Development

Development

Layout

hashbrown/            the package
├── cli.py            argument parsing + the three pipelines (HashBrown.run_long / run_short)
├── methods.py        input discovery, Sample, resource checks, small file utilities
├── nanopore.py       Porechop, Filtlong, Flye, Shasta, Medaka
├── illumina.py       fastp, SKESA, SPAdes, Polypolish, pyPOLCA
└── assembly_qc.py    minimap2 mapping, Qualimap, QUAST, nucmer dot plots, coverage
tests/                pytest suite; conftest.py stubs every external program
validation/           polisher benchmark against a gold standard
docs/wiki/            these pages (published by .github/workflows/wiki.yml)
docs/images/          logo and pipeline diagram (regenerate with make_svgs.py)
recipe/               bioconda recipe

Tests

pip install -e ".[test]"
pytest
ruff check .

The tests run in seconds without conda: tests/conftest.py builds a folder of fake flye, medaka_consensus, samtools, … executables that mimic just enough of each program's command line, and the pipeline tests run the real orchestration end to end against them (including resuming). A change to how a program is called must update its stub to match the real program's behaviour.

CI and releases

• ci.yml: ruff + pytest on Linux (Python 3.10–3.13) and macOS, coverage to Codecov (needs the CODECOV_TOKEN secret).
• wiki.yml: pushes docs/wiki/ to the GitHub wiki on every change (create one page in the web UI once, so the wiki repository exists).
• release.yml: on a v* tag — checks the tag matches hashbrown.__version__ and pyproject.toml, builds the wheel/sdist, and creates a GitHub release with the matching CHANGELOG.md section.

Releasing: bump the version in hashbrown/__init__.py and pyproject.toml, date the section in CHANGELOG.md, update CITATION.cff, then git tag v<version> && git push --tags. For bioconda, update recipe/meta.yaml with the new tarball sha256 and submit to bioconda-recipes.

Adding an assembler or polisher

1. Add the runner in illumina.py/nanopore.py and wire it in cli.py (choices + dependency map in methods.py).
2. Add its stub in tests/conftest.py and a pipeline test.
3. Benchmark it (validation/benchmark_polishers.sh) — a polisher that adds errors in low-coverage regions does not go in, no matter its reputation.
4. Document it in docs/wiki/Methods.md and CHANGELOG.md, and add it to environment.yml and recipe/meta.yaml.