Hardware-Gated Tests

Some tests require hardware or files that are not available in a standard CI environment. These tests are gated behind markers so they are skipped by default and only run when explicitly opted into.

Markers

Marker	Requirement	How to enable
`slow`	No hardware needed, but takes several seconds	`mise run test-slow` or `--run-slow`
`gpu`	CUDA GPU	`mise run test-gpu` or `--run-gpu`
`mlx`	Apple Silicon with `corridorkey-mlx` installed	`--run-mlx`

Unmarked tests must always pass with no special hardware or files.

How Markers Work

The root conftest.py registers the custom CLI flags and automatically skips marked tests unless the corresponding flag is passed. This applies to all packages in a single run.

# conftest.py (repo root)
def pytest_addoption(parser):
    parser.addoption("--run-slow", action="store_true", default=False)
    parser.addoption("--run-gpu", action="store_true", default=False)
    parser.addoption("--run-mlx", action="store_true", default=False)

def pytest_collection_modifyitems(config, items):
    for item in items:
        if "slow" in item.keywords and not config.getoption("--run-slow"):
            item.add_marker(pytest.mark.skip(reason="Pass --run-slow to run"))
        if "gpu" in item.keywords and not config.getoption("--run-gpu"):
            item.add_marker(pytest.mark.skip(reason="Pass --run-gpu to run"))
        if "mlx" in item.keywords and not config.getoption("--run-mlx"):
            item.add_marker(pytest.mark.skip(reason="Pass --run-mlx to run"))

This means you never need to pass -m on the command line. The default mise run test just works.

Applying a Marker

Apply markers at the class or function level.

# Entire class is slow
@pytest.mark.slow
class TestGreenFormer:
    def test_forward_pass(self): ...

# Single test requires GPU
@pytest.mark.gpu
def test_process_frame_on_cuda(): ...

# Combined: slow and requires GPU
@pytest.mark.slow
@pytest.mark.gpu
def test_full_pipeline_4k(): ...

Coverage and Omitted Modules

Hardware-dependent modules are excluded from coverage measurement entirely via omit in each package's pyproject.toml. This prevents the coverage threshold from being dragged down by code that is legitimately untestable in CI.

Current exclusions:

Package	Module	Reason
`corridorkey-core`	`engine.py`	Requires a real checkpoint file
`corridorkey`	`service.py`	Requires GPU, real checkpoint, and filesystem
`corridorkey`	`ffmpeg_tools.py`	Requires FFmpeg installed as a system dependency
`corridorkey`	`frame_io.py`	Requires real video/image files on disk
`corridorkey`	`device_utils.py`	Requires CUDA or MPS hardware

Additionally, # pragma: no cover is used on individual branches within covered files (e.g. the MLX and Torch branches in create_engine) where the branch requires hardware but the surrounding code is otherwise testable.

Do not add # pragma: no cover or omit entries to avoid writing a test. They are only for code that genuinely cannot run in CI.

Coverage Threshold

The fast suite enforces a minimum of 75% combined coverage across all packages. Run mise run test-cov to check. The threshold is defined in the root pyproject.toml:

[tool.coverage.report]
fail_under = 75

When you add new testable code, add tests for it. If coverage drops below 75%, the test-cov task will fail.