Repository navigation
ci: align command contracts and close real-host and lane coverage gaps #811
Description
Context
Verified on main: make docs is weaker than the CI docs gate (no --strict), every target re-runs pip install --upgrade pip and pip install hatch, make check type-checks twice, and make doctor cannot fail for a missing Hatch environment or func. The real-host and lane-coverage gaps are tracked as separate bug issues (host-smoke routes, browser lane, optional-dependency skips).
Acceptance Checklist
- One documented command per purpose:
make install(no per-run pip/Hatch upgrade),make install-ci PYTHON=<interpreter>(installs into exactly that interpreter),make doctor(read-only; non-zero when Python >= 3.11 or Hatch is missing; reportsfunc/azurite),make format/make format-check,make check(lint, format, mypy once, Bandit scope, workflow lints),make test(offline, 95% floor, no e2e/perf/browser/host),make integration(real Functions host + Azurite on free ports; non-zero without prerequisites; no leftover processes),make verify(check + test + integration + strict docs + wheel smoke),make e2e-azure(never silently skips),make clean(generated artifacts only).make check-allkeeps its offline meaning or is documented as an alias. -
make docsequals the CI docs gate (mkdocs build --strict). - Failed CI runs retain JUnit and coverage XML; coverage is summarized in the Job Summary; no secret-bearing file is uploaded.
- Matrix cells run on their own interpreter and enforce coverage >= 95% on 3.11 x latest.
- No existing gate is removed, renamed, made advisory or weakened (required contexts, release chain, wheel, minimum-dependency, 2.x, host-smoke, docs and perf lanes).
Out of scope
A custom evidence/summary schema, generic secret redaction, new orchestration modules under tools/, Azure infrastructure changes, a report command.
References
Real-host routes, browser lane and optional-dependency skips have their own issues.
Activity
[-]chore(dx): standardize OpenAPI development, testing, and verification evidence[/-][+]ci: standardize OpenAPI development, testing, and verification evidence[/+][-]ci: standardize OpenAPI development, testing, and verification evidence[/-][+]ci: align command contracts and close real-host and lane coverage gaps[/+]yeongseon commentedon Oct 24, 2026
Rewrote the body to outcome-focused criteria. The original prescribed a
summary.jsonschema with PASS/FAIL/INCOMPLETE/NOT_APPLICABLE, an evidence tree,report, and thin-Makefile-plus-tested-scripts orchestration. Auditingmainfound the real defects (listed in Context) and three separate bugs, now tracked in their own issues; JUnit/coverage artifacts andcoverage report --format=markdowncover the reporting need. Previous text is in the edit history. PR #813 (about 1,700 added lines) will be replaced by small PRs.build(make): build docs with --strict and the project's docs extra
build(make): make doctor a read-only prerequisite check that can fail
docs(development): document install-ci, doctor, strict docs and the s…
test(make): compact the Makefile contract tests
build(make): align make docs with CI, stop re-bootstrapping, run mypy…
yeongseon commentedon Oct 26, 2026
Closed with one acceptance item not met on
main(verified at 65ad8c4): the checklist saysmake checkcovers "lint, format, mypy once, Bandit scope, workflow lints", butcheckis still justlint; Bandit and the workflow lints run only inmake check-all.tests/test_make_contract.py(from #831) pinscheckto run no tests andcheck-allto run each gate once, so moving them intocheckmeans changing that contract.Everything else on the list is on
main:integration,wheel-smokeandverify(#844), failed-run JUnit/coverage retention and the coverage Job Summary (#843), and the earlierinstall/install-ci/doctor/strictdocswork (#831). If keeping Bandit and workflow lints incheck-allis intended, the checklist item could be reworded instead. Reopening or a follow-up issue is the maintainer's call.yeongseon commentedon Oct 26, 2026
The remaining static-check contract is now addressed by merged PR #851: make check includes lint (format and mypy once), the existing expanded Bandit scope, and all workflow lints. make check-all retains every offline gate exactly once. Contributor/development docs and contract tests were updated deliberately. Local make check-all: 1651 passed, 30 skipped, 22 deselected, 5 warnings in 27.25s; coverage 95.25%; strict docs and build passed. All five required contexts passed on PR head 24e3c25. This follow-up does not claim a new certification of every historical checklist item.