queue-actions.e2e.ts 20 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404
  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, vi } 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.v3.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 WRITER_HELD_EXPECTED = join(SNAPSHOT_DIR, 'writer-held.expected.md')
  30. const FAILED_EXPECTED = join(SNAPSHOT_DIR, 'failed.expected.md')
  31. const MODE = webSnapshotMode()
  32. const ACTIVE_PROMPT = 'Reply with a one-sentence description of event sourcing, then stop.'
  33. const REMOVE = 'Queue item to remove'
  34. const EDIT = 'Queue item to edit'
  35. const EDITED = 'Edited queue item'
  36. const TAIL = 'Queue item preserved after stop'
  37. const WAKE = 'Wake the preserved queue'
  38. const FAILED = 'Queue submission to retry'
  39. /** Durable turn-end classifications observed by the scenario. */
  40. function turnEndReasons(events: readonly SessionEvent[]): string[] {
  41. return events.flatMap(event => event.type === 'turn/end' ? [event.data.reason.kind] : [])
  42. }
  43. describe('web e2e: queue row actions', () => {
  44. let scaffold: WebScaffold | undefined
  45. let browser: Browser | undefined
  46. let page: Page
  47. let overrideDir: string | undefined
  48. afterEach(async () => {
  49. const failures: unknown[] = []
  50. await browser?.close().catch((error: unknown) => failures.push(error))
  51. browser = undefined
  52. const closing = scaffold
  53. scaffold = undefined
  54. await closing?.close().catch((error: unknown) => failures.push(error))
  55. if (overrideDir !== undefined) {
  56. await rm(overrideDir, { recursive: true, force: true })
  57. .catch((error: unknown) => failures.push(error))
  58. }
  59. overrideDir = undefined
  60. if (failures.length === 1) throw failures[0]
  61. if (failures.length > 1) throw new AggregateError(failures, 'queue-actions teardown failed')
  62. })
  63. /** Wait for the exact queue mutation response before observing its unlocked actions. */
  64. async function settleQueueAction(action: () => Promise<void>, remainingText: string): Promise<void> {
  65. const response = page.waitForResponse('**/api/session/updateQueue')
  66. await action()
  67. expect((await response).ok()).toBe(true)
  68. const row = page.locator('[data-queue-dock] li', { hasText: remainingText })
  69. await expect.poll(() => row.getByRole('button', { name: 'Edit queued message' }).isEnabled()).toBe(true)
  70. await expect.poll(() => row.getByRole('button', { name: 'Remove queued message' }).isEnabled()).toBe(true)
  71. }
  72. it.skipIf(MODE === 'record')('edits and removes exact occurrences and preserves Queue across stop', async () => {
  73. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-queue-actions-'))
  74. const readyFile = join(overrideDir, '.hang-ready')
  75. const overridePath = join(overrideDir, 'replay.override.json')
  76. const recorded = deriveReplayScript(parseSessionLog(await readFile(FIXTURE, 'utf8')))
  77. expect(recorded).toHaveLength(1)
  78. const replay: ReplayEntry[] = [
  79. { kind: 'hang', readyFile },
  80. recorded[0]!,
  81. recorded[0]!,
  82. recorded[0]!,
  83. ]
  84. await writeFile(overridePath, JSON.stringify(replay))
  85. const sessionEvents: SessionEvent[] = []
  86. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  87. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  88. browser = await chromium.launch()
  89. page = await newEnglishPage(browser)
  90. const tripwire = watchConsole(page)
  91. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  92. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  93. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  94. onTestFailed(() => saveFailureShot(page, 'web-e2e-queue-actions'))
  95. const input = page.locator('[data-composer-input]').first()
  96. const firstSettled = scaffold.whenTurnSettled()
  97. await input.fill(ACTIVE_PROMPT)
  98. await input.press('Enter')
  99. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  100. const admitted = page.waitForResponse('**/api/session/prompt')
  101. const received = Promise.withResolvers<undefined>()
  102. const release = Promise.withResolvers<undefined>()
  103. await page.route('**/api/session/prompt', async (route) => {
  104. received.resolve(undefined)
  105. await release.promise
  106. await route.continue()
  107. }, { times: 1 })
  108. try {
  109. await input.fill(REMOVE)
  110. await input.press('Enter')
  111. await received.promise
  112. const pending = page.locator('[data-queue-dock] [data-submission-echo]')
  113. await pending.getByRole('status').waitFor()
  114. expect(await pending.getByRole('status').textContent()).toBe('Sending…')
  115. expect(await pending.getByRole('button').count()).toBe(3)
  116. expect(await pending.getByRole('button').evaluateAll(buttons =>
  117. buttons.every(button => (button as HTMLButtonElement).disabled))).toBe(true)
  118. expect(await input.textContent()).toBe('')
  119. expect(await input.getAttribute('contenteditable')).toBe('true')
  120. const sending = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  121. await compareOrRefreshGolden(SENDING_EXPECTED, sending, MODE)
  122. await page.setViewportSize({ width: 390, height: 1000 })
  123. await page.locator('[data-sidebar-collapsed="true"]').waitFor()
  124. await expect.poll(() => pending.evaluate(element => element.scrollWidth <= element.clientWidth)).toBe(true)
  125. await page.setViewportSize({ width: 1680, height: 1000 })
  126. await page.locator('[data-sidebar-collapsed="true"]').waitFor({ state: 'detached' })
  127. } finally {
  128. release.resolve(undefined)
  129. }
  130. expect((await admitted).ok()).toBe(true)
  131. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).isEnabled()).toBe(true)
  132. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  133. expect(await page.locator('[data-queue-dock]').getByRole('status').count()).toBe(0)
  134. await input.fill(EDIT)
  135. await input.press('Enter')
  136. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  137. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  138. .toBe('false')
  139. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  140. const collapsedSnapshot = await captureStableAria(
  141. page,
  142. '[class*="centerCol"]',
  143. scaffold.workspaceCwd,
  144. )
  145. await compareOrRefreshGolden(COLLAPSED_EXPECTED, collapsedSnapshot, MODE)
  146. await queueHeader.click()
  147. await expect.poll(
  148. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  149. { timeout: 10_000 },
  150. ).toBe(2)
  151. await page.setViewportSize({ width: 640, height: 1000 })
  152. await page.locator('[data-sidebar-collapsed="true"]').waitFor()
  153. // The responsive sidebar and composer settle independently; sample both
  154. // rectangles in one browser task so the comparison uses one layout.
  155. await vi.waitFor(async () => {
  156. const metrics = await page.evaluate(() => {
  157. const queue = document.querySelector('[data-queue-dock]')
  158. const composer = document.querySelector('[data-composer-card]')
  159. if (queue === null || composer === null) return undefined
  160. const queueBox = queue.getBoundingClientRect()
  161. const composerBox = composer.getBoundingClientRect()
  162. return {
  163. leftInset: queueBox.left - composerBox.left,
  164. rightInset: composerBox.right - queueBox.right,
  165. dockInset: Number.parseFloat(getComputedStyle(composer).getPropertyValue('--dsh-composer-dock-inset')),
  166. }
  167. })
  168. expect(metrics).toBeDefined()
  169. expect(metrics!.leftInset).toBeGreaterThanOrEqual(0)
  170. expect(metrics!.rightInset).toBeGreaterThanOrEqual(0)
  171. expect(metrics!.leftInset).toBeCloseTo(metrics!.dockInset, 1)
  172. expect(metrics!.rightInset).toBeCloseTo(metrics!.dockInset, 1)
  173. }, { timeout: 10_000 })
  174. await page.setViewportSize({ width: 1680, height: 1000 })
  175. const editRow = page.locator('[data-queue-dock] li', { hasText: EDIT })
  176. await editRow.getByRole('button', { name: 'Edit queued message' }).click()
  177. const editor = page.getByRole('textbox', { name: 'Edit queued message' })
  178. await editor.fill(EDITED)
  179. await page.getByRole('button', { name: 'Save queued message' }).hover()
  180. await page.getByRole('tooltip', { name: 'Save queued message', exact: true }).waitFor()
  181. const editingSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  182. await compareOrRefreshGolden(EDITING_EXPECTED, editingSnapshot, MODE)
  183. await settleQueueAction(() => page.getByRole('button', { name: 'Save queued message' }).click(), EDITED)
  184. await page.getByText(EDITED, { exact: true }).waitFor()
  185. const removeRow = page.locator('[data-queue-dock] li', { hasText: REMOVE })
  186. await settleQueueAction(() => removeRow.getByRole('button', { name: 'Remove queued message' }).click(), EDITED)
  187. await expect.poll(() => page.getByText(REMOVE, { exact: true }).count()).toBe(0)
  188. // The queue stream can remove the row before the mutation reply clears busy.
  189. const remainingEdit = page.getByRole('button', { name: 'Edit queued message', exact: true })
  190. await expect.poll(() => remainingEdit.isEnabled(), { timeout: 10_000 }).toBe(true)
  191. await remainingEdit.hover()
  192. await page.getByRole('tooltip', { name: 'Edit queued message', exact: true }).waitFor()
  193. const snapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  194. await compareOrRefreshGolden(UI_EXPECTED, snapshot, MODE)
  195. expect(sessionEvents.filter(event => event.type === 'user/message' && event.data.source.kind === 'user')).toHaveLength(1)
  196. await page.route('**/api/session/prompt', async (route) => {
  197. const envelope = route.request().postDataJSON() as { rpcId: string }
  198. await route.fulfill({
  199. status: 200,
  200. contentType: 'application/json',
  201. json: {
  202. type: 'server-response', rpcId: envelope.rpcId,
  203. result: {
  204. ok: false,
  205. error: { code: 'session/agent-busy', message: 'Queue submission failed', details: { reason: 'admission refused' } },
  206. },
  207. },
  208. })
  209. }, { times: 1 })
  210. await input.fill(FAILED)
  211. await input.press('Enter')
  212. const failure = page.getByRole('alert').filter({ hasText: 'Queue submission failed' })
  213. await failure.waitFor()
  214. await expect.poll(() => input.textContent()).toBe(FAILED)
  215. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  216. const failed = [
  217. await failure.ariaSnapshot(),
  218. await captureStableAria(page, '[data-composer-card]', scaffold.workspaceCwd),
  219. ].join('\n')
  220. await compareOrRefreshGolden(FAILED_EXPECTED, failed, MODE)
  221. await page.route('**/api/session/prompt', async (route) => {
  222. const envelope = route.request().postDataJSON() as { rpcId: string }
  223. await route.fulfill({
  224. status: 200,
  225. contentType: 'application/json',
  226. json: {
  227. type: 'server-response', rpcId: envelope.rpcId,
  228. result: {
  229. ok: false,
  230. error: { code: 'session/writer-held', message: 'writer held', details: { sessionId: 'held-session' } },
  231. },
  232. },
  233. })
  234. }, { times: 1 })
  235. await input.press('Enter')
  236. const writerHeld = page.getByRole('alert').filter({ hasText: 'This session is already in use' })
  237. await writerHeld.waitFor()
  238. await expect.poll(() => input.textContent()).toBe(FAILED)
  239. await compareOrRefreshGolden(WRITER_HELD_EXPECTED, [
  240. await writerHeld.ariaSnapshot(),
  241. await captureStableAria(page, '[data-composer-card]', scaffold.workspaceCwd),
  242. ].join('\n'), MODE)
  243. await input.fill(TAIL)
  244. await input.press('Enter')
  245. await expect.poll(
  246. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  247. { timeout: 10_000 },
  248. ).toBe(2)
  249. const stopButton = page.getByRole('button', { name: 'Stop generating' })
  250. await stopButton.hover()
  251. await page.getByRole('tooltip', { name: 'Stop generating', exact: true }).waitFor()
  252. await stopButton.click()
  253. await firstSettled
  254. await expect.poll(() => page.getByRole('button', { name: 'Stop generating' }).count())
  255. .toBe(0)
  256. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).count())
  257. .toBe(2)
  258. // The disabled Send button must dismiss the active Stop tooltip without mouseleave.
  259. await expect.poll(() => page.getByRole('tooltip').count()).toBe(0)
  260. const preservedSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  261. await compareOrRefreshGolden(PRESERVED_EXPECTED, preservedSnapshot, MODE)
  262. const expanded = await captureExpandedTurnProcessAria(
  263. page,
  264. '[class*="centerCol"]',
  265. scaffold.workspaceCwd,
  266. )
  267. await compareOrRefreshGolden(PRESERVED_EXPANDED_EXPECTED, expanded, MODE)
  268. const settled = scaffold.whenTurnSettled()
  269. await input.fill(WAKE)
  270. await input.press('Enter')
  271. await settled
  272. await expect.poll(() => turnEndReasons(sessionEvents), { timeout: 15_000 })
  273. .toEqual(['aborted', 'completed', 'completed', 'completed'])
  274. expect(sessionEvents.flatMap(event => event.type === 'user/message' && event.data.source.kind === 'user'
  275. ? event.data.content.flatMap(block => block.type === 'text' ? [block.text] : [])
  276. : [])).toEqual([ACTIVE_PROMPT, EDITED, TAIL, WAKE])
  277. await expect.poll(() => page.locator('[data-queue-dock]').count()).toBe(0)
  278. expect(tripwire.pageErrors).toEqual([])
  279. expect(tripwire.warnings).toEqual([])
  280. }, 120_000)
  281. it.skipIf(MODE === 'record')('orders Todo before Goal and Queue on one responsive card column', async () => {
  282. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-context-layout-'))
  283. const readyFile = join(overrideDir, '.hang-ready')
  284. const overridePath = join(overrideDir, 'replay.override.json')
  285. await writeFile(overridePath, JSON.stringify([{ kind: 'hang', readyFile } satisfies ReplayEntry]))
  286. const sessionEvents: SessionEvent[] = []
  287. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  288. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  289. browser = await chromium.launch()
  290. page = await newEnglishPage(browser)
  291. const tripwire = watchConsole(page)
  292. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  293. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  294. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  295. onTestFailed(() => saveFailureShot(page, 'web-e2e-context-layout'))
  296. const input = page.locator('[data-composer-input]').first()
  297. const settled = scaffold.whenTurnSettled()
  298. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  299. await input.fill('/goal Keep the composer context panels aligned')
  300. await input.press('Enter')
  301. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  302. await page.locator('[data-goal-bar]').waitFor({ timeout: 10_000 })
  303. const sessions = scaffold.ctx.sessions.list()
  304. expect(sessions).toHaveLength(1)
  305. sessions[0]!.append('todo/write', {
  306. todos: [
  307. { content: 'Confirm the panel order', status: 'completed' },
  308. { content: 'Align the panel widths', status: 'in_progress' },
  309. ],
  310. })
  311. await page.locator('[data-testid="todo-panel"]').waitFor({ timeout: 10_000 })
  312. for (const text of ['Layout queue first', 'Layout queue second']) {
  313. // A just-submitted composer is read-only for the prompt round-trip.
  314. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  315. await input.fill(text)
  316. await input.press('Enter')
  317. }
  318. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  319. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  320. .toBe('false')
  321. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  322. const layoutSnapshot = await captureStableAria(
  323. page,
  324. '[class*="centerCol"]',
  325. scaffold.workspaceCwd,
  326. )
  327. await compareOrRefreshGolden(LAYOUT_EXPECTED, layoutSnapshot, MODE)
  328. const expectAlignedContextPanels = async () => {
  329. // Sample one layout: the responsive grid can move between separate browser round trips.
  330. const [queuePanelBox, todoBox, goalBox] = await page.evaluate(() => [
  331. '[data-queue-dock] > div', '[data-testid="todo-panel"]', '[data-goal-bar] > div',
  332. ].map((selector) => {
  333. const box = document.querySelector(selector)?.getBoundingClientRect()
  334. return box === undefined ? null : { x: box.x, y: box.y, width: box.width }
  335. }))
  336. expect(queuePanelBox).not.toBeNull()
  337. expect(todoBox).not.toBeNull()
  338. expect(goalBox).not.toBeNull()
  339. expect(todoBox!.y).toBeLessThan(goalBox!.y)
  340. expect(goalBox!.y).toBeLessThan(queuePanelBox!.y)
  341. expect(todoBox!.x).toBeCloseTo(goalBox!.x, 1)
  342. expect(todoBox!.x).toBeCloseTo(queuePanelBox!.x, 1)
  343. expect(todoBox!.width).toBeCloseTo(goalBox!.width, 1)
  344. expect(todoBox!.width).toBeCloseTo(queuePanelBox!.width, 1)
  345. }
  346. await expectAlignedContextPanels()
  347. await page.setViewportSize({ width: 640, height: 1000 })
  348. await expectAlignedContextPanels()
  349. await page.setViewportSize({ width: 1680, height: 1000 })
  350. await queueHeader.click()
  351. const removeButtons = page.getByRole('button', { name: 'Remove queued message' })
  352. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(2)
  353. await removeButtons.first().click()
  354. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(1)
  355. await removeButtons.first().click()
  356. await expect.poll(() => page.locator('[data-queue-dock]').count(), { timeout: 10_000 }).toBe(0)
  357. await page.getByRole('button', { name: 'Clear goal' }).click()
  358. await expect.poll(() => page.locator('[data-goal-bar]').count(), { timeout: 10_000 }).toBe(0)
  359. await page.getByRole('button', { name: 'Stop generating' }).click()
  360. await settled
  361. expect(turnEndReasons(sessionEvents)).toEqual(['aborted'])
  362. expect(tripwire.pageErrors).toEqual([])
  363. expect(tripwire.warnings).toEqual([])
  364. }, 120_000)
  365. it.skipIf(MODE === 'record')('keeps its snapshot inventory closed', async () => {
  366. await assertFixtureInventory(
  367. SNAPSHOT_DIR,
  368. [
  369. 'collapsed.expected.md', 'editing.expected.md', 'layout.expected.md',
  370. 'preserved.expected.md', 'preserved-expanded.expected.md', 'ui.expected.md',
  371. 'sending.expected.md', 'failed.expected.md', 'writer-held.expected.md',
  372. ],
  373. )
  374. })
  375. })