v2.3.3 — bump version, rebuild all PDFs at 2.3.3 #330
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| name: cross-platform-smoke | |
| # Asserts that ELI imports, the FastAPI server starts headless, and the OS-agnostic code | |
| # paths load + pass their tests on Linux, macOS, AND Windows. Deliberately LIGHTWEIGHT — | |
| # it installs the core + server (CPU), not the heavy GPU/image stack, so it stays fast. | |
| # | |
| # This ran as workflow_dispatch ONLY, on the reasoning that it "avoids metered Actions on | |
| # every push". That reasoning does not apply to this repository: ShadowESC95/ELI_v2.0 is | |
| # PUBLIC, and GitHub Actions on standard runners is free and unmetered for public repos. | |
| # The cost was therefore zero and the gate was off anyway — nothing verified a merge on | |
| # Windows or macOS unless somebody remembered to click the button. | |
| # | |
| # Scope note, so the claim stays honest: this is a curated cross-platform gate, not the | |
| # full 8.6k-test suite. Most of that suite needs a GGUF model, a display, or generated | |
| # artifacts, none of which exist on a runner. What is listed below is what can be | |
| # meaningfully asserted on all three OSes. | |
| # | |
| # Every pytest step below MUST be `python -m pytest`, never the bare `pytest` console | |
| # script. `-m` puts the working directory on sys.path; the console script does not. The | |
| # top-level `api` package is not installed by `pip install -e .`, so under bare `pytest` | |
| # `import api.server` fails, tests/test_api_server.py hits its own module-level skip | |
| # guard, and the step reports "collected 0 items / 1 skipped" — pytest exit code 5. That | |
| # is what this workflow's first ever run did on all six jobs. The identical mistake once | |
| # aborted a release from scripts/build_packages.sh; it is easy to reintroduce and it | |
| # fails in a way that reads like an environment problem rather than a command error. | |
| on: | |
| workflow_dispatch: | |
| pull_request: | |
| paths-ignore: ['**/*.md', '**/*.pdf', 'blueprints/**', 'docs/**'] | |
| push: | |
| branches: [main] | |
| paths-ignore: ['**/*.md', '**/*.pdf', 'blueprints/**', 'docs/**'] | |
| concurrency: | |
| group: smoke-${{ github.ref }} | |
| cancel-in-progress: true | |
| jobs: | |
| smoke: | |
| name: ${{ matrix.os }} · py${{ matrix.python }} | |
| strategy: | |
| fail-fast: false | |
| matrix: | |
| os: [ubuntu-latest, macos-latest, windows-latest] | |
| python: ["3.10", "3.12"] | |
| runs-on: ${{ matrix.os }} | |
| # Windows runners default to PowerShell, which cannot parse the `python - <<'PY'` | |
| # heredoc the dashboard step is built on — it failed with a .ps1 ParserError on both | |
| # Windows jobs the first time this workflow ever ran. Pinning bash everywhere also | |
| # means a step that passes on Linux is running the identical command on Windows, | |
| # which is the entire point of a cross-platform gate. | |
| defaults: | |
| run: | |
| shell: bash | |
| steps: | |
| - uses: actions/checkout@v4 | |
| - uses: actions/setup-python@v5 | |
| with: | |
| python-version: ${{ matrix.python }} | |
| - name: Install (core + server, CPU — no GPU/image stack) | |
| run: | | |
| python -m pip install --upgrade pip | |
| pip install tomli | |
| python tools/validate_pyproject_scripts.py | |
| pip install -e ".[server]" | |
| pip install llama-cpp-python --prefer-binary | |
| pip install httpx pytest | |
| - name: Import smoke — cross-platform modules load on this OS | |
| run: | | |
| python -c "import eli; from eli.utils.platform_compat import normalize_platform; print('platform:', normalize_platform())" | |
| python -c "from eli.execution.executor_enhanced import open_browser; import eli.perception.os_controller; print('executor + os_controller OK')" | |
| python -c "from api.server import app; print('server app OK')" | |
| - name: Headless server smoke — /, /health, /v1/chat shape (no model needed) | |
| run: | | |
| python -c "from fastapi.testclient import TestClient; from api.server import app; c=TestClient(app); assert c.get('/health').status_code==200; assert c.get('/').status_code==200; print('server endpoints reachable (health + web UI)')" | |
| - name: Dashboard read-endpoint smoke — every widget's data source answers (no model) | |
| # The web dashboard is only as alive as the read endpoints it fetches. A blank | |
| # cognition mesh / broken home-mesh shipped this way once (backend fine, endpoint | |
| # not exercised in CI). Hit every OV_SRC source as the operator (tokenless | |
| # loopback) and assert each returns 200 — this is the class of bug the trivial | |
| # /health check above can't see. | |
| env: | |
| ELI_API_ALLOW_TOKENLESS: "1" | |
| run: | | |
| python - <<'PY' | |
| from fastapi.testclient import TestClient | |
| from api.server import app | |
| c = TestClient(app) | |
| eps = [ | |
| "/v1/cognition/orchestration", "/v1/home/mesh/status", "/v1/home/mesh/ping", | |
| "/v1/system", "/v1/tasks", "/v1/autonomy/goals", "/v1/net", "/v1/net/egress", | |
| "/v1/capabilities", "/v1/devices", "/v1/home/scenes", "/v1/home/sun", | |
| "/v1/research/corpora", "/v1/audit", "/v1/connect", "/v1/media", | |
| ] | |
| bad = [] | |
| for e in eps: | |
| try: | |
| r = c.get(e) | |
| if r.status_code != 200: | |
| bad.append(f"{e} -> HTTP {r.status_code}") | |
| except Exception as ex: | |
| bad.append(f"{e} -> raised {type(ex).__name__}: {ex}") | |
| if bad: | |
| raise SystemExit("dashboard endpoints FAILED:\n " + "\n ".join(bad)) | |
| # the mesh specifically must carry real agent data, not an empty shell | |
| o = c.get("/v1/cognition/orchestration").json() | |
| assert o.get("ok") and o.get("agents"), f"cognition mesh returned no agents: {o}" | |
| print(f"all {len(eps)} dashboard read-endpoints OK; mesh has {len(o['agents'])} agents") | |
| PY | |
| - name: Real server-endpoint suite (mesh, home-mesh, config-merge, RBAC — the full API smoke) | |
| # test_api_server.py drives the real FastAPI app end-to-end (78 cases). It needs | |
| # --noconftest because the repo conftest mocks pydantic/PySide6 for the fast unit | |
| # suite, which makes the server un-importable. This is where the home-mesh config | |
| # merge fix and the orchestration/mesh contracts are actually enforced. | |
| run: python -m pytest -q -p no:cacheprovider --noconftest tests/test_api_server.py | |
| - name: First-run DB autobuild (cross-platform paths) | |
| run: python -m eli.core.init_data | |
| - name: Bootstrap claims artifacts (manifest + blueprints — fresh-clone path) | |
| run: python tools/bootstrap_claims_artifacts.py | |
| - name: Cross-platform unit tests (explicit self-contained files — no GGUF / display / generated artifacts) | |
| run: > | |
| python -m pytest -q -p no:cacheprovider | |
| tests/test_dual_gpu_settings.py | |
| tests/test_user_model.py | |
| tests/test_08_hardware.py | |
| tests/test_smart_fit.py | |
| tests/test_settings_extended.py | |
| tests/test_shell_security_gate.py | |
| tests/test_pyproject_packaging.py | |
| tests/test_device_names.py | |
| tests/test_install_script.py | |
| tests/test_document_formats.py | |
| - name: Claims structural smoke (manifest ↔ executor) | |
| run: > | |
| python -m pytest -q -p no:cacheprovider | |
| tests/claims/test_no_silent_swallow.py | |
| tests/claims/test_structural_claims.py | |
| tests/claims/test_readme_counts.py | |
| tests/claims/test_capability_manifest.py::test_manifest_total_matches_capability_count |