queue-actions.e2e.ts 17 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351
  1. // Keyless browser coverage for pending queue actions through the shipped Web
  2. // composition and real HTTP/SSE wire. Replay overrides park consecutive turns
  3. // so the page can edit and remove exact occurrences, then stop the active turn
  4. // while proving the preserved Queue advances in FIFO order.
  5. import { existsSync } from 'node:fs'
  6. import { mkdtemp, readFile, rm, writeFile } from 'node:fs/promises'
  7. import { tmpdir } from 'node:os'
  8. import { fileURLToPath } from 'node:url'
  9. import { join } from 'node:path'
  10. import type { Browser, Page } from 'playwright'
  11. import { chromium } from 'playwright'
  12. import { afterEach, describe, expect, it, onTestFailed } from 'vitest'
  13. import { deriveReplayScript, parseSessionLog, type ReplayEntry } from '@deepseek-ai/dsh-llm-replay'
  14. import type { SessionEvent } from '@deepseek-ai/dsh-session'
  15. import {
  16. assertFixtureInventory, captureExpandedTurnProcessAria, captureStableAria, compareOrRefreshGolden,
  17. launchWebScaffold, watchConsole, webSnapshotMode, type WebScaffold,
  18. } from './scaffold.ts'
  19. import { connectFreshWorkspace, newEnglishPage, saveFailureShot } from './support.ts'
  20. const SNAPSHOT_DIR = fileURLToPath(new URL('../../../snapshots/web/queue-actions', import.meta.url))
  21. const FIXTURE = fileURLToPath(new URL('../../../snapshots/web/live-interactions/session.v2.jsonl', import.meta.url))
  22. const COLLAPSED_EXPECTED = join(SNAPSHOT_DIR, 'collapsed.expected.md')
  23. const EDITING_EXPECTED = join(SNAPSHOT_DIR, 'editing.expected.md')
  24. const LAYOUT_EXPECTED = join(SNAPSHOT_DIR, 'layout.expected.md')
  25. const PRESERVED_EXPECTED = join(SNAPSHOT_DIR, 'preserved.expected.md')
  26. const PRESERVED_EXPANDED_EXPECTED = join(SNAPSHOT_DIR, 'preserved-expanded.expected.md')
  27. const UI_EXPECTED = join(SNAPSHOT_DIR, 'ui.expected.md')
  28. const SENDING_EXPECTED = join(SNAPSHOT_DIR, 'sending.expected.md')
  29. const FAILED_EXPECTED = join(SNAPSHOT_DIR, 'failed.expected.md')
  30. const MODE = webSnapshotMode()
  31. const ACTIVE_PROMPT = 'Reply with a one-sentence description of event sourcing, then stop.'
  32. const REMOVE = 'Queue item to remove'
  33. const EDIT = 'Queue item to edit'
  34. const EDITED = 'Edited queue item'
  35. const TAIL = 'Queue item preserved after stop'
  36. const WAKE = 'Wake the preserved queue'
  37. const FAILED = 'Queue submission to retry'
  38. /** Durable turn-end classifications observed by the scenario. */
  39. function turnEndReasons(events: readonly SessionEvent[]): string[] {
  40. return events.flatMap(event => event.type === 'turn/end' ? [event.data.reason.kind] : [])
  41. }
  42. describe('web e2e: queue row actions', () => {
  43. let scaffold: WebScaffold | undefined
  44. let browser: Browser | undefined
  45. let page: Page
  46. let overrideDir: string | undefined
  47. afterEach(async () => {
  48. const failures: unknown[] = []
  49. await browser?.close().catch((error: unknown) => failures.push(error))
  50. browser = undefined
  51. const closing = scaffold
  52. scaffold = undefined
  53. await closing?.close().catch((error: unknown) => failures.push(error))
  54. if (overrideDir !== undefined) {
  55. await rm(overrideDir, { recursive: true, force: true })
  56. .catch((error: unknown) => failures.push(error))
  57. }
  58. overrideDir = undefined
  59. if (failures.length === 1) throw failures[0]
  60. if (failures.length > 1) throw new AggregateError(failures, 'queue-actions teardown failed')
  61. })
  62. it.skipIf(MODE === 'record')('edits and removes exact occurrences and preserves Queue across stop', async () => {
  63. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-queue-actions-'))
  64. const readyFile = join(overrideDir, '.hang-ready')
  65. const overridePath = join(overrideDir, 'replay.override.json')
  66. const recorded = deriveReplayScript(parseSessionLog(await readFile(FIXTURE, 'utf8')))
  67. expect(recorded).toHaveLength(1)
  68. const replay: ReplayEntry[] = [
  69. { kind: 'hang', readyFile },
  70. recorded[0]!,
  71. recorded[0]!,
  72. recorded[0]!,
  73. ]
  74. await writeFile(overridePath, JSON.stringify(replay))
  75. const sessionEvents: SessionEvent[] = []
  76. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  77. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  78. browser = await chromium.launch()
  79. page = await newEnglishPage(browser)
  80. const tripwire = watchConsole(page)
  81. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  82. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  83. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  84. onTestFailed(() => saveFailureShot(page, 'web-e2e-queue-actions'))
  85. const input = page.locator('[data-composer-input]').first()
  86. const firstSettled = scaffold.whenTurnSettled()
  87. await input.fill(ACTIVE_PROMPT)
  88. await input.press('Enter')
  89. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  90. const received = Promise.withResolvers<undefined>()
  91. const release = Promise.withResolvers<undefined>()
  92. await page.route('**/api/session/prompt', async (route) => {
  93. received.resolve(undefined)
  94. await release.promise
  95. await route.continue()
  96. }, { times: 1 })
  97. try {
  98. await input.fill(REMOVE)
  99. await input.press('Enter')
  100. await received.promise
  101. const pending = page.locator('[data-queue-dock] [data-submission-echo]')
  102. await pending.getByRole('status').waitFor()
  103. expect(await pending.getByRole('status').textContent()).toBe('Sending…')
  104. expect(await pending.getByRole('button').count()).toBe(3)
  105. expect(await pending.getByRole('button').evaluateAll(buttons =>
  106. buttons.every(button => (button as HTMLButtonElement).disabled))).toBe(true)
  107. expect(await input.textContent()).toBe('')
  108. expect(await input.getAttribute('contenteditable')).toBe('true')
  109. const sending = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  110. await compareOrRefreshGolden(SENDING_EXPECTED, sending, MODE)
  111. await page.setViewportSize({ width: 390, height: 1000 })
  112. await page.locator('[data-sidebar-collapsed="true"]').waitFor()
  113. await expect.poll(() => pending.evaluate(element => element.scrollWidth <= element.clientWidth)).toBe(true)
  114. await page.setViewportSize({ width: 1680, height: 1000 })
  115. await page.locator('[data-sidebar-collapsed="true"]').waitFor({ state: 'detached' })
  116. } finally {
  117. release.resolve(undefined)
  118. }
  119. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).isEnabled()).toBe(true)
  120. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  121. expect(await page.locator('[data-queue-dock]').getByRole('status').count()).toBe(0)
  122. await input.fill(EDIT)
  123. await input.press('Enter')
  124. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  125. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  126. .toBe('false')
  127. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  128. const collapsedSnapshot = await captureStableAria(
  129. page,
  130. '[class*="centerCol"]',
  131. scaffold.workspaceCwd,
  132. )
  133. await compareOrRefreshGolden(COLLAPSED_EXPECTED, collapsedSnapshot, MODE)
  134. await queueHeader.click()
  135. await expect.poll(
  136. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  137. { timeout: 10_000 },
  138. ).toBe(2)
  139. await page.setViewportSize({ width: 640, height: 1000 })
  140. const queueBox = await page.locator('[data-queue-dock]').boundingBox()
  141. const composerBox = await page.locator('[data-composer-card]').boundingBox()
  142. expect(queueBox).not.toBeNull()
  143. expect(composerBox).not.toBeNull()
  144. expect(queueBox!.x).toBeGreaterThanOrEqual(composerBox!.x)
  145. expect(queueBox!.x + queueBox!.width)
  146. .toBeLessThanOrEqual(composerBox!.x + composerBox!.width)
  147. const queueLeftInset = queueBox!.x - composerBox!.x
  148. const queueRightInset = composerBox!.x + composerBox!.width - queueBox!.x - queueBox!.width
  149. const composerMetrics = await page.locator('[data-composer-card]').evaluate((element) => {
  150. const style = getComputedStyle(element)
  151. return {
  152. dockInset: Number.parseFloat(style.getPropertyValue('--dsh-composer-dock-inset')),
  153. }
  154. })
  155. expect(queueLeftInset).toBeCloseTo(composerMetrics.dockInset, 1)
  156. expect(queueRightInset).toBeCloseTo(composerMetrics.dockInset, 1)
  157. await page.setViewportSize({ width: 1680, height: 1000 })
  158. const editRow = page.locator('[data-queue-dock] li', { hasText: EDIT })
  159. await editRow.getByRole('button', { name: 'Edit queued message' }).click()
  160. const editor = page.getByRole('textbox', { name: 'Edit queued message' })
  161. await editor.fill(EDITED)
  162. const editingSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  163. await compareOrRefreshGolden(EDITING_EXPECTED, editingSnapshot, MODE)
  164. await page.getByRole('button', { name: 'Save queued message' }).click()
  165. await page.getByText(EDITED, { exact: true }).waitFor()
  166. const removeRow = page.locator('[data-queue-dock] li', { hasText: REMOVE })
  167. await removeRow.getByRole('button', { name: 'Remove queued message' }).click()
  168. await expect.poll(() => page.getByText(REMOVE, { exact: true }).count()).toBe(0)
  169. const snapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  170. await compareOrRefreshGolden(UI_EXPECTED, snapshot, MODE)
  171. expect(sessionEvents.filter(event => event.type === 'user/message' && event.data.source.kind === 'user')).toHaveLength(1)
  172. await page.route('**/api/session/prompt', async (route) => {
  173. const envelope = route.request().postDataJSON() as { rpcId: string }
  174. await route.fulfill({
  175. status: 200,
  176. contentType: 'application/json',
  177. json: {
  178. type: 'server-response', rpcId: envelope.rpcId,
  179. result: {
  180. ok: false,
  181. error: { code: 'session/agent-busy', message: 'Queue submission failed', details: { reason: 'admission refused' } },
  182. },
  183. },
  184. })
  185. }, { times: 1 })
  186. await input.fill(FAILED)
  187. await input.press('Enter')
  188. const failure = page.getByRole('alert').filter({ hasText: 'Queue submission failed' })
  189. await failure.waitFor()
  190. await expect.poll(() => input.textContent()).toBe(FAILED)
  191. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  192. const failed = [
  193. await failure.ariaSnapshot(),
  194. await captureStableAria(page, '[data-composer-card]', scaffold.workspaceCwd),
  195. ].join('\n')
  196. await compareOrRefreshGolden(FAILED_EXPECTED, failed, MODE)
  197. await input.fill(TAIL)
  198. await input.press('Enter')
  199. await expect.poll(
  200. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  201. { timeout: 10_000 },
  202. ).toBe(2)
  203. await page.getByRole('button', { name: 'Stop generating' }).click()
  204. await firstSettled
  205. await expect.poll(() => page.getByRole('button', { name: 'Stop generating' }).count())
  206. .toBe(0)
  207. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).count())
  208. .toBe(2)
  209. const preservedSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  210. await compareOrRefreshGolden(PRESERVED_EXPECTED, preservedSnapshot, MODE)
  211. const expanded = await captureExpandedTurnProcessAria(
  212. page,
  213. '[class*="centerCol"]',
  214. scaffold.workspaceCwd,
  215. )
  216. await compareOrRefreshGolden(PRESERVED_EXPANDED_EXPECTED, expanded, MODE)
  217. const settled = scaffold.whenTurnSettled()
  218. await input.fill(WAKE)
  219. await input.press('Enter')
  220. await settled
  221. await expect.poll(() => turnEndReasons(sessionEvents), { timeout: 15_000 })
  222. .toEqual(['aborted', 'completed', 'completed', 'completed'])
  223. expect(sessionEvents.flatMap(event => event.type === 'user/message' && event.data.source.kind === 'user'
  224. ? event.data.content.flatMap(block => block.type === 'text' ? [block.text] : [])
  225. : [])).toEqual([ACTIVE_PROMPT, EDITED, TAIL, WAKE])
  226. await expect.poll(() => page.locator('[data-queue-dock]').count()).toBe(0)
  227. expect(tripwire.pageErrors).toEqual([])
  228. expect(tripwire.warnings).toEqual([])
  229. }, 120_000)
  230. it.skipIf(MODE === 'record')('orders Todo before Goal and Queue on one responsive card column', async () => {
  231. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-context-layout-'))
  232. const readyFile = join(overrideDir, '.hang-ready')
  233. const overridePath = join(overrideDir, 'replay.override.json')
  234. await writeFile(overridePath, JSON.stringify([{ kind: 'hang', readyFile } satisfies ReplayEntry]))
  235. const sessionEvents: SessionEvent[] = []
  236. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  237. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  238. browser = await chromium.launch()
  239. page = await newEnglishPage(browser)
  240. const tripwire = watchConsole(page)
  241. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  242. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  243. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  244. onTestFailed(() => saveFailureShot(page, 'web-e2e-context-layout'))
  245. const input = page.locator('[data-composer-input]').first()
  246. const settled = scaffold.whenTurnSettled()
  247. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  248. await input.fill('/goal Keep the composer context panels aligned')
  249. await input.press('Enter')
  250. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  251. await page.locator('[data-goal-bar]').waitFor({ timeout: 10_000 })
  252. const sessions = scaffold.ctx.sessions.list()
  253. expect(sessions).toHaveLength(1)
  254. sessions[0]!.append('todo/write', {
  255. todos: [
  256. { content: 'Confirm the panel order', status: 'completed' },
  257. { content: 'Align the panel widths', status: 'in_progress' },
  258. ],
  259. })
  260. await page.locator('[data-testid="todo-panel"]').waitFor({ timeout: 10_000 })
  261. for (const text of ['Layout queue first', 'Layout queue second']) {
  262. // A just-submitted composer is read-only for the prompt round-trip.
  263. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  264. await input.fill(text)
  265. await input.press('Enter')
  266. }
  267. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  268. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  269. .toBe('false')
  270. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  271. const layoutSnapshot = await captureStableAria(
  272. page,
  273. '[class*="centerCol"]',
  274. scaffold.workspaceCwd,
  275. )
  276. await compareOrRefreshGolden(LAYOUT_EXPECTED, layoutSnapshot, MODE)
  277. const expectAlignedContextPanels = async () => {
  278. // Sample one layout: the responsive grid can move between separate browser round trips.
  279. const [queuePanelBox, todoBox, goalBox] = await page.evaluate(() => [
  280. '[data-queue-dock] > div', '[data-testid="todo-panel"]', '[data-goal-bar] > div',
  281. ].map((selector) => {
  282. const box = document.querySelector(selector)?.getBoundingClientRect()
  283. return box === undefined ? null : { x: box.x, y: box.y, width: box.width }
  284. }))
  285. expect(queuePanelBox).not.toBeNull()
  286. expect(todoBox).not.toBeNull()
  287. expect(goalBox).not.toBeNull()
  288. expect(todoBox!.y).toBeLessThan(goalBox!.y)
  289. expect(goalBox!.y).toBeLessThan(queuePanelBox!.y)
  290. expect(todoBox!.x).toBeCloseTo(goalBox!.x, 1)
  291. expect(todoBox!.x).toBeCloseTo(queuePanelBox!.x, 1)
  292. expect(todoBox!.width).toBeCloseTo(goalBox!.width, 1)
  293. expect(todoBox!.width).toBeCloseTo(queuePanelBox!.width, 1)
  294. }
  295. await expectAlignedContextPanels()
  296. await page.setViewportSize({ width: 640, height: 1000 })
  297. await expectAlignedContextPanels()
  298. await page.setViewportSize({ width: 1680, height: 1000 })
  299. await queueHeader.click()
  300. const removeButtons = page.getByRole('button', { name: 'Remove queued message' })
  301. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(2)
  302. await removeButtons.first().click()
  303. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(1)
  304. await removeButtons.first().click()
  305. await expect.poll(() => page.locator('[data-queue-dock]').count(), { timeout: 10_000 }).toBe(0)
  306. await page.getByRole('button', { name: 'Clear goal' }).click()
  307. await expect.poll(() => page.locator('[data-goal-bar]').count(), { timeout: 10_000 }).toBe(0)
  308. await page.getByRole('button', { name: 'Stop generating' }).click()
  309. await settled
  310. expect(turnEndReasons(sessionEvents)).toEqual(['aborted'])
  311. expect(tripwire.pageErrors).toEqual([])
  312. expect(tripwire.warnings).toEqual([])
  313. }, 120_000)
  314. it.skipIf(MODE === 'record')('keeps its snapshot inventory closed', async () => {
  315. await assertFixtureInventory(
  316. SNAPSHOT_DIR,
  317. [
  318. 'collapsed.expected.md', 'editing.expected.md', 'layout.expected.md',
  319. 'preserved.expected.md', 'preserved-expanded.expected.md', 'ui.expected.md',
  320. 'sending.expected.md', 'failed.expected.md',
  321. ],
  322. )
  323. })
  324. })