Releasing icon-uv

The package builds a pure-Python wheel and source archive with setuptools and uv build. The wheel includes the radiation table, the daily JSON schema and the license. Its dependencies may install native libraries. Tests and builds run on Linux with Python 3.11 and 3.13; they do not need forecast-service credentials.

Prepare a GitHub release

1. Set the same version in pyproject.toml and icon_uv/__init__.py, update CHANGELOG.md and any version-specific installation notes, then run uv lock.
2. Use a fresh output directory and run the release checks:

sh uv sync --locked --extra cams uv run --no-sync pytest -q uv build --out-dir work/release/0.3.0 uv run --no-sync python tools/check_distribution.py work/release/0.3.0 --tag v0.3.0 uv run --no-sync twine check --strict work/release/0.3.0/*

1. Commit and push the changes. Wait for Tests and distributions to pass for that commit, then create an annotated vVERSION tag on it and push the tag.
2. Download the tested distributions artifact from that commit's CI run. Attach its wheel and source archive, plus a SHA-256 checksum file, to a GitHub release using that tag. Use the changelog entry as the release notes.

The distribution checker verifies resources, matching wheel/source versions, the release tag and the installed __version__. It installs the wheel outside the source tree and exercises its schema and radiation table. Twine checks the package metadata and README. README links are absolute so they also work on PyPI.

One-time PyPI setup

Sign in to the PyPI account that will own the package and add a pending Trusted Publisher:

Field	Value
PyPI project name	icon-uv
GitHub owner	ofuhrer
GitHub repository	icon-uv
Workflow filename	publish.yml
Environment name	pypi

For an existing project, add the same publisher under its Publishing settings. The repository's pypi GitHub environment should allow deployment from v* tags. The account-side registration must be done on PyPI; adding this workflow alone does not register the publisher or publish the package.

PyPI creates the project on the first successful publish. A pending publisher does not reserve the name. See the official first-project setup guide.

Publish a tested release to PyPI

After the GitHub release exists and the publisher is configured:

gh workflow run publish.yml --ref v0.3.0

Publish to PyPI is manual. It rejects branch refs and draft or missing releases, reruns the CI checks at the selected tag, and publishes the tested wheel and source archive. Only the publishing job receives id-token: write; building and testing happen in separate jobs. No stored PyPI API token is needed. The official PyPA action uses Trusted Publishing and generates provenance attestations by default.

Once publication succeeds, users can install with:

pip install 'icon-uv[cams]==0.3.0'
icon-uv --help

Verify installation in a clean environment before announcing the PyPI release. PyPI versions cannot be overwritten: fix a released package with a new version.