build: modernize forum repo to use uv, pyproject.toml, and semantic-release

[farhan]

Important

PR implemented with the assistance of Claude Code. Refined and validated before being submitted for code review.

Modernize forum
Part of openedx/public-engineering#506

Summary

• Repo publishes to PyPI as openedx-forum
• Move the package into a src/ layout
• Replace setup.py/setup.cfg with pyproject.toml (PEP 621 static metadata)
• Switch from pip-compile to uv with PEP 735 dependency groups; commit uv.lock
• Retain pylint/isort/pycodestyle as on master
• Coverage config moved into pyproject.toml
• Update CI to use astral-sh/setup-uv; SHA-pin all actions
• Add python-semantic-release + release.yml (OIDC trusted publishing)
• Add docs job to CI matrix (builds Sphinx docs as part of every run)
• Wire up Codecov integration (coverage reports now uploaded on each CI run; CODECOV_TOKEN is already configured org-wide)

Removed/Updated

Deleted files: setup.py, setup.cfg, requirements/, .coveragerc, CHANGELOG.rst, mypy.ini, pypi-publish.yml

Removed Makefile targets:

Target	Reason
compile-requirements	Replaced by uv lock
piptools	pip-compile toolchain removed; uv handles environment setup

Updated Makefile targets:

Target	Change
requirements	Now runs uv sync --group dev + uv tool install tox --with tox-uv
upgrade	Now runs edx_lint write_uv_constraints + uv lock --upgrade

Notes on test-format removal

test-format (which runs black --check) was intentionally removed from test-quality. Investigation revealed that on master, black is absent from requirements/quality.txt — the file the quality tox env installs from — so black --check was never actually executable in CI. The removal here maintains parity with what master truly enforces rather than introducing a new failure.

This is tracked in issue #284. The follow-up plan is to replace black with ruff format (the standard formatter for modernized Open edX repos), wire it into the quality dependency group correctly, and restore test-format to the quality gate.

Versioning

setuptools-scm with dynamic = ["version"] — master had a PyPI publish workflow; python-semantic-release controls the version string at release time via git tags.

Important Notes

• pypi-publish.yml was deleted because release.yml replaces it with OIDC trusted publishing via python-semantic-release; the old workflow used a PYPI_UPLOAD_TOKEN secret which is superseded by the OIDC approach.
• PyPI trusted publisher (OIDC) must be configured on PyPI before merging — without it the publish_to_pypi job will fail on the first release.

Testing Notes

This PR has not been manually tested against the repo's own features. Testing relied on CI checks and local agent tooling (make requirements, make lint, make test, python -m build). Repo-owner is encouraged to run the repo's feature tests before merging.

---

🤖 Generated with Claude Code

[farhan]

Review comments on the modernization PR — see individual inline threads for details.

[farhan]

Addressing inline comments — see individual threads for analysis and fixes.

[irfanuddinahmad]

CI passes and the migration is otherwise careful (src-layout, entry points, coverage config, and the release workflow all line up correctly, and I confirmed the docs/package/e2e jobs are genuinely green on the PR). But there's a real, undisclosed regression in the quality gate:

Makefile: test-quality used to run test-lint test-codestyle test-mypy test-format. This PR drops test-format from that list:

-test-quality: test-lint test-codestyle test-mypy test-format ## run static coverage tests
+test-quality: test-lint test-codestyle test-mypy ## run static coverage tests

test-quality is exactly what CI's quality tox env calls (make test-quality selfcheck), so black --check is no longer part of the CI quality gate — a formatting regression could now merge silently. This isn't mentioned anywhere in the PR description. Was this intentional, or a slip during the Makefile restructuring? If intentional, please call it out explicitly; if not, please restore it.

Minor, non-blocking notes:

• The PR body doesn't mention that docs was added to the CI matrix or that Codecov integration is new — both are fine (I verified CODECOV_TOKEN already works org-wide, e.g. on the merged XBlock PR), but worth stating in the description since they're scope beyond "modernize tooling."

[farhan]

Thanks for catching this, @irfanuddinahmad — your point is valid.

The removal of test-format from test-quality was intentional, but I should have called it out explicitly in the PR description (now updated). Here's the context:

On master, black is present in requirements/ci.txt and requirements/dev.txt but not in requirements/quality.txt — which is what the quality tox env installs from. So black --check could never actually execute in CI's quality gate; it has silently been unenforced. Removing test-format from test-quality in this PR maintains parity with what master actually enforces in CI, rather than introducing a new failure.

I've opened issue #284 to track this properly. The plan is to replace black with ruff format (the standard formatter for modernized Open edX repos, as established in openedx-platform#38076), wire it up correctly in the quality dependency group, and restore test-format to the quality gate — all as a follow-up after this PR merges.

[irfanuddinahmad]

CI is fully green (11/11) and the migration is careful — src-layout, entry points, coverage config, and the semantic-release/OIDC publish workflow all line up correctly. The test-format removal from test-quality is confirmed non-regressive (black was never enforced on master either, since it was missing from requirements/quality.txt) and is tracked with a concrete follow-up plan in #284. LGTM.