chore(deps): update openai requirement from <3,>=2.52.0 to >=3.6.0,<4 #7298

Workflow file for this run

.github/workflows/backend.yml at d888b91

	name: Backend CI

	# Enable Buildkit and let compose use it to speed up image building
	env:
	DOCKER_BUILDKIT: 1
	COMPOSE_DOCKER_CLI_BUILD: 1
	SKLEARN_ALLOW_DEPRECATED_SKLEARN_PACKAGE_INSTALL: True

	defaults:
	run:
	working-directory: ./

	on:
	# NOTE: no `paths-ignore` on pull_request, deliberately. The `gate` job at
	# the bottom of this file is the one meant to be REQUIRED by `main`'s branch
	# protection, and a workflow skipped by path filtering never reports its
	# checks at all — the required check would hang Pending and the PR would be
	# permanently unmergeable. So the workflow always starts on a PR; the `changes` path
	# filter below still keeps the expensive jobs from running, and a
	# docs-only PR costs two ubuntu-latest jobs of a few seconds each.
	# `push` keeps its filter: branch protection does not gate pushes.
	pull_request:
	branches: [ "master", "main", "v*" ]

	push:
	branches: [ "master", "main", "v*" ]
	paths-ignore: [ "docs/**" ]

	concurrency:
	group: ${{ github.head_ref \|\| github.run_id }}
	cancel-in-progress: true

	jobs:
	changes:
	# Path filtering only gates pull_request events; pushes to protected
	# branches always run downstream regardless of which files changed.
	# dorny/paths-filter@v4 additionally needs a checked-out repo on push
	# events (it shells out to `git branch --show-current`), so scoping
	# this job to pull_request avoids the "fatal: not a git repository"
	# red X that would otherwise appear on every push to main.
	if: github.event_name == 'pull_request'
	runs-on: ubuntu-latest
	# Fail open: if this job errors (e.g., transient GitHub API failure),
	# downstream jobs should still run rather than being silently skipped.
	continue-on-error: true
	permissions:
	contents: read
	pull-requests: read
	outputs:
	backend: ${{ steps.filter.outputs.backend }}
	steps:
	# No checkout needed: dorny/paths-filter@v4 reads diffs from the
	# pull_request event payload.
	- uses: dorny/paths-filter@v4
	id: filter
	with:
	filters: \|
	backend:
	- 'opencontractserver/**'
	- 'config/**'
	- '*.py'
	- 'requirements/**'
	- 'test.yml'
	- 'compose/**'
	- 'Dockerfile*'
	- '.pre-commit-config.yaml'
	- 'setup.cfg'
	- 'mypy.ini'
	- 'pyproject.toml'
	- '.github/workflows/backend.yml'
	# The gate's own scripts. A change to the thing that decides
	# whether a merge is allowed should be validated by the full
	# suite, not waved through as "not backend code".
	- '.github/scripts/**'

	linter:
	needs: changes
	# Run on every push, and on PRs unless the filter explicitly reports
	# no backend changes ('false'). `always()` is required because the
	# `changes` job is skipped on push events (see its `if:` above), and
	# a skipped `needs` target would otherwise cascade-skip this job.
	# The `!= 'false'` form also preserves the fail-open behaviour on
	# PRs where `changes` errors transiently (outputs.backend is '').
	if: always() && (github.event_name == 'push' \|\| needs.changes.outputs.backend != 'false')
	runs-on: ubuntu-latest
	permissions:
	contents: read
	steps:
	- name: Checkout Code Repository
	uses: actions/checkout@v7

	- name: Set up Python
	uses: actions/setup-python@v7
	with:
	python-version: "3.12"
	cache: pip
	cache-dependency-path: \|
	requirements/base.txt
	requirements/local.txt

	- name: Install dependencies
	run: pip install -r requirements/local.txt

	- name: Run pre-commit
	run: pre-commit run --all-files

	- name: Run mypy
	# Runs from requirements/local.txt (not via pre-commit) so CI is
	# authoritative even when contributors skip the hook. Keep the
	# mypy pin in requirements/local.txt in sync with the `rev:` of
	# the mirrors-mypy hook in .pre-commit-config.yaml — a drift
	# means the hook and CI can disagree silently. See
	# docs/typing/README.md.
	run: python -m mypy --config-file mypy.ini opencontractserver config

	pytest:
	needs: [changes, linter]
	# Same fail-open semantics as the linter gate above.
	if: always() && needs.linter.result == 'success' && (github.event_name == 'push' \|\| needs.changes.outputs.backend != 'false')
	runs-on: yuge
	timeout-minutes: 180
	permissions:
	contents: read
	actions: read
	# ``packages: write`` lets the pytest job push the buildx layer cache to
	# ``ghcr.io/{org}/{repo}/django-local-cache:main`` on ``main`` pushes (the
	# ``--cache-to`` step below is gated to that event). PRs only read the
	# cache — the ``write`` scope is harmless for them since their token is
	# still PR-scoped (no cross-fork push).
	packages: write

	steps:
	- name: Checkout Code Repository
	uses: actions/checkout@v7

	- name: Store Codecov Env Flags
	continue-on-error: true
	run: \|
	ci_env=`bash <(curl -s https://codecov.io/env)`
	echo "$ci_env"

	- name: Set up Docker Buildx
	uses: docker/setup-buildx-action@v4

	# Use the ghcr.io registry as the buildx layer-cache backend instead of
	# the GitHub Actions cache (``type=gha``). The registry cache is portable:
	# the same layers CI writes are pullable from a local dev machine and any
	# prod-deploy host (see ``local.yml`` / ``production.yml`` ``cache_from``).
	# GHA cache only works inside GitHub Actions, so it would not help
	# contributors building from a fresh checkout.
	#
	# The cache image inherits the repo's visibility (public for this repo,
	# so no auth is required to pull on first publish — confirm in
	# GitHub Packages settings if the package was created private).
	- name: Login to ghcr.io
	uses: docker/login-action@v4.6.0
	with:
	registry: ghcr.io
	username: ${{ github.actor }}
	password: ${{ secrets.GITHUB_TOKEN }}

	- name: Build the Stack
	# The django image rebuild was the biggest pre-test cost on every PR
	# (~2:40 wall) because the runner ships from scratch every time. The
	# buildx layer cache restores most layers (apt-get, pip wheel, spaCy
	# model fetch) in seconds when their inputs (Dockerfile, requirements,
	# COPY'd source) are unchanged.
	#
	# ``cache-from`` is always set so PRs benefit from main's cache.
	# ``cache-to`` only writes on ``push`` events (i.e. main-branch merges)
	# — PR runs from forks can't push to ghcr packages anyway, and pushing
	# from PRs would pollute main's cache key.
	#
	# The image is tagged to match the ``image:`` directive on the
	# ``django`` service in test.yml, so the subsequent ``docker compose
	# build`` recognises it as already built and only compiles the small
	# ``postgres`` image (~5 s).
	env:
	CACHE_REF: ghcr.io/${{ github.repository_owner }}/${{ github.event.repository.name }}/django-local-cache:main
	run: \|
	# Lowercase the cache ref — ghcr requires lowercase paths but the
	# repo name on GitHub can be mixed-case (e.g., "OpenContracts").
	CACHE_REF_LC="$(echo "$CACHE_REF" \| tr '[:upper:]' '[:lower:]')"
	CACHE_TO_FLAG=""
	if [ "${{ github.event_name }}" = "push" ]; then
	CACHE_TO_FLAG="--cache-to=type=registry,ref=${CACHE_REF_LC},mode=max"
	fi
	docker buildx build \
	--load \
	--cache-from="type=registry,ref=${CACHE_REF_LC}" \
	${CACHE_TO_FLAG} \
	--tag opencontractserver_local_django \
	--file compose/local/django/Dockerfile \
	--build-arg GITHUB_ACTIONS=true \
	.
	docker compose -f test.yml build

	# Note: a ``manage.py migrate`` step previously ran here against the
	# ``opencontractserver`` default database. That database is never read
	# by the test suite — pytest-django creates per-worker test databases
	# (``test_opencontractserver_gw0``, ``_gw1``, …) on first invocation
	# and runs migrations against those. ``DATABASES['default']['TEST']``
	# does not declare a ``TEMPLATE``, so the default DB is not used as a
	# createdb template either. The step was pure waste (~3 min wall) and
	# has been removed; ``pytest --reuse-db`` (in pytest.ini) handles cold
	# creation and warm reuse of the test DBs by itself.

	# ``collectstatic`` previously ran here (~1:40 wall on a cold container
	# start). The test suite does not serve static files — ``config/settings/
	# test.py`` deliberately uses ``StaticFilesStorage`` (the manifest-less
	# variant) so admin tests can resolve ``{% static %}`` URLs at template
	# render time without needing the post-collectstatic manifest file. The
	# step produced no test-relevant artefacts; the wall time was dominated
	# by the django container cold start, not by the actual collection work.

	- name: Verify Docker Containers
	run: \|
	docker compose -f test.yml ps

	- name: Capture Docker Compose Logs
	if: failure()
	run: \|
	docker compose -f test.yml logs --no-color > docker-compose-logs.txt

	- name: Upload Docker Compose Logs
	if: failure()
	uses: actions/upload-artifact@v7
	with:
	name: docker-compose-logs
	path: docker-compose-logs.txt

	- name: Check Container Health
	run: \|
	echo "=== Docker containers status ==="
	docker compose -f test.yml ps
	echo "=== Container logs (last 20 lines each) ==="
	for container in $(docker compose -f test.yml ps -q); do
	name=$(docker inspect -f '{{.Name}}' $container \| sed 's/^\///')
	echo "--- Logs for $name ---"
	docker logs --tail 20 $container 2>&1 \|\| true
	done
	echo "=== Memory usage ==="
	docker stats --no-stream
	free -h

	- name: Run Backend Test Suite
	timeout-minutes: 100
	run: \|
	# Coverage instrumentation runs on both push and pull_request events
	# so Codecov's PR patch status has a fresh report for the PR head
	# commit. The Django image is built on Python 3.12 with
	# ``COVERAGE_CORE=sysmon`` (set in .envs/.test/.django), so coverage
	# uses ``sys.monitoring`` rather than the legacy C trace function —
	# an order-of-magnitude cheaper instrumentation path on Python 3.12+.
	#
	# ``--dist loadscope`` is retained: a worksteal trial (PR #1767,
	# initial commit) exposed ~39 test-order-dependency failures across
	# plain ``TestCase`` subclasses that worked under loadscope's
	# class-pinning. The underlying isolation bugs are tracked in
	# docs/refactor_plans/2026-05-23-test-suite-speedup-A1-B1.md §6
	# and need to be resolved before worksteal can be re-attempted.
	#
	# pytest-cov + pytest-xdist handle per-worker coverage merging.
	#
	# ``--timeout=600 --timeout-method=signal``: a per-test hang guard.
	# Without it a single test that blocks on a starved service (the
	# 2-core runner runs ``-n auto`` workers alongside the docling/embedder
	# ML containers) hangs its worker until the 100-min step ceiling, with
	# no traceback. 600s is ~10x the slowest legitimate test (~57s) so it
	# never false-positives, while turning a silent hang into a failure
	# that identifies the offending test. On Linux xdist workers the
	# signal method raises a pytest failure in the worker's main thread.
	# The thread method hard-exits the worker, which can leave the xdist
	# controller waiting until the outer 100-minute timeout.
	docker compose -f test.yml run django pytest --cov --cov-report=xml -n auto --dist loadscope --timeout=600 --timeout-method=signal

	- name: Verify Coverage File Exists
	run: \|
	# Verify coverage.xml exists in the working directory
	ls -la coverage.xml

	- name: Upload Coverage Reports to Codecov
	uses: codecov/codecov-action@v7
	with:
	token: ${{ secrets.CODECOV_TOKEN }}
	files: ./coverage.xml
	flags: backend
	name: backend-coverage
	fail_ci_if_error: false

	- name: Tear down the Stack
	run: docker compose -f test.yml down

	gate:
	# The job to require on `main` — this one, not `pytest`. See
	# .github/scripts/backend_ci_gate.sh for why requiring `pytest` directly
	# leaves a red linter mergeable (a job skipped by its own `if:` reports
	# SUCCESS to branch protection, and a red linter skips pytest).
	#
	# Required on `main` as of 2026-08-20 (PR #2267). Note `enforce_admins` is
	# false, so an admin can still bypass it. Re-apply or move it with
	# .github/scripts/require_backend_ci_gate.sh --apply, and confirm the
	# current state with
	# `gh api repos/{owner}/{repo}/branches/main/protection`
	# rather than assuming this comment is still accurate.
	#
	# `if: always()` is what makes this requirable: it reports on every PR,
	# including ones where every job above it was skipped.
	name: backend-ci-gate
	needs: [changes, linter, pytest]
	if: always()
	runs-on: ubuntu-latest
	permissions:
	contents: read
	steps:
	- name: Checkout Code Repository
	uses: actions/checkout@v7

	- name: Self-test the gate logic
	# Runs first, and on every invocation: a gate whose own decision
	# table has silently inverted is worse than no gate. Costs ~0.1s.
	run: bash .github/scripts/backend_ci_gate.sh --self-test

	- name: Evaluate Backend CI results
	run: \|
	bash .github/scripts/backend_ci_gate.sh \
	"${{ github.event_name }}" \
	"${{ needs.changes.outputs.backend }}" \
	"${{ needs.linter.result }}" \
	"${{ needs.pytest.result }}"

Navigation Menu

Uh oh!

Uh oh!

chore(deps): update openai requirement from <3,>=2.52.0 to >=3.6.0,<4 #7298

Workflow file

chore(deps): update openai requirement from <3,>=2.52.0 to >=3.6.0,<4 #7298

Uh oh!

Workflow file for this run