Skip to content

releasing.md

Latest commit

 

History

History
64 lines (49 loc) · 2.73 KB

releasing.md

File metadata and controls

64 lines (49 loc) · 2.73 KB
Raw

Publishing BEND 2.0

The code, Python package and leaderboard are released from this repository. Versioned benchmark inputs remain on Hugging Face; their revisions and checksums are recorded in src/bend2/assets.json.

Website

Enable GitHub Pages in the repository settings and choose GitHub Actions as the source. The Deploy GitHub Pages workflow publishes the website from main at https://machinelearninglifescience.github.io/BEND2/. It can also be run manually. The workflow tests the benchmark and builds reviewed submissions before deploying.

PyPI with the local credential

The first release uses the existing PyPI token in macOS Keychain, stored for https://upload.pypi.org/legacy/ with username __token__. Twine reads it directly; do not put the token in a command, file or repository secret.

python -m build
python -m twine check --strict dist/*
TWINE_USERNAME=__token__ python -m twine upload --non-interactive dist/*

Use a clean dist/ containing only the intended version. Test installation from the built wheel before uploading.

Optional GitHub trusted publishing

To activate the prepared Publish to PyPI workflow, add a GitHub trusted publisher in the bend2 project's PyPI settings with these exact values. For a project that does not exist yet, use a pending publisher at https://pypi.org/manage/account/publishing/ instead.

Field Value
PyPI project name bend2
GitHub owner MachineLearningLifeScience
Repository BEND2
Workflow filename publish.yml
Environment pypi

Create the pypi GitHub environment with a maintainer as a required reviewer. Publishing uses short-lived GitHub credentials; no PyPI API token is stored in the repository. See PyPI's setup instructions.

Release a version

  1. Update the version in pyproject.toml, CITATION.cff and src/bend2/__init__.py.
  2. Confirm the benchmark tests and website build pass.
  3. Create a vVERSION tag and a GitHub release with brief release notes for the same commit.
  4. Upload locally, or run Publish to PyPI with that tag after trusted publishing is configured and approve its pypi environment deployment.

The publishing workflow builds the source distribution and wheel, checks their metadata, installs the wheel and runs the workflow tests. A separate job uploads those artifacts and their provenance attestations to PyPI. This workflow is manual; creating a GitHub release does not trigger an upload.

After publication, verify installation from PyPI in a fresh environment. The package can be released before the website. PyPI does not allow replacing an uploaded version; package corrections need a new version.