d3ro-voice/apps/desktop/tests/main/services/voice-mode-redteam-r1-3.test.ts

// tests/main/services/voice-mode-redteam-r1-3.test.ts
// VoiceModeService 세션 런(SessionRun) 회귀 테스트:
//  - 취소된 세션의 늦은 STT 결과가 다음 세션에 삽입되지 않는다
//  - 지연 전사 중 두 번째 release 가 받아쓰기를 취소하지 않는다
//  - 오디오 캡처 참조는 세션당 정확히 한 번 놓는다 (start 대기 중 취소 포함)
//  - 녹음 중 캡처가 죽으면(장치 분리) 에러로 알린다
//  - 회의 모드가 소유한 자막은 자막 핫키로 끌 수 없다
//  - 표시 계층 이벤트(phase / insert-failed)

import { describe, it, expect, beforeEach, vi } from 'vitest'
import { EventEmitter } from 'events'
import { ErrorCode } from '@d3ro/core/errors'
import type { KeyBindingTriggerPayload } from '../../../src/main/services/KeyBindingService'

vi.mock('../../../src/main/services/LoggerService', () => ({
  getLogger: () => ({ info: vi.fn(), warn: vi.fn(), error: vi.fn(), debug: vi.fn() }),
}))

interface Deferred<T> {
  promise: Promise<T>
  resolve: (value: T) => void
  reject: (err: unknown) => void
}

function deferred<T>(): Deferred<T> {
  let resolve!: (value: T) => void
  let reject!: (err: unknown) => void
  const promise = new Promise<T>((res, rej) => {
    resolve = res
    reject = rej
  })
  return { promise, resolve, reject }
}

const result = (text: string): { text: string; segments: never[]; language: string; duration: number; processingTime: number } => ({
  text,
  segments: [],
  language: 'ko',
  duration: 1,
  processingTime: 1,
})

const mockSTT = vi.hoisted(() => ({
  initialize: vi.fn(),
  transcribe: vi.fn(),
  transcribePartial: vi.fn(async () => ''),
  getModels: vi.fn(() => [
    { id: 'base', name: 'Base', sizeBytes: 0, downloaded: true, languages: [], accuracy: 2, speed: 4 },
  ]),
  getStatus: vi.fn(() => ({ engineState: 'ready', activeModel: 'base', engineVersion: null, gpuAccelerated: false })),
}))

vi.mock('../../../src/main/services/LocalSTTService', () => ({
  getLocalSTTService: () => mockSTT,
  resetLocalSTTServiceForTests: () => undefined,
}))

const audioBus = new EventEmitter()
const mockAudio = vi.hoisted(() => ({
  start: vi.fn(),
  stop: vi.fn(),
  on: vi.fn(),
  off: vi.fn(),
}))

vi.mock('../../../src/main/services/AudioCaptureService', () => ({
  getAudioCaptureService: () => mockAudio,
}))

const keyBinding = vi.hoisted(() => ({
  handler: null as ((payload: KeyBindingTriggerPayload) => void) | null,
}))

vi.mock('../../../src/main/services/KeyBindingService', () => ({
  getKeyBindingService: () => ({
    on: (_ev: string, fn: (payload: KeyBindingTriggerPayload) => void) => {
      keyBinding.handler = fn
    },
    off: vi.fn(),
  }),
}))

const config = vi.hoisted(() => ({ values: {} as Record<string, unknown> }))
vi.mock('../../../src/main/services/ConfigService', () => ({
  configGet: vi.fn((key: string) => config.values[key]),
}))

const mockInsert = vi.hoisted(() => ({ insertText: vi.fn() }))
vi.mock('../../../src/main/services/TextInsertService', () => ({
  getTextInsertService: () => mockInsert,
}))

vi.mock('../../../src/main/services/LocalLLMService', () => ({
  getLocalLLMService: () => ({
    isAvailable: () => true,
    processText: vi.fn(async (text: string) => text),
    generate: vi.fn(),
    chatStream: vi.fn(),
    cancelGeneration: vi.fn(),
  }),
}))

const caption = vi.hoisted(() => ({
  state: 'inactive' as string,
  stop: vi.fn(),
  start: vi.fn(async () => undefined),
}))
vi.mock('../../../src/main/services/CaptionService', () => ({
  getCaptionService: () => ({
    getState: () => caption.state,
    stop: caption.stop,
    start: caption.start,
  }),
}))

vi.mock('../../../src/main/services/MeetingModeService', () => ({
  getMeetingModeService: () => ({ getState: () => 'recording' }),
}))

type VoiceModeModule = typeof import('../../../src/main/services/VoiceModeService')
let getVoiceModeService: VoiceModeModule['getVoiceModeService']

const SPEECH = Buffer.alloc(16000 * 2) // 1초

async function flush(times = 5): Promise<void> {
  for (let i = 0; i < times; i++) await new Promise((r) => setImmediate(r))
}

/** Date.now 를 앞으로 당겨 accidental-press(700ms) 판정을 건너뛴다 */
function advanceClock(ms: number): void {
  const base = Date.now()
  vi.spyOn(Date, 'now').mockReturnValue(base + ms)
}

beforeEach(async () => {
  vi.restoreAllMocks()
  vi.resetModules()
  vi.clearAllMocks()
  audioBus.removeAllListeners()
  config.values = { sttModelId: 'base', defaultLLMAction: 'none', autoInsert: true, llmBackend: 'local' }
  caption.state = 'inactive'
  keyBinding.handler = null
  // once 큐가 테스트 사이로 새지 않게 완전히 초기화한다
  for (const fn of [mockSTT.initialize, mockSTT.transcribe, mockSTT.transcribePartial, mockSTT.getModels, mockAudio.start, mockAudio.stop, mockInsert.insertText, caption.stop]) {
    fn.mockReset()
  }
  mockSTT.initialize.mockResolvedValue(undefined)
  mockSTT.transcribe.mockResolvedValue(result('기본 전사'))
  mockSTT.transcribePartial.mockResolvedValue('')
  mockSTT.getModels.mockReturnValue([
    { id: 'base', name: 'Base', sizeBytes: 0, downloaded: true, languages: [], accuracy: 2, speed: 4 },
  ])
  mockAudio.start.mockResolvedValue(undefined)
  mockAudio.stop.mockResolvedValue(undefined)
  mockAudio.on.mockImplementation((ev: string, fn: (...args: unknown[]) => void) => {
    audioBus.on(ev, fn)
  })
  mockAudio.off.mockImplementation((ev: string, fn: (...args: unknown[]) => void) => {
    audioBus.off(ev, fn)
  })
  mockInsert.insertText.mockResolvedValue({ success: true, method: 'clipboard', textLength: 1, durationMs: 1 })
  const mod = await import('../../../src/main/services/VoiceModeService')
  mod.resetVoiceModeServiceForTests()
  getVoiceModeService = mod.getVoiceModeService
})

describe('세션 런 — 늦게 도착한 결과', () => {
  it('취소된 세션 A 의 STT 결과는 새 세션 B 에 삽입되지 않는다', async () => {
    const svc = getVoiceModeService()
    const sttA = deferred<ReturnType<typeof result>>()
    mockSTT.transcribe.mockReturnValueOnce(sttA.promise)
    const completed: string[] = []
    svc.on('session-completed', ({ finalText }) => completed.push(finalText))

    // 세션 A: 녹음 → 정지 → 전사 진행 중
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    const stopA = svc.stopSession()
    await vi.waitFor(() => expect(mockSTT.transcribe).toHaveBeenCalledTimes(1))

    // A 취소 → B 시작
    svc.cancelSession()
    await svc.startSession('dictation')
    const sessionB = svc.currentSession
    expect(sessionB).not.toBeNull()

    // A 의 결과가 늦게 도착
    sttA.resolve(result('세션 A 텍스트'))
    await stopA
    await flush()

    expect(completed).toEqual([])
    expect(mockInsert.insertText).not.toHaveBeenCalled()
    expect(svc.currentSession?.id).toBe(sessionB?.id)
    expect(svc.isActive).toBe(true)
  })

  it('취소된 세션의 늦은 STT 실패는 새 세션을 에러로 끝내지 않는다', async () => {
    const svc = getVoiceModeService()
    const sttA = deferred<ReturnType<typeof result>>()
    mockSTT.transcribe.mockReturnValueOnce(sttA.promise)
    const errors: number[] = []
    svc.on('error', ({ error }) => errors.push(error.code))

    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    const stopA = svc.stopSession()
    await vi.waitFor(() => expect(mockSTT.transcribe).toHaveBeenCalledTimes(1))
    svc.cancelSession()
    await svc.startSession('dictation')

    sttA.reject(new Error('sidecar exploded'))
    await stopA
    await flush()

    expect(errors).toEqual([])
    expect(svc.isActive).toBe(true)
  })
})

describe('세션 런 — 정지 멱등성', () => {
  it('지연 전사 중 두 번째 정지는 받아쓰기를 취소하지 않는다', async () => {
    const init = deferred<void>()
    mockSTT.initialize.mockReturnValueOnce(init.promise)
    const stt = deferred<ReturnType<typeof result>>()
    mockSTT.transcribe.mockReturnValueOnce(stt.promise)

    const svc = getVoiceModeService()
    const cancelled: string[] = []
    const completed: string[] = []
    svc.on('session-cancelled', ({ reason }) => cancelled.push(reason))
    svc.on('session-completed', ({ finalText }) => completed.push(finalText))

    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession() // STT 미준비 → 대기

    init.resolve() // 모델 로딩 완료 → 지연 전사 시작 (버퍼 비움)
    await vi.waitFor(() => expect(mockSTT.transcribe).toHaveBeenCalledTimes(1))

    await svc.stopSession() // 두 번째 release
    expect(cancelled).toEqual([])

    stt.resolve(result('살아남은 받아쓰기'))
    await vi.waitFor(() => expect(completed).toEqual(['살아남은 받아쓰기']))
    expect(mockInsert.insertText).toHaveBeenCalledWith('살아남은 받아쓰기', undefined)
  })
})

describe('세션 런 — 오디오 참조 소유', () => {
  it('정지 후 "No speech" 에러가 나도 audio.stop 은 세션당 한 번이다', async () => {
    mockSTT.transcribe.mockResolvedValueOnce(result('   '))
    const svc = getVoiceModeService()
    const errors: number[] = []
    svc.on('error', ({ error }) => errors.push(error.code))

    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    await vi.waitFor(() => expect(errors).toEqual([ErrorCode.STTNoAudioData]))
    expect(mockAudio.start).toHaveBeenCalledTimes(1)
    expect(mockAudio.stop).toHaveBeenCalledTimes(1)
  })

  it('audio.start 대기 중 취소돼도 늦게 잡힌 참조를 정확히 한 번 놓는다', async () => {
    const start = deferred<void>()
    mockAudio.start.mockReturnValueOnce(start.promise)
    const svc = getVoiceModeService()

    const starting = svc.startSession('dictation')
    await vi.waitFor(() => expect(mockAudio.start).toHaveBeenCalledTimes(1))
    svc.cancelSession()
    expect(mockAudio.stop).not.toHaveBeenCalled()

    start.resolve()
    await starting
    await flush()
    expect(mockAudio.stop).toHaveBeenCalledTimes(1)
    expect(audioBus.listenerCount('audio-data')).toBe(0)
  })

  it('세션이 끝나면 오디오 핸들러가 모두 해제된다 (다음 세션 프레임 중복 없음)', async () => {
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    await vi.waitFor(() => expect(svc.isActive).toBe(false))

    await svc.startSession('dictation')
    expect(audioBus.listenerCount('audio-data')).toBe(1)
    expect(audioBus.listenerCount('stopped')).toBe(1)
  })
})

describe('캡처 손실', () => {
  it('오디오 없이 캡처가 죽으면 AudioCaptureFailed 에러를 낸다', async () => {
    const svc = getVoiceModeService()
    const errors: number[] = []
    svc.on('error', ({ error }) => errors.push(error.code))

    await svc.startSession('dictation')
    audioBus.emit('stopped', { reason: 'device-lost' })
    await flush()

    expect(errors).toEqual([ErrorCode.AudioCaptureFailed])
    expect(svc.isActive).toBe(false)
    // 캡처 쪽이 이미 정리했으므로 다른 소비자의 참조를 깎지 않는다
    expect(mockAudio.stop).not.toHaveBeenCalled()
  })

  it("수동 정지('manual')는 캡처 손실로 보지 않는다", async () => {
    const svc = getVoiceModeService()
    const errors: number[] = []
    svc.on('error', ({ error }) => errors.push(error.code))
    await svc.startSession('dictation')
    audioBus.emit('stopped', { reason: 'manual' })
    await flush()
    expect(errors).toEqual([])
    expect(svc.isActive).toBe(true)
  })
})

describe('표시 이벤트', () => {
  it("모델 확인을 통과한 뒤에만 'recording' phase 를 내보낸다", async () => {
    mockSTT.getModels.mockReturnValueOnce([
      { id: 'base', name: 'Base', sizeBytes: 0, downloaded: false, languages: [], accuracy: 2, speed: 4 },
    ])
    const svc = getVoiceModeService()
    const phases: string[] = []
    const errors: number[] = []
    svc.on('phase', ({ phase }) => phases.push(phase))
    svc.on('error', ({ error }) => errors.push(error.code))

    await svc.startSession('dictation')

    expect(phases).toEqual([])
    expect(errors).toEqual([ErrorCode.STTModelNotFound])
  })

  it("녹음 → 정지에서 'recording' 다음 'thinking' phase 가 나간다", async () => {
    const svc = getVoiceModeService()
    const phases: string[] = []
    svc.on('phase', ({ phase }) => phases.push(phase))
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    expect(phases).toEqual(['recording', 'thinking'])
  })

  it('자동 삽입이 실패하면 insert-failed 를 내보낸다', async () => {
    mockInsert.insertText.mockRejectedValueOnce(new Error('clipboard busy'))
    const svc = getVoiceModeService()
    const failed: string[] = []
    svc.on('insert-failed', ({ text }) => failed.push(text))

    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    await vi.waitFor(() => expect(failed).toEqual(['기본 전사']))
  })
})

describe('자막 핫키', () => {
  it('회의 모드가 소유한 자막 세션은 자막 핫키로 멈추지 않는다', async () => {
    caption.state = 'active'
    caption.stop.mockResolvedValue(false)
    const svc = getVoiceModeService()
    svc.connectKeyBindings()
    expect(keyBinding.handler).not.toBeNull()

    keyBinding.handler?.({
      actionId: 'caption',
      type: 'pressed',
      timestamp: Date.now(),
      isDoublePress: false,
      holdMode: false,
    } as KeyBindingTriggerPayload)
    // 소유자('user')를 밝혀 요청하므로 CaptionService 가 거부할 수 있다
    await vi.waitFor(() => expect(caption.stop).toHaveBeenCalledWith('user'))
    expect(caption.start).not.toHaveBeenCalled()
  })
})