Testing

The test suite has two tiers. The lightweight checks stay fast, and the heavyweight HiveMind-side round-trips do not pull the full alpha stack into every CI matrix cell.

Tier	Path	Extra	What it covers
Smoke	tests/test_smoke.py	[test]	Packaging, version, the VoiceClient surface, console-script import. Network-free, no hivescope.
E2E	tests/e2e/	[e2e]	Real hivemind-core hub + real satellite client over a real HiveMessageBusClient, hardware mocked.

Running locally

Use a clean virtualenv (the e2e extra pulls the full 2.x alpha stack):

python3 -m venv .venv && . .venv/bin/activate

# smoke only (fast)
pip install -e ".[test]"
pytest tests/test_smoke.py

# full end-to-end suite
pip install -e ".[e2e]"
pytest tests/e2e/

No --pre is needed: the prerelease dependencies are pinned as minimum versions in pyproject.toml, so pip resolves the 2.x stack on its own.

How the e2e tests work

tests/e2e/ boots a real hivemind-core hub in-process using hivescope's loopback WebSocket harness (TopologyBuilder().add_master(use_loopback=True)). It registers a satellite key and connects the satellite over a real HiveMessageBusClient. The HiveMind protocol, encryption handshake, ACL/policy chain, and hub dispatch are all genuine production code running over a localhost WebSocket.

Only the audio hardware is mocked, and only at the seams that ovos-dinkum-listener and ovos-audio already expose as constructor arguments:

• Microphone: a FakeMicrophone that returns silent fixed-size chunks.
• STT: a FakeSTT stub. Tests inject utterances on the bus directly.
• VAD: a FakeVAD that reports silence.
• TTS: a CapturingTTS that records synthesis requests instead of writing to a sound device. Injecting a TTS sets PlaybackService.disable_reload, so the real audio service never loads a real TTS plugin or opens an audio device.

The tests carry no importorskip or skipif guards on missing dependencies. The [e2e] extra is a hard requirement, so the tests always run.

What the e2e suite asserts

• test_satellite_hivemind_e2e.py covers the satellite-side round-trips:
  • an utterance on the satellite's real bus reaches the real hub agent bus with a stamped source (satellite to hub).
  • the real VoiceClient constructs and binds to the HiveMind bus with mocked hardware, and its _connect_to_bus no-op leaves the supplied bus in place.
  • a hub speak routed to the satellite peer drives the real PlaybackService and its injected fake TTS (hub to satellite).
• test_bridge1_conformance.py covers OVOS-BRIDGE-1 / SESSION-1 conformance (envelope, source/destination, session fidelity, FIFO ordering).
• test_acl_policy.py covers policy admission paths (allowed-types denial, skill-blacklist injection, reserved-session-id gate).

CI

Workflow	Job
build_tests.yml	smoke (tests/test_smoke.py, [test]) across Python 3.10–3.13
e2e_tests.yml	e2e (tests/e2e/, [e2e]) on Python 3.11
coverage.yml	full suite under coverage ([e2e])
lint.yml, pip_audit.yml, license_tests.yml, repo-health.yml	static / supply-chain / health gates

All jobs use the shared OpenVoiceOS/gh-automations reusable workflows at @dev.

---

← Troubleshooting · Home