| 12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478147914801481148214831484148514861487148814891490149114921493149414951496149714981499150015011502150315041505150615071508150915101511151215131514151515161517151815191520152115221523152415251526152715281529153015311532153315341535153615371538153915401541154215431544154515461547154815491550155115521553155415551556155715581559156015611562156315641565156615671568156915701571157215731574157515761577157815791580158115821583158415851586158715881589159015911592159315941595159615971598159916001601160216031604160516061607160816091610161116121613161416151616161716181619162016211622162316241625162616271628162916301631163216331634163516361637163816391640164116421643164416451646164716481649165016511652165316541655165616571658165916601661166216631664166516661667166816691670167116721673167416751676167716781679168016811682168316841685168616871688168916901691169216931694169516961697169816991700170117021703170417051706170717081709171017111712171317141715171617171718171917201721172217231724172517261727172817291730173117321733173417351736173717381739174017411742174317441745174617471748174917501751175217531754175517561757175817591760176117621763176417651766176717681769177017711772177317741775177617771778177917801781178217831784178517861787178817891790179117921793179417951796179717981799180018011802180318041805180618071808180918101811181218131814181518161817181818191820182118221823182418251826182718281829183018311832183318341835183618371838183918401841184218431844184518461847184818491850185118521853185418551856185718581859186018611862186318641865186618671868186918701871187218731874187518761877187818791880188118821883188418851886188718881889189018911892189318941895189618971898189919001901190219031904190519061907190819091910191119121913191419151916191719181919192019211922192319241925192619271928192919301931193219331934193519361937193819391940194119421943194419451946194719481949195019511952195319541955195619571958195919601961196219631964196519661967196819691970197119721973197419751976197719781979198019811982198319841985198619871988198919901991199219931994199519961997199819992000200120022003200420052006200720082009201020112012201320142015201620172018201920202021202220232024202520262027202820292030203120322033203420352036203720382039204020412042204320442045204620472048204920502051205220532054205520562057205820592060206120622063206420652066206720682069207020712072207320742075207620772078207920802081208220832084208520862087208820892090209120922093209420952096209720982099210021012102210321042105210621072108210921102111211221132114211521162117211821192120212121222123212421252126212721282129213021312132213321342135213621372138213921402141214221432144214521462147214821492150215121522153215421552156215721582159216021612162216321642165216621672168216921702171217221732174217521762177217821792180218121822183218421852186218721882189219021912192219321942195219621972198 |
- """Tests for the assistant service (Phase 2) using synthetic event streams."""
- import itertools
- import json
- import re
- import threading
- import time
- from typing import Any
- from stt_daemon import events
- from stt_daemon.assistant.inference import InferenceJob, InferenceQueue
- from stt_daemon.assistant.service import EXECUTE_BUSY, AssistantService
- from stt_daemon.assistant.triage_dump import TriageDumper
- from stt_daemon.config import Config
- _sequence = itertools.count(1)
- class Clock:
- def __init__(self, start=1000.0):
- self.t = float(start)
- def __call__(self):
- return self.t
- def advance(self, dt):
- self.t += float(dt)
- class FakeSubscriber:
- def events(self):
- return iter(())
- def close(self):
- pass
- def _service():
- cfg = Config()
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- service = AssistantService(cfg, FakeSubscriber(), emit=emit, time_fn=clock)
- return service, emitted, clock
- def _ev(type_, clock, **fields):
- event = {"v": 1, "type": type_, "seq": next(_sequence), "ts": clock()}
- event.update(fields)
- return event
- def _transcript(clock, text, ts=None):
- return {
- "v": 1,
- "type": events.TRANSCRIPT,
- "seq": next(_sequence),
- "ts": clock() if ts is None else ts,
- "text": text,
- }
- def test_ready_requires_matching_segment_done():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.ready() is False
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- assert service.ready() is False # awaiting segment 1
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is True
- def test_stale_segment_done_does_not_satisfy_later_turn():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SPEECH_START, clock))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=2))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is False # still awaiting segment 2
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=2, emitted=True))
- assert service.ready() is True
- def test_gated_segment_done_clears_readiness():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=3))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=3, emitted=False))
- assert service.ready() is True
- def test_speech_stop_without_seg_id_clears_readiness():
- """A discarded turn closes with speech_stop but no seg_id; readiness must recover."""
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.ready() is False
- service.handle_event(_ev(events.SPEECH_STOP, clock)) # discarded: no seg_id
- assert service.speech_open is False
- assert service.awaited_seg_id is None
- assert service.ready() is True
- def test_status_resets_stale_speech_state_on_reconnect():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.speech_open is True
- service.handle_event(_ev(events.STATUS, clock, capture="x", segments_done=0))
- assert service.speech_open is False
- assert service.awaited_seg_id is None
- assert service.ready() is True
- def test_command_buffer_uses_arm_timestamp():
- service, _, clock = _service()
- clock.advance(10)
- assert service.arm() is True
- arm_ts = service.arm_ts
- assert arm_ts is not None
- # Replayed transcript from before arming is ignored.
- service.handle_event(_transcript(clock, "old", ts=arm_ts - 5))
- assert service.command_text == ""
- service.handle_event(_transcript(clock, "what time is it"))
- assert service.command_text == "what time is it"
- def test_mutual_exclusion_between_command_and_chat():
- service, _, _ = _service()
- assert service.toggle_chat() is True
- assert service.arm() is False # cannot arm in chat
- assert service.toggle_chat() is True # back to idle
- assert service.arm() is True
- assert service.toggle_chat() is False # cannot chat while armed
- def test_arm_timeout_disarms_and_emits_error():
- service, emitted, clock = _service()
- service.arm()
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 0.1)
- service.tick()
- assert service.modes.mode.value == "idle"
- assert service.command_text == ""
- assert any(e["type"] == "error" and e["code"] == "arm_timeout" for e in emitted)
- def test_long_speech_does_not_disarm():
- service, emitted, clock = _service()
- service.arm()
- service.handle_event(_ev(events.SPEECH_START, clock))
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 30.0)
- service.tick()
- assert service.modes.mode.value == "command_armed"
- assert not any(e["type"] == "error" for e in emitted)
- def test_captured_command_survives_arm_timeout():
- service, emitted, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "what time is it"))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- clock.advance(service.cfg.assistant.command_arm_timeout_s + 30.0)
- service.tick()
- assert service.modes.mode.value == "command_armed"
- assert service.command_text == "what time is it"
- assert not any(e["type"] == "error" for e in emitted)
- def test_stt_stall_cancels_armed_command():
- service, emitted, clock = _service()
- service.handle_event(_ev(events.HEARTBEAT, clock, worker_stall_s=0.0))
- service.arm()
- clock.advance(service.cfg.assistant.stall_timeout_ms / 1000.0 + 0.5)
- service.tick()
- assert service.modes.mode.value == "idle"
- assert any(e["type"] == "error" and e["code"] == "stt_stall" for e in emitted)
- def test_worker_stall_marks_stt_degraded():
- service, _, clock = _service()
- long_stall = service.cfg.assistant.worker_stall_ms / 1000.0 + 1.0
- service.handle_event(_ev(events.HEARTBEAT, clock, worker_stall_s=long_stall))
- assert service.stt_ok() is False
- assert service.ready() is False
- def test_chat_idle_timeout_exits_chat():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- clock.advance(service.cfg.assistant.chat_idle_timeout_s + 0.1)
- service.tick()
- assert service.modes.mode.value == "idle"
- def test_chat_exit_phrase_exits_chat():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "That's all."))
- assert service.modes.mode.value == "idle"
- assert service.chat_history == []
- def test_viewer_detached_exits_chat_and_forgets_history():
- service, _, clock = _service()
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "hello"))
- assert service.viewer_detached() is True
- assert service.modes.mode.value == "idle"
- assert service.chat_history == []
- def test_viewer_detached_is_noop_outside_chat():
- service, _, _ = _service()
- assert service.viewer_detached() is False
- assert service.modes.mode.value == "idle"
- def test_exit_chat_invokes_chat_reset():
- cfg = Config()
- resets: list[int] = []
- service = AssistantService(
- cfg, FakeSubscriber(), time_fn=Clock(), chat_reset=lambda: resets.append(1)
- )
- service.toggle_chat()
- service.viewer_detached()
- assert resets == [1]
- def test_chat_accumulates_turns():
- service, _, clock = _service()
- service.toggle_chat()
- for text in ("hello there", "how are you"):
- service.handle_event(_transcript(clock, text))
- assert service.chat_history == ["Unknown: hello there", "Unknown: how are you"]
- def test_execute_requires_readiness():
- service, _, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "do it"))
- service.handle_event(_ev(events.SPEECH_START, clock))
- assert service.execute() is None # speech still open
- assert service.modes.mode.value == "command_armed"
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- assert service.execute() is None # awaiting segment 1
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.execute() == "do it"
- assert service.modes.mode.value == "executing"
- assert service.finish() is True
- assert service.modes.mode.value == "idle"
- def test_execute_reports_busy_when_queue_rejects():
- service, emitted, clock = _service()
- # A zero-capacity lane rejects the priority command job on submit.
- service._inference = InferenceQueue(size=0)
- service.executor = object()
- service.arm()
- service.handle_event(_transcript(clock, "do it"))
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.execute() is EXECUTE_BUSY
- assert service.modes.mode.value == "idle"
- assert not any(e.get("code") == "busy" for e in emitted)
- def test_turn_ended_requires_quiet_gap():
- service, _, clock = _service()
- service.handle_event(_ev(events.SPEECH_STOP, clock, seg_id=1))
- service.handle_event(_ev(events.SEGMENT_DONE, clock, seg_id=1, emitted=True))
- assert service.ready() is True
- assert service.turn_ended() is False
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- assert service.turn_ended() is True
- def test_emits_mode_and_command_text():
- service, emitted, clock = _service()
- service.arm()
- assert {"type": "mode", "mode": "command_armed"} in emitted
- service.handle_event(_transcript(clock, "hi"))
- assert any(e["type"] == "command_text" and e["text"] == "hi" for e in emitted)
- def test_cancel_clears_command_buffer():
- service, _, clock = _service()
- service.arm()
- service.handle_event(_transcript(clock, "hi"))
- assert service.cancel() is True
- assert service.modes.mode.value == "idle"
- assert service.command_text == ""
- assert service.arm_ts is None
- def test_pending_chat_is_not_lost_across_threads():
- """Transcripts appended on the reader thread while the ticker dispatches
- must all reach the model: the join-and-clear in _dispatch_chat is atomic,
- and the user bubble is emitted by the worker in _run_chat."""
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.01
- dispatched: list[str] = []
- lock = threading.Lock()
- def emit(type_, **fields):
- if type_ == "chat_message" and fields.get("role") == "user":
- with lock:
- dispatched.append(fields["text"])
- service = AssistantService(cfg, FakeSubscriber(), emit=emit, chat_fn=lambda text: "ok")
- service.toggle_chat()
- service.start()
- try:
- total = 200
- producer = threading.Thread(
- target=lambda: [
- service.handle_event(
- {
- "v": 1,
- "type": events.TRANSCRIPT,
- "seq": i,
- "ts": time.time(),
- "text": f"w{i}",
- }
- )
- for i in range(total)
- ],
- daemon=True,
- )
- producer.start()
- deadline = time.time() + 10.0
- while time.time() < deadline:
- service.tick()
- with lock:
- words = sum(len(text.split()) for text in dispatched)
- done = words == 2 * total
- if not producer.is_alive() and done:
- break
- time.sleep(0.002)
- producer.join(timeout=5)
- with lock:
- words = sum(len(text.split()) for text in dispatched)
- assert words == 2 * total
- finally:
- service.stop()
- def _full_queue_service():
- service, emitted, clock = _service()
- service.toggle_chat()
- service._inference = InferenceQueue(size=1)
- service._inference.submit(
- InferenceJob(name="blocker", run=lambda: None, evictable=False)
- )
- return service, emitted, clock
- def test_chat_refusal_rebuffers_and_emits_queue_full_once():
- """A refused chat submit re-buffers under lock and emits queue_full on flip."""
- service, emitted, _ = _full_queue_service()
- service._pending_chat = ["hello world"]
- service._dispatch_chat()
- assert service._pending_chat == ["hello world"]
- assert service._chat_queue_full is True
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "queue_full"]
- assert len(errors) == 1
- assert [e for e in emitted if e["type"] == "chat_message"] == []
- service._dispatch_chat()
- assert service._pending_chat == ["hello world"]
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "queue_full"]
- assert len(errors) == 1
- def test_chat_dispatch_success_clears_latch_without_user_bubble():
- """Success clears the latch; the user bubble comes from _run_chat, not dispatch."""
- service, emitted, _ = _full_queue_service()
- service._pending_chat = ["hello world"]
- service._chat_queue_full = True
- service._inference.drop(lambda j: True)
- service._dispatch_chat()
- assert service._pending_chat == []
- assert service._chat_queue_full is False
- assert [e for e in emitted if e["type"] == "chat_message"] == []
- def test_run_chat_emits_user_first_then_reply():
- service, emitted, _ = _service()
- service.chat_fn = lambda text: "reply!"
- service._run_chat("hello")
- assert emitted == [
- {"type": "chat_message", "role": "user", "text": "hello"},
- {"type": "chat_message", "role": "assistant", "text": "reply!"},
- ]
- def test_chat_notify_emits_compaction_note():
- service, emitted, _ = _service()
- service.chat_notify("compacted", turns=4)
- assert {
- "type": "chat_message",
- "role": "note",
- "text": "Compacted 4 earlier turns into a summary.",
- } in emitted
- def test_chat_notify_emits_truncated_error():
- service, emitted, _ = _service()
- service.chat_notify("truncated", message="cut off")
- assert {"type": "error", "code": "truncated", "message": "cut off"} in emitted
- def test_clear_empties_chat_history():
- service, emitted, clock = _service()
- service.toggle_chat()
- for text in ("hello", "how are you"):
- service.handle_event(_transcript(clock, text))
- assert service.chat_history == ["Unknown: hello", "Unknown: how are you"]
- assert len(service._pending_chat) == 2
- assert service.clear() is True
- assert service.chat_history == []
- assert service._pending_chat == []
- assert {"type": "cleared"} in emitted
- def test_clear_invokes_chat_reset():
- cfg = Config()
- resets: list[int] = []
- service = AssistantService(
- cfg, FakeSubscriber(), time_fn=Clock(), chat_reset=lambda: resets.append(1)
- )
- service.toggle_chat()
- service.clear()
- assert resets == [1]
- def test_clear_does_not_change_mode():
- service, _, clock = _service()
- service.toggle_chat()
- assert service.modes.mode.value == "chat"
- service.clear()
- assert service.modes.mode.value == "chat"
- def test_clear_always_returns_true():
- service, _, _ = _service()
- assert service.clear() is True
- service.toggle_chat()
- service.handle_event(_ev(events.TRANSCRIPT, Clock(), text="test"))
- assert service.clear() is True
- def test_heartbeat_emits_inference_depth_and_in_flight():
- """Calling tick() emits heartbeat with inference_depth (int) and inference_in_flight (bool)."""
- service, emitted, clock = _service()
- service.tick()
- heartbeats = [e for e in emitted if e["type"] == "heartbeat"]
- assert len(heartbeats) == 1
- heartbeat = heartbeats[0]
- assert "inference_depth" in heartbeat
- assert isinstance(heartbeat["inference_depth"], int)
- assert "inference_in_flight" in heartbeat
- assert isinstance(heartbeat["inference_in_flight"], bool)
- def test_status_contains_inference_depth_and_in_flight():
- """status() snapshot contains inference_depth (int) and inference_in_flight (bool)."""
- service, _, _ = _service()
- status = service.status()
- assert "inference_depth" in status
- assert isinstance(status["inference_depth"], int)
- assert "inference_in_flight" in status
- assert isinstance(status["inference_in_flight"], bool)
- _AUTO = object()
- class FakeTriage:
- def __init__(
- self,
- questions=_AUTO,
- state_pair=("ok", ""),
- escalate=False,
- parsed=_AUTO,
- normalized=_AUTO,
- gate=True,
- ):
- if questions is _AUTO:
- questions = {"urgency": {"type": "noul", "instructions": "how urgent"}}
- self._questions = questions
- self._state_pair = state_pair
- self._escalate = escalate
- self._parsed = parsed
- self._normalized = normalized
- self._gate = gate
- self.build_state_calls: list[tuple] = []
- self.normalize_calls: list = []
- self.prompt_calls: list[tuple] = []
- self.gate_calls: list = []
- @property
- def questions(self):
- return self._questions
- def state(self):
- return self._state_pair
- def build_state(self, window):
- self.build_state_calls.append(tuple(window))
- return "\n".join("- " + str(item) for item in (window or []))
- def normalize(self, payload):
- self.normalize_calls.append(payload)
- if self._normalized is not _AUTO:
- return self._normalized
- if not isinstance(payload, dict):
- return {}
- signals = {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- signals.update(payload)
- if self._escalate:
- signals["escalated"] = True
- return signals
- def build_brief_prompt(self, deltas, signals):
- self.prompt_calls.append(
- (tuple(deltas), dict(signals) if isinstance(signals, dict) else signals)
- )
- return "BRIEF-PROMPT"
- def parse_brief(self, text):
- if self._parsed is _AUTO:
- return text if text else None
- return self._parsed
- def should_update_brief(self, signals):
- self.gate_calls.append(signals)
- return self._gate
- class FakeBriefFile:
- """Recording stand-in for BriefFile (write/clear never raise)."""
- def __init__(self):
- self.writes: list[str] = []
- self.clears = 0
- def write(self, text):
- self.writes.append(text)
- def clear(self):
- self.clears += 1
- class FakeLane:
- def __init__(self, ready=True):
- self._ready = ready
- self.submits: list[tuple] = []
- self.drops = 0
- self.predict_gate = threading.Event()
- def ready(self):
- return self._ready
- def state(self):
- return ("ok", "") if self._ready else ("loading", "")
- def submit(self, state, questions, lang, generation):
- self.submits.append((state, questions, lang, generation))
- return True
- def drop_pending(self):
- self.drops += 1
- def _ok_reply(text):
- return "ok"
- def _triage_service(
- triage=None,
- lane=None,
- brief_fn=None,
- chat_fn: Any = _AUTO,
- brief_file=None,
- brief_reset=None,
- chat_reset=None,
- ):
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.5
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- if triage is None:
- triage = FakeTriage()
- if lane is None:
- lane = FakeLane()
- if chat_fn is _AUTO:
- chat_fn = _ok_reply
- service = AssistantService(
- cfg,
- FakeSubscriber(),
- emit=emit,
- time_fn=clock,
- triage=triage,
- laya_lane=lane,
- brief_fn=brief_fn,
- brief_reset=brief_reset,
- brief_file=brief_file,
- chat_fn=chat_fn,
- chat_reset=chat_reset,
- )
- return service, emitted, clock, triage, lane
- def _triage_events(emitted):
- return [e for e in emitted if e["type"] == "triage"]
- def _escalated_payload():
- return {"order": [], "items": {}, "escalated": True, "input_tokens": 3}
- def test_triage_status_disabled_when_no_analyzer():
- service, _, _ = _service()
- status = service.status()
- assert status["triage_state"] == "disabled"
- assert status["triage_reason"] == ""
- assert status["triage"] is None
- def test_triage_status_reflects_analyzer():
- service, _, _, _, _ = _triage_service(
- triage=FakeTriage(state_pair=("loading", "warming up"))
- )
- status = service.status()
- assert status["triage_state"] == "loading"
- assert status["triage_reason"] == "warming up"
- assert status["triage"] is None
- def test_status_contains_questions_from_analyzer():
- questions = {
- "urgency": {"type": "noul", "instructions": "how urgent"},
- "sentiment": {"type": "score", "instructions": "tone", "criteria": ["low", "high"]},
- }
- service, _, _, triage, _ = _triage_service(triage=FakeTriage(questions=questions))
- status = service.status()
- assert status["questions"] == questions
- assert status["questions"] is triage.questions
- def test_status_questions_none_without_analyzer():
- service, _, _ = _service()
- assert service.status()["questions"] is None
- def test_status_event_carries_questions():
- service, emitted, _, triage, _ = _triage_service()
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 1
- assert statuses[0]["questions"] is triage.questions
- def test_triage_captures_transcript_lang():
- service, _, clock, _, _ = _triage_service()
- assert service._last_transcript_lang is None
- event = _transcript(clock, "hola")
- event["lang"] = "es"
- service.handle_event(event)
- assert service._last_transcript_lang == "es"
- service.handle_event(_transcript(clock, "hello"))
- assert service._last_transcript_lang is None
- def test_maybe_triage_passes_lang_to_lane():
- service, _, clock, _, lane = _triage_service()
- event = _transcript(clock, "hola")
- event["lang"] = "es"
- service.handle_event(event)
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, lang, _ = lane.submits[0]
- assert lang == "es"
- def test_maybe_triage_runs_in_idle_mode():
- service, _, clock, triage, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- state, questions, lang, gen = lane.submits[0]
- assert state == "- hello"
- assert questions is triage.questions
- assert lang is None
- assert gen == 0
- assert triage.build_state_calls == [("hello",)]
- assert service.triage_history == ["hello"]
- def test_maybe_triage_uses_transcript_buffer_not_chat_history():
- service, _, clock, triage, lane = _triage_service(chat_fn=None)
- service.chat_history = ["stale chat only"]
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- assert triage.build_state_calls == []
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- assert triage.build_state_calls == [("hello",)]
- def test_maybe_triage_requires_settled_turn():
- service, _, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- service.tick()
- assert lane.submits == []
- def test_maybe_triage_skips_unready_lane():
- service, _, clock, _, lane = _triage_service(lane=FakeLane(ready=False))
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- def test_maybe_triage_skips_without_questions():
- service, _, clock, triage, lane = _triage_service(
- triage=FakeTriage(questions=None)
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert lane.submits == []
- assert triage.build_state_calls == []
- def test_maybe_triage_dedupes_identical_window():
- service, _, clock, triage, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- state, questions, lang, gen = lane.submits[0]
- assert state == "- hello"
- assert questions is triage.questions
- assert lang is None
- assert gen == 0
- service.tick()
- service.tick()
- assert len(lane.submits) == 1
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- assert lane.submits[1][0] == "- hello\n- second"
- def test_second_transcript_produces_second_submit():
- service, _, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- assert lane.submits[1][0] == "- first\n- second"
- def test_tick_under_50ms_with_blocked_lane():
- service, _, clock, _, lane = _triage_service()
- assert not lane.predict_gate.is_set()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- start = time.perf_counter()
- service.tick()
- elapsed = time.perf_counter() - start
- assert len(lane.submits) == 1
- assert elapsed < 0.05
- def test_on_triage_result_emits_signals():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- payload = {
- "order": ["urgency"],
- "items": {"urgency": {"kind": "noul"}},
- "escalated": False,
- "input_tokens": 12,
- }
- service._on_triage_result(gen, payload)
- staged = _triage_events(emitted)
- assert len(staged) == 1
- assert staged[0]["stage"] == "signals"
- assert staged[0]["triage"] == service.last_triage
- assert staged[0]["triage"]["escalated"] is False
- assert staged[0]["triage"]["input_tokens"] == 12
- assert service.status()["triage"] == service.last_triage
- def test_triage_events_carry_the_server_timestamp():
- service, emitted, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- expected_ts = clock()
- service._on_triage_result(gen, _escalated_payload())
- signals = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert len(signals) == 1
- assert isinstance(signals[0]["ts"], (int, float))
- assert signals[0]["ts"] == expected_ts
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- job.run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert isinstance(contents[0]["ts"], (int, float))
- assert contents[0]["ts"] == expected_ts # content reuses the signals timestamp
- def test_on_triage_result_suppresses_stale_generation():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service.clear()
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert [e for e in _triage_events(emitted) if e.get("stage") == "signals"] == []
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- def test_pending_brief_capped_by_char_budget():
- """The buffer keeps the newest deltas within budget, dropping the oldest."""
- service, _, clock, _, _ = _triage_service(brief_fn=lambda prompt: "RAW")
- service.cfg.triage.brief_pending_char_budget = 30
- service.handle_event(_transcript(clock, "aaa"))
- service.handle_event(_transcript(clock, "bbbb"))
- assert service._pending_brief == ["Unknown: aaa", "Unknown: bbbb"]
- service.handle_event(_transcript(clock, "ccccc"))
- assert service._pending_brief == ["Unknown: bbbb", "Unknown: ccccc"]
- assert service.triage_history == ["aaa", "bbbb", "ccccc"]
- def test_pending_brief_untouched_without_triage():
- service, _, clock = _service()
- service.handle_event(_transcript(clock, "hello"))
- assert service._pending_brief == []
- assert service.triage_history == ["hello"]
- def test_on_triage_result_accepts_non_chat_mode():
- service, emitted, clock, _, lane = _triage_service()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- window = service._last_triage_window
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service._last_triage_window == window
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- def test_on_triage_result_ignores_none_payload():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- def test_none_payload_dispatches_nothing_and_keeps_deltas():
- """A Laya predict failure never dispatches: no job, no events, deltas stay."""
- calls: list[str] = []
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: calls.append(prompt) or "RAW"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- assert service._pending_brief == ["Unknown: hello"]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert [e.get("stage") for e in _triage_events(emitted)] == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert calls == []
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_none_payload_without_brief_fn_enqueues_nothing():
- service, emitted, clock, _, lane = _triage_service()
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, None)
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- def test_on_triage_result_ignores_empty_signals():
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(normalized={})
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert _triage_events(emitted) == []
- def test_substantive_turn_dispatches_one_brief_job():
- """A substantive settled turn calls the gate once and queues one brief job."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(escalate=True), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "we might churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(
- gen, {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- )
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert service._last_brief_signals == service.last_triage
- assert service._inference.depth() == 2
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert len(triage.gate_calls) == 1
- assert service._brief_outstanding is True
- assert service._brief_token == 1
- def test_brief_dispatches_without_escalation():
- service, _, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "all calm"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(
- gen, {"order": [], "items": {}, "escalated": False, "input_tokens": 0}
- )
- assert service.last_triage is not None
- assert service._inference.depth() == 2
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- def test_gated_out_turn_skips_brief():
- """A gated-out turn stores signals but never calls brief_fn."""
- calls: list[str] = []
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(gate=False),
- brief_fn=lambda prompt: calls.append(prompt) or "RAW",
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert "brief" not in service.last_triage # no prior brief to retain
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert calls == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._brief_outstanding is False
- assert len(triage.gate_calls) == 1
- def test_gated_out_turn_retains_previous_brief():
- """A gated-out settled turn keeps the last accepted brief in the panel."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(gate=False), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- assert "brief" not in service._last_brief_signals # raw signals stay raw
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert len(staged) == 1
- assert staged[0]["triage"] == service.last_triage
- assert staged[0]["triage"]["brief"] == "RAW"
- assert len(triage.gate_calls) == 1
- def test_gated_out_turn_retains_brief_when_extract_disabled():
- """`extract_enabled = false` skips the dispatch but still retains the brief."""
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(gate=False), brief_fn=lambda prompt: "RAW"
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- service.cfg.triage.extract_enabled = False
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- assert [j.name for j in service._inference._queue] == ["chat"]
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert staged[0]["triage"]["brief"] == "RAW"
- def test_gated_out_turn_retains_brief_without_brief_fn():
- """No brief_fn means no dispatch, but a retained brief is still carried."""
- service, emitted, clock, _, lane = _triage_service(triage=FakeTriage(gate=False))
- service.toggle_chat()
- service.last_triage = {**_escalated_payload(), "brief": "KEEP"}
- emitted.clear()
- service.handle_event(_transcript(clock, "small talk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage["brief"] == "KEEP"
- staged = [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- assert staged[0]["triage"]["brief"] == "KEEP"
- def test_later_accepted_update_replaces_retained_brief():
- """A later accepted update overwrites the retained brief text."""
- texts = iter(["OLD", "NEW"])
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: next(texts))
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "OLD"
- token = _seed_brief(service, "Unknown: second")
- service._run_brief_update(0, signals, service._now(), token)
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "NEW"
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert contents[-1]["triage"]["brief"] == "NEW"
- service.clear()
- assert service.last_triage is None
- def test_no_brief_when_disabled_in_config():
- calls: list[str] = []
- service, _, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True),
- brief_fn=lambda prompt: calls.append(prompt) or "RAW",
- )
- service.cfg.triage.extract_enabled = False
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert calls == []
- assert [j.name for j in service._inference._queue] == ["chat"]
- def test_no_brief_without_brief_fn():
- service, _, clock, _, lane = _triage_service(triage=FakeTriage(escalate=True))
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._brief_outstanding is False
- def test_brief_refused_when_queue_full():
- """A refused submit leaves deltas buffered and the outstanding flag clear."""
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True), brief_fn=lambda prompt: "RAW"
- )
- service._inference = InferenceQueue(size=2)
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn risk"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert service._inference.depth() == 1
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service._inference.depth() == 1
- assert service._inference._queue[0].name == "chat"
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is True
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert service._pending_brief == ["Unknown: churn risk"]
- assert service._brief_outstanding is False
- def test_second_result_while_outstanding_is_refused():
- """At most one brief job is outstanding: the second turn only stores signals."""
- service, emitted, clock, _, lane = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(gen0, _escalated_payload())
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals", "signals"]
- assert service._brief_outstanding is True
- assert service._pending_brief == ["Unknown: first", "Unknown: second"]
- def _seed_brief(service, *deltas):
- service._pending_brief = list(deltas)
- service._brief_outstanding = True
- return service._brief_token
- def test_brief_merges_content_and_writes_file():
- brief_file = FakeBriefFile()
- service, emitted, _, triage, _ = _triage_service(
- brief_fn=lambda prompt: "the brief", brief_file=brief_file
- )
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, signals, ts, token)
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert contents[0]["triage"] == {**signals, "brief": "the brief"}
- assert service.last_triage == contents[0]["triage"]
- assert triage.prompt_calls == [(("Unknown: hello",), signals)]
- assert brief_file.writes == ["the brief"]
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_empty_brief_requeues_deltas():
- """A `""` reply keeps signals, re-buffers the deltas, and clears the flag."""
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: "")
- service.toggle_chat()
- signals = _escalated_payload()
- service.last_triage = signals
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, signals, ts, token)
- assert [e for e in _triage_events(emitted) if e.get("stage") == "content"] == []
- assert service.last_triage == signals
- assert "brief" not in service.last_triage
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_brief_error_requeues_and_clears_flag():
- """A raising brief_fn leaves the flag clear and the deltas buffered."""
- def boom(prompt):
- raise RuntimeError("llm down")
- service, emitted, _, _, _ = _triage_service(brief_fn=boom)
- service.toggle_chat()
- token = _seed_brief(service, "Unknown: hello")
- ts = service._now()
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert _triage_events(emitted) == []
- assert service.last_triage is None
- assert service._pending_brief == ["Unknown: hello"]
- assert service._brief_outstanding is False
- def test_brief_empty_buffer_returns_early():
- """An empty buffer returns early: no call, no event, flag cleared."""
- calls: list[str] = []
- service, emitted, _, _, _ = _triage_service(
- brief_fn=lambda prompt: calls.append(prompt) or "RAW"
- )
- service.toggle_chat()
- service._brief_outstanding = True
- token = service._brief_token
- ts = service._now()
- assert service._pending_brief == []
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert calls == []
- assert _triage_events(emitted) == []
- assert service._brief_outstanding is False
- def test_brief_suppressed_after_reset():
- service, emitted, _, _, _ = _triage_service(brief_fn=lambda prompt: "RAW")
- service.toggle_chat()
- service._pending_brief = ["Unknown: hello"]
- service._reset_triage()
- emitted.clear()
- service._run_brief_update(0, _escalated_payload(), service._now(), 0)
- assert _triage_events(emitted) == []
- assert service.last_triage is None
- assert service._pending_brief == ["Unknown: hello"]
- def test_brief_works_after_leaving_chat():
- seen = []
- service, emitted, _, _, _ = _triage_service(
- brief_fn=lambda prompt: seen.append(prompt) or "RAW"
- )
- service.toggle_chat()
- assert service.modes.exit_chat() is True
- token = _seed_brief(service, "Unknown: hello")
- signals = _escalated_payload()
- service._last_brief_signals = signals
- service._run_brief_update(0, signals, service._now(), token)
- assert seen != []
- assert [e.get("stage") for e in _triage_events(emitted)] == ["content"]
- assert service.last_triage is not None
- assert service.last_triage["brief"] == "RAW"
- def test_deltas_arriving_during_call_survive_and_flush():
- """Deltas appended mid-call are not consumed; the finally flushes them."""
- service, emitted, clock, _, _ = _triage_service()
- service.toggle_chat()
- signals = _escalated_payload()
- service._last_brief_signals = signals
- token = _seed_brief(service, "Unknown: first")
- appended: list[int] = []
- def _brief(prompt):
- if not appended:
- appended.append(1)
- service.handle_event(_transcript(clock, "during call"))
- return "first brief"
- service._brief_fn = _brief
- service._run_brief_update(0, signals, clock(), token)
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert contents[0]["triage"]["brief"] == "first brief"
- flushed = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(flushed) == 1
- assert service._brief_outstanding is True
- assert service._pending_brief == ["Unknown: during call"]
- flushed[0].run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 2
- assert contents[1]["triage"]["brief"] == "first brief"
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_in_flight_brief_after_clear_emits_and_writes_nothing():
- """Stale in-flight job: no event, no file write, no requeue, flag untouched."""
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW", brief_file=brief_file
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- assert service._brief_outstanding is True
- service.clear()
- assert service._brief_outstanding is False
- token_after_clear = service._brief_token
- emitted.clear()
- job.run()
- assert [e for e in _triage_events(emitted)] == []
- assert brief_file.writes == []
- assert service._pending_brief == []
- assert service.last_triage is None
- assert service._brief_token == token_after_clear
- assert service._brief_outstanding is False
- def test_clear_resets_triage():
- resets: list[int] = []
- chats: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- triage=FakeTriage(escalate=True),
- brief_fn=lambda prompt: "RAW",
- brief_file=brief_file,
- brief_reset=lambda: resets.append(1),
- chat_reset=lambda: chats.append(1),
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service._inference.depth() == 2
- assert service._pending_brief == ["Unknown: churn"]
- assert service._brief_outstanding is True
- token_before = service._brief_token
- assert service.clear() is True
- assert service._triage_generation == 1
- assert service.last_triage is None
- assert service._last_triage_window == ()
- assert service.triage_history == []
- assert service._pending_brief == []
- assert service._brief_token == token_before + 1
- assert service._brief_outstanding is False
- assert service._last_brief_signals is None
- assert resets == [1]
- assert brief_file.clears == 1
- assert chats == [1]
- assert lane.drops == 1
- assert service._inference.depth() == 0
- assert list(service._inference._queue) == []
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert len(cleared) == 1
- assert cleared[0]["triage"] is None
- assert {"type": "cleared"} in emitted
- signals_before = len(
- [e for e in _triage_events(emitted) if e.get("stage") == "signals"]
- )
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is None
- assert (
- len([e for e in _triage_events(emitted) if e.get("stage") == "signals"])
- == signals_before
- )
- submits_before = len(lane.submits)
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == submits_before
- def test_exit_chat_does_not_reset_triage():
- chats: list[int] = []
- bresets: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW",
- chat_reset=lambda: chats.append(1),
- brief_reset=lambda: bresets.append(1),
- brief_file=brief_file,
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- assert service.last_triage is not None
- assert service._triage_generation == 0
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "idle"
- assert service._triage_generation == 0
- assert service.last_triage is not None
- assert service._last_triage_window == ("hello",)
- assert service.triage_history == ["hello"]
- assert service._pending_brief == ["Unknown: hello"]
- assert service._last_brief_signals is not None
- assert service._brief_outstanding is True
- assert lane.drops == 0
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert cleared == []
- assert chats == [1]
- assert bresets == []
- assert brief_file.clears == 0
- def test_toggle_chat_enter_does_not_touch_triage():
- """Entering chat resets only the chat buffers; the brief engine is independent."""
- bresets: list[int] = []
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW",
- brief_reset=lambda: bresets.append(1),
- brief_file=brief_file,
- )
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- triage_before = service.last_triage
- pending_before = list(service._pending_brief)
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "chat"
- assert service._triage_generation == 0
- assert service.last_triage is triage_before
- assert service._last_triage_window == ("hello",)
- assert service.triage_history == ["hello"]
- assert service._pending_brief == pending_before
- assert service._last_brief_signals is not None
- assert service._brief_outstanding is True
- assert lane.drops == 0
- assert bresets == []
- assert brief_file.clears == 0
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert cleared == []
- def test_clear_without_triage_emits_no_triage_event():
- service, emitted, _ = _service()
- brief_file = FakeBriefFile()
- bresets: list[int] = []
- service._brief_file = brief_file
- service._brief_reset = lambda: bresets.append(1)
- service.clear()
- assert {"type": "cleared"} in emitted
- assert _triage_events(emitted) == []
- assert brief_file.clears == 1
- assert bresets == []
- def test_stale_brief_after_clear_does_not_clear_newer_dispatch_flag():
- """A stale in-flight job must not clear a newer dispatch's outstanding flag."""
- brief_file = FakeBriefFile()
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW", brief_file=brief_file
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, _escalated_payload())
- stale_job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- assert service._brief_outstanding is True
- service.clear()
- # A fresh turn dispatches a new brief AFTER clear bumped the token.
- service.handle_event(_transcript(clock, "fresh"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen2 = lane.submits[1]
- service._on_triage_result(gen2, _escalated_payload())
- assert service._brief_outstanding is True
- newer_token = service._brief_token
- pending_before = list(service._pending_brief)
- triage_before = service.last_triage
- emitted.clear()
- stale_job.run()
- assert _triage_events(emitted) == []
- assert brief_file.writes == []
- assert service._pending_brief == pending_before
- assert service._brief_outstanding is True
- assert service._brief_token == newer_token
- assert service.last_triage is triage_before
- def test_clear_resets_chat_latch():
- service, emitted, _ = _service()
- service.toggle_chat()
- service._chat_queue_full = True
- service._pending_chat = ["stuck text"]
- service.clear()
- assert service._chat_queue_full is False
- assert service._pending_chat == []
- assert {"type": "cleared"} in emitted
- def test_triage_end_to_end_signals_content_cleared():
- """Two settled turns collapse into ONE brief job (single-outstanding guard).
- The second `_on_triage_result` is refused as `"outstanding"`, so the queue
- holds one `brief-update` that consumes both deltas and emits one `content`.
- The panel keeps the FIRST turn's signals until the next settled turn (lag).
- """
- prompts: list[str] = []
- service, emitted, clock, triage, lane = _triage_service(
- brief_fn=lambda prompt: prompts.append(prompt) or "the brief"
- )
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first thought"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 1
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(
- gen0, {"order": ["urgency"], "items": {}, "escalated": False, "input_tokens": 1}
- )
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals"]
- assert prompts == []
- assert service.last_triage is not None
- assert service.last_triage["escalated"] is False
- assert [j.name for j in service._inference._queue] == ["chat", "brief-update"]
- triage._escalate = True
- service.handle_event(_transcript(clock, "we might churn"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 2
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- assert [e.get("stage") for e in _triage_events(emitted)] == ["signals", "signals"]
- assert [j.name for j in service._inference._queue] == [
- "chat",
- "brief-update",
- "chat",
- ]
- for job in [j for j in service._inference._queue if j.name == "brief-update"]:
- job.run()
- contents = [e for e in _triage_events(emitted) if e.get("stage") == "content"]
- assert len(contents) == 1
- assert len(prompts) == 1
- deltas, _ = triage.prompt_calls[0]
- assert deltas == ("Unknown: first thought", "Unknown: we might churn")
- assert contents[0]["triage"]["escalated"] is False
- assert contents[0]["triage"]["brief"] == "the brief"
- assert service.last_triage == contents[0]["triage"]
- assert service.status()["triage"] == contents[0]["triage"]
- service.clear()
- cleared = [e for e in _triage_events(emitted) if e.get("stage") == "cleared"]
- assert len(cleared) == 1
- assert cleared[0]["triage"] is None
- assert service.last_triage is None
- service._on_triage_result(gen1, _escalated_payload())
- assert [e.get("stage") for e in _triage_events(emitted)] == [
- "signals",
- "signals",
- "content",
- "cleared",
- ]
- assert service.toggle_chat() is True
- service.handle_event(_transcript(clock, "idle thought"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- assert len(lane.submits) == 3
- assert lane.submits[2][0].endswith("- idle thought")
- def test_tick_emits_status_on_triage_state_change():
- service, emitted, _, triage, _ = _triage_service(
- triage=FakeTriage(state_pair=("loading", "warming up"))
- )
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 1
- assert statuses[0]["triage_state"] == "loading"
- assert statuses[0]["triage_reason"] == "warming up"
- assert "triage" in statuses[0]
- triage._state_pair = ("ok", "")
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert len(statuses) == 2
- assert statuses[1]["triage_state"] == "ok"
- assert statuses[1]["triage_reason"] == ""
- assert "triage" in statuses[1]
- service.tick()
- assert len([e for e in emitted if e["type"] == "status"]) == 2
- def test_no_status_emitted_when_triage_disabled():
- service, emitted, _ = _service()
- service.tick()
- service.tick()
- assert [e for e in emitted if e["type"] == "status"] == []
- def test_tick_heals_triage_panel_after_lane_ready():
- import time as _time
- from pathlib import Path
- from stt_daemon.assistant.laya_lane import LayaLane
- from stt_daemon.assistant.triage import TriageAnalyzer
- from stt_daemon.tui.state import ViewState
- class _FakeRouter:
- def preload(self, names):
- return self
- def predict(self, state, questions, model=None, lang=None, **kwargs):
- if "_warmup" in questions:
- return {"model": model, "answers": {}, "usage": {}, "routing": {}}
- return {
- "model": model,
- "answers": {
- "frustration": {
- "noul": 0.1,
- "confidence": 0.9,
- "answer_confidence": 0.9,
- },
- "urgency": {
- "choice": "low",
- "confidence": 0.9,
- "answer_confidence": 0.9,
- },
- },
- "usage": {"input_tokens": 5},
- }
- def _factory(**kwargs):
- return _FakeRouter()
- cfg = Config()
- cfg.assistant.turn_gap_s = 0.5
- cfg.triage.questions_file = str(
- Path(__file__).resolve().parent.parent / "questions.example.json"
- )
- clock = Clock()
- emitted: list[dict] = []
- def emit(type_, **fields):
- emitted.append({"type": type_, **fields})
- analyzer = TriageAnalyzer(cfg.triage)
- assert analyzer.questions, "example questions must load"
- holder: dict = {}
- lane = LayaLane(
- cfg.triage,
- agent_factory=_factory,
- on_result=lambda gen, payload: holder["svc"]._on_triage_result(gen, payload),
- on_state=lambda state, reason: analyzer.refresh(state, reason),
- )
- service = AssistantService(
- cfg,
- FakeSubscriber(),
- emit=emit,
- time_fn=clock,
- triage=analyzer,
- laya_lane=lane,
- chat_fn=_ok_reply,
- )
- holder["svc"] = service
- try:
- lane.start()
- deadline = _time.monotonic() + 5.0
- while not lane.ready() and _time.monotonic() < deadline:
- _time.sleep(0.005)
- assert lane.ready(), "lane did not become ready"
- assert analyzer.state() == ("ok", "")
- service.tick()
- statuses = [e for e in emitted if e["type"] == "status"]
- assert statuses, "tick emitted no status after lane ready"
- assert statuses[-1]["triage_state"] == "ok"
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- deadline = _time.monotonic() + 5.0
- while service.last_triage is None and _time.monotonic() < deadline:
- _time.sleep(0.005)
- assert service.last_triage is not None, "no triage signals from lane"
- view = ViewState()
- for event in emitted:
- view.ingest(event)
- assert view.triage_state == "ok"
- lines = view.triage_lines()
- assert lines
- assert not any("unavailable" in line for line in lines)
- assert not any("loading" in line for line in lines)
- finally:
- lane.stop()
- def _dump_brief(prompt, on_io=None):
- """Brief fake that mimics LlmClient.chat's on_io phases."""
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {"prompt": prompt}})
- on_io({"phase": "response", "payload": {"content": "parsed"}, "error": None})
- return "BRIEF"
- def _run_triage_flow(service, emitted, clock, lane, payload):
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- service._on_triage_result(gen, payload)
- return gen
- def _dump_files(tmp_path):
- return list(tmp_path.rglob("*.json"))
- def test_dump_writes_full_laya_and_llm_io(tmp_path):
- service, emitted, clock, triage, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- jobs[0].run()
- files = _dump_files(tmp_path)
- assert len(files) == 1
- assert re.fullmatch(r"\d{6}-\d{3}\.json", files[0].name)
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["ts"] == clock()
- assert isinstance(data["iso"], str) and data["iso"]
- assert data["laya"]["request"]["state"] == "- hello"
- assert data["laya"]["request"]["questions"] == triage.questions
- assert data["laya"]["request"]["model"] == "multilingual"
- assert data["laya"]["request"]["max_len"] == 1024
- assert data["laya"]["response"] == _escalated_payload()
- assert data["laya"]["error"] is None
- assert data["llm"]["request"]["url"] == "http://llm/v1"
- assert data["llm"]["request"]["body"] == {"prompt": "BRIEF-PROMPT"}
- assert data["llm"]["response"] == {"content": "parsed"}
- assert data["llm"]["error"] is None
- assert service._triage_dump._pending == {}
- def test_dump_laya_error_writes_laya_only_and_keeps_deltas(tmp_path):
- """A Laya predict failure finalizes the Laya half with `llm: null`.
- Nothing is dispatched, so no LLM half is ever recorded and the buffered
- deltas survive for the next settled turn.
- """
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, None)
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] is None
- assert data["laya"]["error"] == "predict failed"
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_without_brief_has_laya_only(tmp_path):
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.cfg.triage.extract_enabled = False
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_queue_full_has_laya_only(tmp_path):
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service._inference = InferenceQueue(size=2)
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- assert service._inference.depth() == 1
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert service._pending_brief == ["Unknown: hello"]
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_disabled_writes_nothing(tmp_path):
- service, emitted, clock, _, lane = _triage_service(
- brief_fn=lambda prompt: "RAW"
- )
- assert service._triage_dump.enabled is False
- service._triage_dump = TriageDumper(False, directory=str(tmp_path))
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- jobs = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(jobs) == 1
- jobs[0].run()
- assert _dump_files(tmp_path) == []
- def test_dump_stale_result_discards_without_file(tmp_path):
- service, emitted, clock, triage, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.toggle_chat()
- service.handle_event(_transcript(clock, "hello"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen = lane.submits[0]
- real_normalize = triage.normalize
- def stale_normalize(payload):
- # Simulate a clear landing between the first check/begin and the re-check.
- service._triage_generation += 1
- return real_normalize(payload)
- triage.normalize = stale_normalize
- service._on_triage_result(gen, _escalated_payload())
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- def test_dump_stale_brief_discards_without_file(tmp_path):
- service, _, clock, _, _ = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- ts = clock()
- service._triage_dump.begin(ts, laya={"request": {}, "response": {}, "error": None})
- service._pending_brief = ["Unknown: hello"]
- service._triage_generation += 1 # stale before the brief even starts
- service._run_brief_update(0, _escalated_payload(), ts, service._brief_token)
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- assert service._pending_brief == ["Unknown: hello"]
- def test_dump_brief_cleared_mid_flight_writes_nothing(tmp_path):
- service, _, clock, _, _ = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- def stale_brief(prompt, on_io=None):
- # Simulate a clear landing mid-LLM-call (after the request phase fired).
- service._triage_generation += 1
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {}})
- on_io({"phase": "response", "payload": {"c": 1}, "error": None})
- return "RAW"
- service._brief_fn = stale_brief
- service._pending_brief = ["Unknown: hello"]
- service._brief_outstanding = True
- token = service._brief_token
- ts = clock()
- service._triage_dump.begin(ts, laya={"request": {}, "response": {}, "error": None})
- service._run_brief_update(0, _escalated_payload(), ts, token)
- assert _dump_files(tmp_path) == []
- assert service._triage_dump._pending == {}
- assert service._pending_brief == []
- assert service._brief_outstanding is False
- def test_dump_gated_out_writes_laya_only_and_no_pending(tmp_path):
- """A turn the gate leaves out writes `llm: null` and leaks no pending record."""
- service, emitted, clock, triage, lane = _triage_service(
- triage=FakeTriage(escalate=True, gate=False), brief_fn=_dump_brief
- )
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- _run_triage_flow(service, emitted, clock, lane, _escalated_payload())
- assert [j.name for j in service._inference._queue] == ["chat"]
- assert len(triage.gate_calls) == 1
- files = _dump_files(tmp_path)
- assert len(files) == 1
- data = json.loads(files[0].read_text(encoding="utf-8"))
- assert data["laya"]["response"] == _escalated_payload()
- assert data["llm"] is None
- assert service._triage_dump._pending == {}
- def test_dump_coalesced_update_writes_laya_only_then_llm_only_flush(tmp_path):
- """A coalesced turn writes a Laya-only record; the later flush a separate llm-only one."""
- service, emitted, clock, _, lane = _triage_service(brief_fn=_dump_brief)
- service._triage_dump = TriageDumper(True, directory=str(tmp_path))
- service.cfg.triage.dump_io = True
- service.toggle_chat()
- service.handle_event(_transcript(clock, "first"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen0 = lane.submits[0]
- service._on_triage_result(gen0, _escalated_payload())
- service.handle_event(_transcript(clock, "second"))
- clock.advance(service.cfg.assistant.turn_gap_s + 0.1)
- service.tick()
- _, _, _, gen1 = lane.submits[1]
- service._on_triage_result(gen1, _escalated_payload())
- assert service._brief_outstanding is True
- appended: list[int] = []
- def _brief(prompt, on_io=None):
- if not appended:
- appended.append(1)
- service.handle_event(_transcript(clock, "during call"))
- if on_io is not None:
- on_io({"phase": "request", "url": "http://llm/v1", "body": {"prompt": prompt}})
- on_io({"phase": "response", "payload": {"content": "parsed"}, "error": None})
- return "BRIEF"
- service._brief_fn = _brief
- clock.advance(1.0)
- job = [j for j in service._inference._queue if j.name == "brief-update"][0]
- service._inference._queue.remove(job)
- job.run()
- flushed = [j for j in service._inference._queue if j.name == "brief-update"]
- assert len(flushed) == 1
- flushed[0].run()
- assert service._triage_dump._pending == {}
- records = [
- json.loads(path.read_text(encoding="utf-8")) for path in _dump_files(tmp_path)
- ]
- assert len(records) == 3
- coalesced = [r for r in records if r["laya"] is not None and r["llm"] is None]
- flushed_records = [r for r in records if r["laya"] is None and r["llm"] is not None]
- both = [r for r in records if r["laya"] is not None and r["llm"] is not None]
- assert len(coalesced) == 1
- assert len(flushed_records) == 1
- assert len(both) == 1
- assert coalesced[0]["ts"] != flushed_records[0]["ts"]
- assert flushed_records[0]["llm"]["request"]["body"] == {"prompt": "BRIEF-PROMPT"}
- def test_chat_client_failure_emits_llm_error_and_service_survives():
- """Adversarial (a): a chat client raising on `chat` surfaces as an `llm_error`
- event and the service keeps functioning afterwards."""
- from stt_daemon.assistant.llm import ConversationEngine, LlmError
- class BoomClient:
- def chat(self, messages, max_tokens=None, timeout_s=None, **kwargs):
- raise LlmError("boom")
- engine = ConversationEngine(Config().llm, BoomClient(), system_prompt="SYS")
- service, emitted, clock = _service()
- service.chat_fn = engine.send
- assert service.toggle_chat() is True
- service._run_chat("hello")
- errors = [e for e in emitted if e["type"] == "error" and e.get("code") == "llm_error"]
- assert errors and "boom" in errors[-1]["message"]
- assert service.toggle_chat() is True
- assert service.modes.mode.value == "idle"
|