Skip to content

Testing

SimView has three independent test suites: Python (pytest), JS unit tests (vitest), and Playwright end-to-end tests.

Python tests (tests/)

uv run pytest -q                                    # full suite
uv run pytest tests/test_scene.py -q                # one file
uv run pytest tests/test_scene.py::test_name -q     # one test
uv run pytest --cov=simview --cov-report=term-missing --cov-fail-under=83 -q

Tests requiring torch use pytest.importorskip("torch") before torch-dependent imports (see tests/conftest.py's HAS_TORCH flag), which pushes those imports below a statement — hence the tests/*.py ruff E402 ignore in pyproject.toml. Follow the same pattern for new torch-dependent test modules so they skip cleanly on a base install (no authoring extra).

tests/conftest.py's build_scene() is the shared fixture builder for a small representative scene (shared terrain + one box, a few states with an "energy" scalar) — prefer it over hand-rolling scenes in new tests.

Test files are grouped roughly by area:

  • Core authoring: test_scene.py, test_trajectory.py, test_roundtrip.py, test_contacts.py, test_episodes.py, test_pointcloud_features.py, test_columnar_states.py, test_columnar_ondisk.py
  • CLI-backed queries: test_terrain.py, test_terrain_query.py, test_diff.py, test_info.py, test_cli.py
  • Server/viewer lifecycle: test_server.py, test_live.py, test_launcher.py, test_show.py
  • Multi-file workflows: test_merge.py, test_batch_selection.py
  • Shared helpers: test_utils.py, test_remote.py

JS unit tests (tests/js/)

npm test                                             # vitest run
npx vitest run tests/js/blobCodec.test.js            # one file

Target pure logic in utils/ and a few components//objects//ui/ classes — no DOM/browser needed, run under Node. One *.test.js per module covered: blobCodec, blobWindow, bodyTransforms, interpolate, errorMath, csv, viewState, liveFollow, episodes, terrainSample, batchPresets, batchVisibility, cameraRange, similarity, objectsUtils, AnimationController, StateStore, BatchManager, WindowedField, InteractionController, Body, Terrain, TerrainFeatures.

End-to-end tests (tests/e2e/)

uv run python example.py --no-launch   # writes example_sim.json first
npx playwright test                     # auto-starts the server
npx playwright test tests/e2e/smoke.spec.js   # one spec

tests/e2e/smoke.spec.js is a tripwire for wiring-level regressions (bad imports, server 500s), not a substitute for unit coverage — see the comment in playwright.config.js. Alongside it, batchVisibility, episodes, scalarChart and windowedFields cover behavior that only shows up in a real browser (per-batch objects built lazily, episode navigation, a chart actually painting its data, range requests for a long trajectory's fields). playwright.config.js starts two servers for these: one on example_sim.json, one on the episodic scene make_episodic_scene.py generates.

CI

.github/workflows/ci.yml runs four independent jobs on every push/PR:

  1. Python (matrix 3.12/3.13/3.14) — lint (ruff check), format check (ruff format --check), type check (pyright), and pytest with coverage, installed via uv sync --extra authoring.
  2. JS unit tests — npm test.
  3. Playwright e2e — generates example_sim.json, then runs the e2e specs.
  4. Base-install-only — uv sync (no authoring extra), confirms import simview and the test suite still work without torch/numpy. A change that only works with torch installed will pass every other job but fail this one.