d3ro-voice/apps/desktop/tests/main/services/voice-mode-redteam-r2-7.test.ts

// tests/main/services/voice-mode-redteam-r2-7.test.ts
// VoiceModeService 회귀 테스트 (round 2 #7):
//  - 더블탭으로 핸즈프리를 끄면 두 번째 탭이 처리 뒤 새 녹음을 시작하지 않는다 (큐가 막히지 않음)
//  - 스크린 컨텍스트의 창 조회가 세션·마이크 시작을 막지 않는다
//  - 원문 삽입 경로(none)에선 선택 텍스트용 복사 단축키를 보내지 않는다
//  - '체인' 기본 액션에서도 음성 단축키가 지목한 지시문이 이긴다
//  - 정지 시 캡처가 흘려보내는 꼬리 오디오를 받는다

import { describe, it, expect, beforeEach, vi } from 'vitest'
import { EventEmitter } from 'events'
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
}

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

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

const gateway = vi.hoisted(() => ({ processText: vi.fn() }))
vi.mock('../../../src/main/services/llm/LlmGateway', () => ({
  getLlmGateway: () => gateway,
}))

vi.mock('../../../src/main/services/CaptionService', () => ({
  getCaptionService: () => ({ getState: () => 'inactive', stop: vi.fn(), start: vi.fn() }),
}))

const chain = vi.hoisted(() => ({ execute: vi.fn() }))
vi.mock('../../../src/main/services/ChainService', () => ({
  getChainService: () => chain,
}))

const instructions = vi.hoisted(() => ({
  byId: {} as Record<string, { id: string; name: string; prompt: string }>,
}))
vi.mock('../../../src/main/services/CustomInstructionService', () => ({
  getCustomInstructionService: () => ({ getById: (id: string) => instructions.byId[id] ?? null }),
}))

const voiceCommand = vi.hoisted(() => ({ instructionId: null as string | null }))
vi.mock('../../../src/main/services/VoiceCommandService', () => ({
  getVoiceCommandService: () => ({
    isEnabled: () => voiceCommand.instructionId !== null,
    match: (text: string) =>
      voiceCommand.instructionId
        ? { matched: true, ruleId: 'r', instructionId: voiceCommand.instructionId, cleanedText: text, matchedKeyword: 'k' }
        : { matched: false, ruleId: null, instructionId: null, cleanedText: text, matchedKeyword: null },
  }),
}))

const screen = vi.hoisted(() => ({
  enabled: false,
  captureContext: vi.fn(),
  captureSelectedText: vi.fn(),
}))
vi.mock('../../../src/main/services/ScreenContextService', () => ({
  getScreenContextService: () => ({
    isEnabled: () => screen.enabled,
    captureContext: screen.captureContext,
    captureSelectedText: screen.captureSelectedText,
    buildContextPrompt: (ctx: { appName: string | null; selectedText: string | null }) =>
      `[ctx app=${ctx.appName ?? '-'} sel=${ctx.selectedText ?? '-'}]\n`,
  }),
}))

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

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

async function flush(times = 8): Promise<void> {
  for (let i = 0; i < times; i++) await new Promise((r) => setImmediate(r))
}

function advanceClock(ms: number): void {
  const base = Date.now()
  vi.spyOn(Date, 'now').mockReturnValue(base + ms)
}

function key(
  actionId: 'dictation' | 'hands-free',
  type: 'pressed' | 'released',
  isDoublePress: boolean,
): KeyBindingTriggerPayload {
  return {
    actionId,
    type,
    isDoublePress,
    holdMode: actionId === 'dictation',
    timestamp: Date.now(),
    durationMs: 0,
  } as unknown as KeyBindingTriggerPayload
}

function screenContext(appName: string): { context: { appName: string; windowTitle: string; selectedText: null; capturedAt: number }; selectedTextAttempted: boolean } {
  return { context: { appName, windowTitle: 'w', selectedText: null, capturedAt: 0 }, selectedTextAttempted: false }
}

beforeEach(async () => {
  vi.restoreAllMocks()
  vi.resetModules()
  audioBus.removeAllListeners()
  config.values = { sttModelId: 'base', defaultLLMAction: 'none', autoInsert: true, llmBackend: 'local' }
  keyBinding.handler = null
  screen.enabled = false
  voiceCommand.instructionId = null
  instructions.byId = {}
  for (const fn of [
    mockSTT.initialize, mockSTT.transcribe, mockSTT.transcribePartial, mockSTT.getModels,
    mockAudio.start, mockAudio.stop, mockAudio.on, mockAudio.off, mockInsert.insertText,
    gateway.processText, chain.execute, screen.captureContext, screen.captureSelectedText,
  ]) {
    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 })
  gateway.processText.mockImplementation(async (text: string) => `LLM(${text})`)
  chain.execute.mockResolvedValue({ finalText: '체인 결과' })
  screen.captureContext.mockResolvedValue(screenContext('Code'))
  screen.captureSelectedText.mockResolvedValue('선택 문장')
  const mod = await import('../../../src/main/services/VoiceModeService')
  mod.resetVoiceModeServiceForTests()
  getVoiceModeService = mod.getVoiceModeService
})

describe('더블탭 핸즈프리 정지', () => {
  it('정지 더블탭의 두 번째 탭은 전사가 끝난 뒤 새 핸즈프리 녹음을 시작하지 않는다', async () => {
    const stt = deferred<SttResult>()
    mockSTT.transcribe.mockReturnValueOnce(stt.promise)
    const svc = getVoiceModeService()
    svc.connectKeyBindings()
    const completed: string[] = []
    svc.on('session-completed', ({ finalText }) => completed.push(finalText))
    const emit = (p: KeyBindingTriggerPayload): void => keyBinding.handler?.(p)

    // 더블탭으로 핸즈프리 켜기
    emit(key('hands-free', 'pressed', true))
    emit(key('hands-free', 'released', true))
    await vi.waitFor(() => expect(mockAudio.start).toHaveBeenCalledTimes(1))
    await vi.waitFor(() => expect(svc.isActive).toBe(true))
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(3000)

    // 더블탭으로 끄기: 첫 탭(dictation press/release) → 정지, 두 번째 탭(hands-free press)
    emit(key('dictation', 'pressed', false))
    emit(key('dictation', 'released', false))
    await vi.waitFor(() => expect(mockSTT.transcribe).toHaveBeenCalledTimes(1))
    emit(key('hands-free', 'pressed', true))
    emit(key('hands-free', 'released', true))
    await flush()

    // 전사 완료
    stt.resolve(result('핸즈프리 받아쓰기'))
    await vi.waitFor(() => expect(completed).toEqual(['핸즈프리 받아쓰기']))
    await flush()

    expect(mockAudio.start).toHaveBeenCalledTimes(1)
    expect(svc.isActive).toBe(false)
  })

  it('공개 stopSession() 은 여전히 전사·삽입이 끝날 때까지 기다린다', async () => {
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    expect(mockSTT.transcribe).toHaveBeenCalledTimes(1)
    expect(svc.isActive).toBe(false)
  })
})

describe('스크린 컨텍스트 — 창 조회가 세션을 막지 않는다', () => {
  it('창 조회가 끝나기 전에 세션과 마이크가 시작되고, 결과는 LLM 입력에 합쳐진다', async () => {
    screen.enabled = true
    config.values.defaultLLMAction = 'refine'
    const probe = deferred<ReturnType<typeof screenContext>>()
    screen.captureContext.mockReturnValueOnce(probe.promise)
    const svc = getVoiceModeService()
    const started: string[] = []
    svc.on('session-started', ({ session }) => started.push(session.id))

    const starting = svc.startSession('dictation')
    await Promise.race([starting, new Promise((r) => setTimeout(r, 300))])
    expect(started).toHaveLength(1)
    expect(mockAudio.start).toHaveBeenCalledTimes(1)
    expect(screen.captureContext).toHaveBeenCalledWith(false)

    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    const stopping = svc.stopSession()
    probe.resolve(screenContext('Code'))
    await stopping
    await starting

    expect(gateway.processText).toHaveBeenCalledTimes(1)
    const [text, action] = gateway.processText.mock.calls[0] as [string, string]
    expect(action).toBe('refine')
    expect(text).toBe('[ctx app=Code sel=선택 문장]\n기본 전사')
  })
})

describe('스크린 컨텍스트 — 선택 텍스트 캡처', () => {
  it("원문 삽입(none) 경로에선 복사 단축키(선택 텍스트 캡처)를 보내지 않는다", async () => {
    screen.enabled = true
    config.values.defaultLLMAction = 'none'
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    await flush()

    expect(mockInsert.insertText).toHaveBeenCalledWith('기본 전사', undefined)
    expect(screen.captureSelectedText).not.toHaveBeenCalled()
  })

  it('LLM 경로면 정지 시점에 선택 텍스트를 캡처한다', async () => {
    screen.enabled = true
    config.values.defaultLLMAction = 'refine'
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()
    expect(screen.captureSelectedText).toHaveBeenCalledTimes(1)
  })

  it("none 이어도 음성 단축키로 LLM 경로가 되면 그때 선택 텍스트를 캡처해 쓴다", async () => {
    screen.enabled = true
    config.values.defaultLLMAction = 'none'
    voiceCommand.instructionId = 'i1'
    instructions.byId.i1 = { id: 'i1', name: '요약', prompt: '요약해줘' }
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()

    expect(screen.captureSelectedText).toHaveBeenCalledTimes(1)
    expect(gateway.processText).toHaveBeenCalledTimes(1)
    const [text, action] = gateway.processText.mock.calls[0] as [string, string]
    expect(action).toBe('custom')
    expect(text).toContain('[ctx app=Code sel=선택 문장]')
  })
})

describe('LLM 경로 — 체인 + 음성 단축키', () => {
  it('체인 기본 액션에서도 음성 단축키가 지목한 지시문이 이긴다', async () => {
    config.values.defaultLLMAction = 'chain'
    config.values.activeChainId = 'c1'
    voiceCommand.instructionId = 'i1'
    instructions.byId.i1 = { id: 'i1', name: '요약', prompt: '요약해줘' }
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()

    expect(chain.execute).not.toHaveBeenCalled()
    expect(gateway.processText).toHaveBeenCalledTimes(1)
    const [, action, , systemPrompt] = gateway.processText.mock.calls[0] as [string, string, unknown, string]
    expect(action).toBe('custom')
    expect(systemPrompt).toContain('요약해줘')
  })

  it('음성 단축키가 없으면 활성 체인을 실행한다', async () => {
    config.values.defaultLLMAction = 'chain'
    config.values.activeChainId = 'c1'
    const svc = getVoiceModeService()
    const completed: string[] = []
    svc.on('session-completed', ({ finalText }) => completed.push(finalText))
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()

    expect(chain.execute).toHaveBeenCalledWith('c1', '기본 전사')
    expect(gateway.processText).not.toHaveBeenCalled()
    expect(completed).toEqual(['체인 결과'])
  })
})

describe('정지 시 꼬리 오디오', () => {
  it('캡처가 정지 중에 흘려보낸 꼬리 오디오까지 전사에 넣는다', async () => {
    const TAIL = Buffer.alloc(3200) // 100ms
    mockAudio.stop.mockImplementation(async () => {
      // AudioCaptureService.stop() 은 SoX 를 죽이기 전에 남은 오디오를 emit 한다
      audioBus.emit('audio-data', { buffer: TAIL })
    })
    const svc = getVoiceModeService()
    await svc.startSession('dictation')
    audioBus.emit('audio-data', { buffer: SPEECH })
    advanceClock(1000)
    await svc.stopSession()

    expect(mockSTT.transcribe).toHaveBeenCalledTimes(1)
    const [audio] = mockSTT.transcribe.mock.calls[0] as [Buffer]
    expect(audio.length).toBe(SPEECH.length + TAIL.length)
    expect(audioBus.listenerCount('audio-data')).toBe(0)
  })
})