; prefer the single-channel or REVOLT configurations for tests that only need a reconstructor.

Test validation failures explicitly, since raising the right exception type is part of the parameter classes’ contract:

import pytest

def test_negative_diameter_rejected(config):
    config["lgs_wfs_parameters"]["D"] = -1
    with pytest.raises(ValueError):
        lgsWfsParameters(config, asterism_params)

GPU code paths cannot run on the CI runner. Guard any CuPy-dependent test with pytest.importorskip("cupy").
Importing a module gives you the module. No module shares a name with a class it defines, so ordinary imports and unittest.mock string targets both behave normally:
```
from pyTomoAO import reconstructor

with patch.object(reconstructor, "atmosphereParameters"):
    ...
```
Before 2.0 the package re-exported tomographicReconstructor and fitting under their own modules’ names, so those dotted paths resolved to the classes. Tests had to reach for importlib.import_module, and string patch targets silently failed on Python 3.9 and 3.10. Both modules were renamed (reconstructor, dm_fitting); the workarounds are gone.

Note

Coverage depends on which backend the machine selects. tomographyUtilsCPU and tomographyUtilsGPU are alternative implementations chosen at import, so whichever one is not loaded reports 0%: the CI runner has no GPU and covers the CPU kernels, while a CUDA-equipped workstation covers the GPU kernels instead. Construct a reconstructor with force_cpu=True if you want to exercise the CPU path on a GPU machine.