git clone https://github.com/duceppemo/primer-finder && cd primer-finder
conda env create -f environment.yml && conda activate primer-finder
pip install -e ".[test]"
pre-commit install

primer_finder/
├── cli.py        the command line: `find` (the default), `design` and `idt`
├── pipeline.py   the steps of `find`, in order, and what each writes
├── kmers.py      KMC: counting, subtracting, dumping
├── assemble.py   the unitig walk, SKESA and SPAdes, and the fallback between them
├── mapping.py    minimap2, cigar strings, which contigs are candidates
├── blast.py      makeblastdb/blastn, presence, and what differs where
├── design.py     the steps of `design`: Primer3, the amplicon checks, the order
├── insilico.py   the primer files insilicoPCR reads, running it, and its report
├── idt.py        IDT order sheets (.xlsx without any library, .csv, .tsv)
├── seqio.py      fasta reading and writing, gzipped or not
├── tools.py      running external programs, their versions, and their time limits
└── system.py     how many CPUs and how much memory may be used
example/          the simulated dataset, the runner and the checks
validation/       the published assays this is checked against, and dated records
tests/            the test suite, with stand-ins for the external programs
docs/wiki/        the sources of this wiki
recipe/           a copy of the bioconda recipe

primer-finder uses the Python standard library only. Everything else is an external program called on the command line, which keeps the conda environment small and the tests fast. Please keep it that way.

pytest -q                                  # the whole suite, a few seconds
pytest --cov=primer_finder --cov-report=term-missing
pytest -m tools                            # only the example, with the real programs
ruff check .

tests/stub_programs.py holds a stand-in for every external program, and tests/stub_insilico_pcr.py one for insilicoPCR; the stubs fixture puts them first on PATH and lets each test say what they should produce (the kmers KMC dumps, the contigs the assembler writes, the SAM records, the blast hits, the assays Primer3 proposes, which samples amplify). The whole pipeline is therefore tested without conda, in seconds, including its error paths. The test marked tools runs example/run_example.sh with the real programs and is skipped when they are not installed.

Add a test with every fix, and check that it fails without the fix.

.github/workflows/:

The wiki is maintained in docs/wiki/ in this repository and published by wiki.yml. Do not edit the pages on GitHub directly: the next push overwrites them.

1. Bump the version in pyproject.toml, primer_finder/__init__.py, CITATION.cff (version and date-released) and recipe/meta.yaml; add the section to CHANGELOG.md. tests/test_cli.py checks that all five agree.
2. pytest and bash example/run_example.sh green; update the tarball version in the README and in docs/wiki/Installation.md.
3. Update what the five-file test cannot see: date-released and doi in CITATION.cff (the concept DOI until Zenodo mints the version one), the recipe's sha256 (only computable once the tag exists), and the version-dependent prose — any "since the release after X" in the FAQ, the sample log in docs/wiki/Usage.md, and the expected output in docs/wiki/Example.md if the run's output moved.
4. Tag and push: git tag v1.0.1 && git push --tags. release.yml publishes the release, and publish.yml then uploads the wheel and the sdist to PyPI.
5. Zenodo archives the release and mints a DOI; add the new version DOI to CITATION.cff.
6. Update the bioconda recipe: bump version, recompute the sha256 of the release tarball (curl -sL <url> | sha256sum), reset number to 0, and open a pull request against bioconda-recipes (its autobump bot often does it first). Keep recipe/meta.yaml in this repository the same as the one there.