Local Workflows¶

This page provides practical local workflows for development and validation.

Setup¶

Install dependencies:

uv sync --group dev

Standard Quality Workflow¶

Run the same core checks used during development:

uv run python -m ruff check .
uv run python -m ruff format --check online_fdr tests
uv run python -m mypy online_fdr
uv run python -m pytest -q

Focused Test Runs¶

Run only the new API contract checks:

uv run python -m pytest tests/test_public_api_surface.py tests/test_import_smoke.py -q

Run a single module:

uv run python -m pytest tests/test_static.py -q

Working with R Parity Tests¶

Parity tests depend on a working R + rpy2 setup and pinned Bioconductor onlineFDR package.

Install parity dependencies:

uv sync --group dev --group parity

Run parity tests directly:

uv run python -m pytest -m live_r_parity tests/test_onlinefdr_parity.py tests/test_async_methods.py -q

The pinned onlineFDR==2.18.0 implementation stores async ADDIS discovery positions as booleans, collapsing start indices above 1. The repeated 20-block case directly checks decision equality, the matching threshold prefix, and the known stage-9 difference. An independent frozen-reference regression requires exact preservation of prior Python behavior. Supported 20-block live R comparisons keep pending tests selected and discoveries at positions 0 and/or 1, covering both single and multiple discoveries. The scalability optimization preserves the prior Python accounting.

If parity setup is not available yet, run the standard suite:

uv run python -m pytest -q

Async Scalability Benchmark¶

Compare asynchronous SAFFRON and ADDIS with the frozen scan-based reference:

uv run python benchmarks/async_scaling.py

The benchmark reports median times over five fresh instances for no-discovery, candidate-heavy, mixed, and discovery-heavy streams. Reference runs stop at 800 tests; optimized no-discovery runs extend through 10,000 tests. The default run takes several minutes because the reference implementation repeatedly scans history. For a shorter check of the original SAFFRON bottleneck:

uv run python benchmarks/async_scaling.py --methods saffron --scenarios no-discovery --sizes 800

Timing results are descriptive. CI uses deterministic record-inspection tests to detect repeated history scans without imposing machine-specific time limits. The optimized methods retain all records and still sum past discovery contributions when assigning a new level, so memory grows with the number of tests and discovery-heavy streams require more work.

Packaging Smoke¶

Build artifacts and verify package metadata locally:

uv run python -m build

Release Smoke¶

Run the package build, wheel install, and import smoke check used by release CI:

uv run python -m build
uv venv .smoke --python 3.12
uv pip install --python .smoke/bin/python dist/*.whl
.smoke/bin/python -c "import online_fdr; print(online_fdr.__version__)"

Documentation Workflow¶

Serve docs locally:

uv run mkdocs serve

Build strict docs:

uv run mkdocs build --strict