Publishing meds-algorithm-template¶

Release meds-algorithm first. This repository publishes a small Copier frontend named meds-algorithm-template; it is not another copy of the shared algorithm library. The command fetches https://github.com/McDermottHealthAI/MEDS_algorithm_template.git at v<installed-version> and records that source/tag in .copier-answers.yml, preserving normal Copier updates. Git and network access are required. The generator wheel contains the frontend, not the template payload.

Use the shared publishing checklist for the exact PyPI/TestPyPI publisher fields and GitHub environment setup. The template values are: project meds-algorithm-template, owner McDermottHealthAI, repository MEDS_algorithm_template, workflow publish.yml, and environments testpypi / pypi. The workflow must be merged into main before GitHub exposes its manual dispatch action.

1. Switch generated projects to the published shared package¶

This development version pins core revision 99b3b59798d796360d1eff74fa75db58d35c4e90. First publish meds-algorithm==0.4.0a5, then replace the development Git requirements with the exact PyPI requirements below before tagging generator 0.4.0a2:

File	Change
`pyproject.toml`, development dependencies	Pin the shared conformance dependency to `meds-algorithm[conformance]==0.4.0a5`
`template/pyproject.toml.jinja`, project dependencies	Use `meds-algorithm[data]==0.4.0a5`
`template/pyproject.toml.jinja`, test extra	Use `meds-algorithm[conformance]==0.4.0a5`
`README.md`	Document the shared PyPI version and update the generator install example to `0.4.0a2`
`uv.lock`	Regenerate with `uv lock`
`template/pyproject.toml.jinja`, Hatch metadata	Remove `allow-direct-references` after replacing all Git requirements

Do not change the generator's runtime dependencies to require the shared library; it only generates files. Preserve historical tutorial/source revision pins as recorded provenance. If updating the manual tutorial to demonstrate the new PyPI dependency, pin it to a real template commit after this cutover, then execute the entire script again.

Validate before merging:

uv lock
uv sync --locked --group dev --group docs
uv run ruff check src tests scripts docs/hooks.py
uv run ruff format --check src tests scripts docs/hooks.py
uv run pytest -q
uv run mkdocs build --strict
uv run python scripts/check_release.py
uv build
uvx --from 'twine>=6' twine check --strict dist/*
wheel_env=$(mktemp -d)
uv venv "$wheel_env"
uv pip install --python "$wheel_env/bin/python" dist/*.whl
"$wheel_env/bin/python" -I tests/wheel_smoke.py "$PWD"

check_release.py enforces matching exact index pins and verifies that the shared release has files on PyPI. It rejects Git requirements and inconsistent version pins. The wheel smoke test uses a temporary local tagged mirror to exercise the installed command, single/all-profile generation and the canonical update URL without creating a real release tag or fetching a nonexistent release.

2. Tag and rehearse on TestPyPI¶

Candidate generator version: 0.4.0a2. Once the release changes are merged and CI passes:

git switch main
git pull --ff-only
git tag v0.4.0a2
git push origin v0.4.0a2
gh workflow run publish.yml --repo McDermottHealthAI/MEDS_algorithm_template \
  --ref v0.4.0a2 -f index=testpypi
gh run list --repo McDermottHealthAI/MEDS_algorithm_template --workflow publish.yml --limit 5
gh run watch RUN_ID --repo McDermottHealthAI/MEDS_algorithm_template --exit-status

The workflow verifies the tag/version and reviewed main history, checks the shared release prerequisite, runs tests and strict docs, and validates the built and installed wheel before upload. It rebuilds the same tag for each index; it does not reuse the previous workflow's exact distribution bytes.

Install dependencies from PyPI, then install only this distribution from TestPyPI:

check_env=$(mktemp -d)
uv venv "$check_env"
uv pip install --python "$check_env/bin/python" 'copier>=9.17.1,<10' 'pyyaml>=6'
uv pip install --python "$check_env/bin/python" --no-deps \
  --index-url https://test.pypi.org/simple/ 'meds-algorithm-template==0.4.0a2'
"$check_env/bin/meds-algorithm-template" --version
"$check_env/bin/meds-algorithm-template" "$check_env/generated" --defaults --trust
uv pip install --python "$check_env/bin/python" "$check_env/generated[test]"
"$check_env/bin/meds-algorithm" profiles
uv pip check --python "$check_env/bin/python"

Generation alone does not implement an algorithm; the generated completion gates remain in place. Do not move a published tag to repair a release. Increment the package version and use a new tag.

3. Publish the generator to PyPI¶

After the TestPyPI rehearsal succeeds:

gh workflow run publish.yml --repo McDermottHealthAI/MEDS_algorithm_template \
  --ref v0.4.0a2 -f index=pypi
# Find and watch the new run, then approve the pypi environment deployment.
# After it succeeds, verify in another clean environment:
pypi_env=$(mktemp -d)
uv venv "$pypi_env"
uv pip install --python "$pypi_env/bin/python" 'meds-algorithm-template==0.4.0a2'
"$pypi_env/bin/meds-algorithm-template" "$pypi_env/generated" --defaults --trust
uv pip install --python "$pypi_env/bin/python" "$pypi_env/generated[test]"
"$pypi_env/bin/meds-algorithm" profiles
uv pip check --python "$pypi_env/bin/python"

Verify the PyPI badge and install commands after publication. Future versions retain the same order whenever the template requires a new shared-package version. Pin both generator and shared-package versions for reproducible runs.