fix: resolve unbounded get_response() hang on FakeBus

[JarbasAl]

🤖 Auto-generated by Claude Fable 5 (claude-fable-5) via Claude Code — NOT human-reviewed. Verify before acting.

Revised after adversarial review — see "Concurrency defect found in review" below.

Field evidence

OpenVoiceOS/ovos-skill-alerts#138 ("Update 3") root-caused, via py-spy stack
captures on a real CI hang, two mechanisms by which real skill handlers
block forever under ovoscope's synchronous FakeBus:

• (a) wait_while_speaking() (ovos_bus_client/session.py:710) burns
  its full 15s timeout for nested/cascading speak_dialog(..., wait=True)
  calls if recognizer_loop:audio_output_end never arrives. Already
  mitigated in ovoscope/__init__.py's _mock_tts mock-TTS handler
  (per-call threading.Timer unduck) — verified still correct for nested
  calls (see regression test below).
• (b) OVOSSkill._wait_response() (ovos_workshop/skills/ovos.py
  ~1802-1809, ovos-workshop==7.0.6) has no deadline at all:
  while not ans: time.sleep(0.1). With the get_response()/ask_yesno()
  default num_retries=-1, the background _real_wait_response thread
  re-prompts forever if nothing ever answers it. A real voice satellite
  eventually emits mycroft.skills.abort_question on user silence;
  ovoscope's FakeBus never did. This is what leaked non-answered-thread
  state into ovos-skill-alerts' shared-MiniCroft e2e suite and made CI
  hang to the 30-minute job kill on 4 of 5 Python versions on PR fix: Han audit round 2 — SkillApi retention, accuracy gate, false-green assertions, race windows #123.

Fix

MiniCroft arms a short watchdog threading.Timer on
skill.converse.get_response.enable that emits the existing
mycroft.skills.abort_question bus message (no new message type — this is
the same message a real listener/GUI sends on cancel/silence, and
OVOSSkill._real_wait_response is already decorated with
@killable_event("mycroft.skills.abort_question", ...) to handle it) if
.disable doesn't fire first. .disable (emitted once get_response()
actually returns, answered or not) disarms the watchdog. Tracked/cancelled
in stop() the same way the existing mock-TTS timers already are, so no
orphaned timer can fire onto a closed bus and poison a later test.

Concurrency defect found in review

The first version of this fix kept ONE flat list of pending watchdog
timers. Adversarial review reproduced a real regression: with two
concurrent get_response() flows in flight (skill A and skill B, a
fleet-style MiniCroft running multiple skills), skill A's .disable
drained and cancelled the entire list, including skill B's still-armed
watchdog — if B's question was never answered, B hung forever again,
defeating the fix for exactly the multi-skill case it needs to hold up
under. Repro that pinned it: enable(A), enable(B), disable(A) → 0
timers left instead of 1.

Fixed by scoping each watchdog by (skill_id, session_id) — the same
two-part scope ovos_workshop's own
@killable_event("mycroft.skills.abort_question", check_skill_id=True)
already uses to decide which stalled thread an abort is actually for.
.disable now cancels only its own (skill_id, session_id) entry; the
abort emit carries the same skill_id/session context so only the
stalled question dies. stop() still drains and cancels every remaining
entry on teardown.

Smallest correct change: ovoscope/__init__.py only (MiniCroft),
mirroring the existing _mock_tts timer-tracking pattern already in the
file.

Red → green proof

test/unittests/test_tts_lifecycle_nested_and_get_response.py:

• TestNestedSpeakDialogWait / TestUnansweredGetResponse — the original
  two regression tests. Before the fix: both hang (a 90s timeout kill was
  needed to get a shell back). After: both pass in ~3s.
• TestConcurrentGetResponseWatchdogScoping (new, pins the review finding):
  • test_disabling_one_skill_does_not_cancel_another_skills_watchdog —
    whitebox, drives the enable/disable protocol messages directly.
    On the flat-list version this FAILS: AssertionError: 1 != 0
    (skill B's watchdog is gone after only skill A disabled). On the fixed
    version it passes.
  • test_unanswered_skill_still_aborted_when_another_skill_answers_first
    — end-to-end: two skills, two sessions, concurrent get_response()
    calls; skill A is answered via an injected utterance, skill B is
    deliberately left unanswered. On the flat-list version this hangs
    (verified via a 20s timeout kill on the whole test class). On the
    fixed version: skill A's real answer survives untouched, skill B is
    aborted by its own watchdog and resolves to None — both complete in
    ~7s total for the 4-test file.

Full ovoscope suite (after fix)

1 failed, 626 passed, 13 skipped, 3200 warnings in 151.77s
FAILED test/unittests/test_tts_intelligibility.py::TestHarnessPlaybackMode::test_playback_captures_wav_and_scores

That one failure is a pre-existing order-dependent flake, unrelated to
this change: it passes in isolation and doesn't touch
get_response/speak_dialog mocking at all. Reproduces the same way with
the fix reverted.

Real-world gate: ovos-skill-alerts PR #123

Checked out OpenVoiceOS/ovos-skill-alerts PR #123
(feat/ovoscope-tests) into a separate throwaway clone/venv (Python
3.14 in this sandbox), installed this branch's ovoscope via local path,
and ran its test/end2end suite (ovos-workshop/ovos-core/padacioso
pipeline — ovos-padatious still not installable on this box, matching
the maintainer's own note in #138).

• Before (issue fix: raise clear TypeError for bare-string expected_messages #138's documented CI evidence, Python 3.10/3.12-3.14):
  job hangs to the 30-minute timeout-minutes kill.
• After (this fix, this sandbox, serial): 30 failed, 3 passed, 133 skipped in ~12s — terminates, no hang. The failures are the
  already-documented, unrelated auto-generation defects from fix: raise clear TypeError for bare-string expected_messages #138 Update 1
  (utterances that don't satisfy the pinned adapt pipeline's
  .require(...) vocab) — out of scope here.

Note: this sandbox's exact pass/fail counts differ from the numbers seen
in review logs on a Python 3.11 + pytest-xdist setup (1 failed, 32 passed, 133 skipped in ~24s) — re-ran with pytest-xdist here too and
still saw the pre-existing vocab-mismatch failures, just distributed
differently across workers (30 failed, 2 passed, 52 skipped in ~19s, same
"terminates, no hang" outcome). The discrepancy tracks Python version /
worker distribution affecting which of the already-documented, unrelated
vocab-pinning tests execute — it does not affect the property this PR
claims: the suite reliably terminates instead of hanging.

Scope

Only ovoscope/__init__.py (MiniCroft) changed, plus the regression
test. No new bus message types. No changes to ovos-workshop (out of
scope for this repo — the _wait_response() no-deadline issue is a
separate upstream defect worth its own report there, not fixed here).

[coderabbitai]

Important

Review skipped

Draft detected.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 5ab5366a-f501-4151-b310-438894a6178d

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

I've gathered some intelligence on your latest changes. 🕵️‍♀️

I've aggregated the results of the automated checks for this PR below.

🔍 Lint

A detailed summary of the latest automation run. 📝

❌ ruff: issues found — see job log

🏷️ Release Preview

Evaluating the impact of these changes on our release schedule. 📉

Current: 1.6.4a1 → Next: 1.6.5a1

Signal	Value
Label	fix
PR title	fix: resolve unbounded get_response() hang on FakeBus
Bump	build

✅ PR title follows conventional commit format.

---

🚀 Release Channel Compatibility

Predicted next version: 1.6.5a1

Channel	Status	Note	Current Constraint
Stable	⚪	Not in channel	-
Testing	❌	Too new (must be <1.0.0)	ovoscope>=0.7.2,<1.0.0
Alpha	✅	Compatible	ovoscope>=1.6.4a1

🔒 Security (pip-audit)

Our digital defenses have been updated. 🛡️

✅ No known vulnerabilities found (79 packages scanned).

📋 Repo Health

Evaluating the longevity of the project. 🌳

✅ All required files present.

Latest Version: 1.6.4a1

✅ ovoscope/version.py — Version file
✅ README.md — README
✅ LICENSE — License file
✅ pyproject.toml — pyproject.toml
⚠️ setup.py — setup.py
✅ CHANGELOG.md — Changelog
✅ ovoscope/version.py has valid version block markers

⚖️ License Check

Scanning for any potential trademark infringements. ™️

✅ No license violations found.

Policy: Apache 2.0 (universal donor). StrongCopyleft / NetworkCopyleft / WeakCopyleft / Other / Error categories fail. MPL allowed.

🔨 Build Tests

From source to binary, let's see how it holds up. 🧱

✅ All versions pass

Python	Build	Install	Tests
3.10	✅	✅	✅
3.11	✅	✅	✅
3.12	✅	✅	✅
3.13	✅	✅	✅
3.14	✅	✅	✅

📊 Coverage

Measuring the reach of our test cases. 📏

❌ 59.0% total coverage

Files below 80% coverage (15 files)

File	Coverage	Missing lines
ovoscope/simple_listener.py	0.0%	63
ovoscope/tts_intelligibility.py	0.0%	200
ovoscope/version.py	0.0%	5
ovoscope/classic_listener.py	18.2%	117
ovoscope/intent_cases.py	21.4%	151
ovoscope/pytest_plugin.py	39.8%	237
ovoscope/cli.py	47.8%	132
ovoscope/ocp.py	47.9%	61
ovoscope/e2e.py	53.5%	66
ovoscope/media.py	57.0%	104
ovoscope/listener.py	57.1%	127
ovoscope/voice_loop.py	59.2%	118
ovoscope/audio.py	63.4%	126
ovoscope/__init__.py	64.2%	347
ovoscope/media_provider.py	67.6%	23

Full report: download the coverage-report artifact.

---

Generated with ❤️ by OVOS Automations