queue-actions.e2e.ts 19 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380
  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 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. /** Wait for the exact queue mutation response before observing its unlocked actions. */
  63. async function settleQueueAction(action: () => Promise<void>, remainingText: string): Promise<void> {
  64. const response = page.waitForResponse('**/api/session/updateQueue')
  65. await action()
  66. expect((await response).ok()).toBe(true)
  67. const row = page.locator('[data-queue-dock] li', { hasText: remainingText })
  68. await expect.poll(() => row.getByRole('button', { name: 'Edit queued message' }).isEnabled()).toBe(true)
  69. await expect.poll(() => row.getByRole('button', { name: 'Remove queued message' }).isEnabled()).toBe(true)
  70. }
  71. it.skipIf(MODE === 'record')('edits and removes exact occurrences and preserves Queue across stop', async () => {
  72. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-queue-actions-'))
  73. const readyFile = join(overrideDir, '.hang-ready')
  74. const overridePath = join(overrideDir, 'replay.override.json')
  75. const recorded = deriveReplayScript(parseSessionLog(await readFile(FIXTURE, 'utf8')))
  76. expect(recorded).toHaveLength(1)
  77. const replay: ReplayEntry[] = [
  78. { kind: 'hang', readyFile },
  79. recorded[0]!,
  80. recorded[0]!,
  81. recorded[0]!,
  82. ]
  83. await writeFile(overridePath, JSON.stringify(replay))
  84. const sessionEvents: SessionEvent[] = []
  85. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  86. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  87. browser = await chromium.launch()
  88. page = await newEnglishPage(browser)
  89. const tripwire = watchConsole(page)
  90. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  91. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  92. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  93. onTestFailed(() => saveFailureShot(page, 'web-e2e-queue-actions'))
  94. const input = page.locator('[data-composer-input]').first()
  95. const firstSettled = scaffold.whenTurnSettled()
  96. await input.fill(ACTIVE_PROMPT)
  97. await input.press('Enter')
  98. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  99. const admitted = page.waitForResponse('**/api/session/prompt')
  100. const received = Promise.withResolvers<undefined>()
  101. const release = Promise.withResolvers<undefined>()
  102. await page.route('**/api/session/prompt', async (route) => {
  103. received.resolve(undefined)
  104. await release.promise
  105. await route.continue()
  106. }, { times: 1 })
  107. try {
  108. await input.fill(REMOVE)
  109. await input.press('Enter')
  110. await received.promise
  111. const pending = page.locator('[data-queue-dock] [data-submission-echo]')
  112. await pending.getByRole('status').waitFor()
  113. expect(await pending.getByRole('status').textContent()).toBe('Sending…')
  114. expect(await pending.getByRole('button').count()).toBe(3)
  115. expect(await pending.getByRole('button').evaluateAll(buttons =>
  116. buttons.every(button => (button as HTMLButtonElement).disabled))).toBe(true)
  117. expect(await input.textContent()).toBe('')
  118. expect(await input.getAttribute('contenteditable')).toBe('true')
  119. const sending = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  120. await compareOrRefreshGolden(SENDING_EXPECTED, sending, MODE)
  121. await page.setViewportSize({ width: 390, height: 1000 })
  122. await page.locator('[data-sidebar-collapsed="true"]').waitFor()
  123. await expect.poll(() => pending.evaluate(element => element.scrollWidth <= element.clientWidth)).toBe(true)
  124. await page.setViewportSize({ width: 1680, height: 1000 })
  125. await page.locator('[data-sidebar-collapsed="true"]').waitFor({ state: 'detached' })
  126. } finally {
  127. release.resolve(undefined)
  128. }
  129. expect((await admitted).ok()).toBe(true)
  130. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).isEnabled()).toBe(true)
  131. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  132. expect(await page.locator('[data-queue-dock]').getByRole('status').count()).toBe(0)
  133. await input.fill(EDIT)
  134. await input.press('Enter')
  135. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  136. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  137. .toBe('false')
  138. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  139. const collapsedSnapshot = await captureStableAria(
  140. page,
  141. '[class*="centerCol"]',
  142. scaffold.workspaceCwd,
  143. )
  144. await compareOrRefreshGolden(COLLAPSED_EXPECTED, collapsedSnapshot, MODE)
  145. await queueHeader.click()
  146. await expect.poll(
  147. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  148. { timeout: 10_000 },
  149. ).toBe(2)
  150. await page.setViewportSize({ width: 640, height: 1000 })
  151. await page.locator('[data-sidebar-collapsed="true"]').waitFor()
  152. // The responsive sidebar and composer settle independently; sample both
  153. // rectangles in one browser task so the comparison uses one layout.
  154. await vi.waitFor(async () => {
  155. const metrics = await page.evaluate(() => {
  156. const queue = document.querySelector('[data-queue-dock]')
  157. const composer = document.querySelector('[data-composer-card]')
  158. if (queue === null || composer === null) return undefined
  159. const queueBox = queue.getBoundingClientRect()
  160. const composerBox = composer.getBoundingClientRect()
  161. return {
  162. leftInset: queueBox.left - composerBox.left,
  163. rightInset: composerBox.right - queueBox.right,
  164. dockInset: Number.parseFloat(getComputedStyle(composer).getPropertyValue('--dsh-composer-dock-inset')),
  165. }
  166. })
  167. expect(metrics).toBeDefined()
  168. expect(metrics!.leftInset).toBeGreaterThanOrEqual(0)
  169. expect(metrics!.rightInset).toBeGreaterThanOrEqual(0)
  170. expect(metrics!.leftInset).toBeCloseTo(metrics!.dockInset, 1)
  171. expect(metrics!.rightInset).toBeCloseTo(metrics!.dockInset, 1)
  172. }, { timeout: 10_000 })
  173. await page.setViewportSize({ width: 1680, height: 1000 })
  174. const editRow = page.locator('[data-queue-dock] li', { hasText: EDIT })
  175. await editRow.getByRole('button', { name: 'Edit queued message' }).click()
  176. const editor = page.getByRole('textbox', { name: 'Edit queued message' })
  177. await editor.fill(EDITED)
  178. await page.getByRole('button', { name: 'Save queued message' }).hover()
  179. await page.getByRole('tooltip', { name: 'Save queued message', exact: true }).waitFor()
  180. const editingSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  181. await compareOrRefreshGolden(EDITING_EXPECTED, editingSnapshot, MODE)
  182. await settleQueueAction(() => page.getByRole('button', { name: 'Save queued message' }).click(), EDITED)
  183. await page.getByText(EDITED, { exact: true }).waitFor()
  184. const removeRow = page.locator('[data-queue-dock] li', { hasText: REMOVE })
  185. await settleQueueAction(() => removeRow.getByRole('button', { name: 'Remove queued message' }).click(), EDITED)
  186. await expect.poll(() => page.getByText(REMOVE, { exact: true }).count()).toBe(0)
  187. // The queue stream can remove the row before the mutation reply clears busy.
  188. const remainingEdit = page.getByRole('button', { name: 'Edit queued message', exact: true })
  189. await expect.poll(() => remainingEdit.isEnabled(), { timeout: 10_000 }).toBe(true)
  190. await remainingEdit.hover()
  191. await page.getByRole('tooltip', { name: 'Edit queued message', exact: true }).waitFor()
  192. const snapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  193. await compareOrRefreshGolden(UI_EXPECTED, snapshot, MODE)
  194. expect(sessionEvents.filter(event => event.type === 'user/message' && event.data.source.kind === 'user')).toHaveLength(1)
  195. await page.route('**/api/session/prompt', async (route) => {
  196. const envelope = route.request().postDataJSON() as { rpcId: string }
  197. await route.fulfill({
  198. status: 200,
  199. contentType: 'application/json',
  200. json: {
  201. type: 'server-response', rpcId: envelope.rpcId,
  202. result: {
  203. ok: false,
  204. error: { code: 'session/agent-busy', message: 'Queue submission failed', details: { reason: 'admission refused' } },
  205. },
  206. },
  207. })
  208. }, { times: 1 })
  209. await input.fill(FAILED)
  210. await input.press('Enter')
  211. const failure = page.getByRole('alert').filter({ hasText: 'Queue submission failed' })
  212. await failure.waitFor()
  213. await expect.poll(() => input.textContent()).toBe(FAILED)
  214. expect(await page.locator('[data-queue-dock] [data-submission-echo]').count()).toBe(0)
  215. const failed = [
  216. await failure.ariaSnapshot(),
  217. await captureStableAria(page, '[data-composer-card]', scaffold.workspaceCwd),
  218. ].join('\n')
  219. await compareOrRefreshGolden(FAILED_EXPECTED, failed, MODE)
  220. await input.fill(TAIL)
  221. await input.press('Enter')
  222. await expect.poll(
  223. () => page.getByRole('button', { name: 'Remove queued message', disabled: false }).count(),
  224. { timeout: 10_000 },
  225. ).toBe(2)
  226. const stopButton = page.getByRole('button', { name: 'Stop generating' })
  227. await stopButton.hover()
  228. await page.getByRole('tooltip', { name: 'Stop generating', exact: true }).waitFor()
  229. await stopButton.click()
  230. await firstSettled
  231. await expect.poll(() => page.getByRole('button', { name: 'Stop generating' }).count())
  232. .toBe(0)
  233. await expect.poll(() => page.getByRole('button', { name: 'Remove queued message' }).count())
  234. .toBe(2)
  235. // The disabled Send button must dismiss the active Stop tooltip without mouseleave.
  236. await expect.poll(() => page.getByRole('tooltip').count()).toBe(0)
  237. const preservedSnapshot = await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd)
  238. await compareOrRefreshGolden(PRESERVED_EXPECTED, preservedSnapshot, MODE)
  239. const expanded = await captureExpandedTurnProcessAria(
  240. page,
  241. '[class*="centerCol"]',
  242. scaffold.workspaceCwd,
  243. )
  244. await compareOrRefreshGolden(PRESERVED_EXPANDED_EXPECTED, expanded, MODE)
  245. const settled = scaffold.whenTurnSettled()
  246. await input.fill(WAKE)
  247. await input.press('Enter')
  248. await settled
  249. await expect.poll(() => turnEndReasons(sessionEvents), { timeout: 15_000 })
  250. .toEqual(['aborted', 'completed', 'completed', 'completed'])
  251. expect(sessionEvents.flatMap(event => event.type === 'user/message' && event.data.source.kind === 'user'
  252. ? event.data.content.flatMap(block => block.type === 'text' ? [block.text] : [])
  253. : [])).toEqual([ACTIVE_PROMPT, EDITED, TAIL, WAKE])
  254. await expect.poll(() => page.locator('[data-queue-dock]').count()).toBe(0)
  255. expect(tripwire.pageErrors).toEqual([])
  256. expect(tripwire.warnings).toEqual([])
  257. }, 120_000)
  258. it.skipIf(MODE === 'record')('orders Todo before Goal and Queue on one responsive card column', async () => {
  259. overrideDir = await mkdtemp(join(tmpdir(), 'dsh-web-context-layout-'))
  260. const readyFile = join(overrideDir, '.hang-ready')
  261. const overridePath = join(overrideDir, 'replay.override.json')
  262. await writeFile(overridePath, JSON.stringify([{ kind: 'hang', readyFile } satisfies ReplayEntry]))
  263. const sessionEvents: SessionEvent[] = []
  264. scaffold = await launchWebScaffold({ replayFixture: FIXTURE, replayOverride: overridePath, compareReplaySession: false })
  265. scaffold.ctx.on('session/event', (_session, event: SessionEvent) => { sessionEvents.push(event) })
  266. browser = await chromium.launch()
  267. page = await newEnglishPage(browser)
  268. const tripwire = watchConsole(page)
  269. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  270. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  271. await connectFreshWorkspace(page, scaffold.workspaceCwd)
  272. onTestFailed(() => saveFailureShot(page, 'web-e2e-context-layout'))
  273. const input = page.locator('[data-composer-input]').first()
  274. const settled = scaffold.whenTurnSettled()
  275. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  276. await input.fill('/goal Keep the composer context panels aligned')
  277. await input.press('Enter')
  278. await expect.poll(() => existsSync(readyFile), { timeout: 15_000 }).toBe(true)
  279. await page.locator('[data-goal-bar]').waitFor({ timeout: 10_000 })
  280. const sessions = scaffold.ctx.sessions.list()
  281. expect(sessions).toHaveLength(1)
  282. sessions[0]!.append('todo/write', {
  283. todos: [
  284. { content: 'Confirm the panel order', status: 'completed' },
  285. { content: 'Align the panel widths', status: 'in_progress' },
  286. ],
  287. })
  288. await page.locator('[data-testid="todo-panel"]').waitFor({ timeout: 10_000 })
  289. for (const text of ['Layout queue first', 'Layout queue second']) {
  290. // A just-submitted composer is read-only for the prompt round-trip.
  291. await page.locator('[data-composer-input][contenteditable="true"]').first().waitFor({ timeout: 10_000 })
  292. await input.fill(text)
  293. await input.press('Enter')
  294. }
  295. const queueHeader = page.getByRole('button', { name: '2 queued messages' })
  296. await expect.poll(() => queueHeader.getAttribute('aria-expanded'), { timeout: 10_000 })
  297. .toBe('false')
  298. await expect.poll(() => queueHeader.getByRole('status').count(), { timeout: 10_000 }).toBe(0)
  299. const layoutSnapshot = await captureStableAria(
  300. page,
  301. '[class*="centerCol"]',
  302. scaffold.workspaceCwd,
  303. )
  304. await compareOrRefreshGolden(LAYOUT_EXPECTED, layoutSnapshot, MODE)
  305. const expectAlignedContextPanels = async () => {
  306. // Sample one layout: the responsive grid can move between separate browser round trips.
  307. const [queuePanelBox, todoBox, goalBox] = await page.evaluate(() => [
  308. '[data-queue-dock] > div', '[data-testid="todo-panel"]', '[data-goal-bar] > div',
  309. ].map((selector) => {
  310. const box = document.querySelector(selector)?.getBoundingClientRect()
  311. return box === undefined ? null : { x: box.x, y: box.y, width: box.width }
  312. }))
  313. expect(queuePanelBox).not.toBeNull()
  314. expect(todoBox).not.toBeNull()
  315. expect(goalBox).not.toBeNull()
  316. expect(todoBox!.y).toBeLessThan(goalBox!.y)
  317. expect(goalBox!.y).toBeLessThan(queuePanelBox!.y)
  318. expect(todoBox!.x).toBeCloseTo(goalBox!.x, 1)
  319. expect(todoBox!.x).toBeCloseTo(queuePanelBox!.x, 1)
  320. expect(todoBox!.width).toBeCloseTo(goalBox!.width, 1)
  321. expect(todoBox!.width).toBeCloseTo(queuePanelBox!.width, 1)
  322. }
  323. await expectAlignedContextPanels()
  324. await page.setViewportSize({ width: 640, height: 1000 })
  325. await expectAlignedContextPanels()
  326. await page.setViewportSize({ width: 1680, height: 1000 })
  327. await queueHeader.click()
  328. const removeButtons = page.getByRole('button', { name: 'Remove queued message' })
  329. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(2)
  330. await removeButtons.first().click()
  331. await expect.poll(() => removeButtons.count(), { timeout: 10_000 }).toBe(1)
  332. await removeButtons.first().click()
  333. await expect.poll(() => page.locator('[data-queue-dock]').count(), { timeout: 10_000 }).toBe(0)
  334. await page.getByRole('button', { name: 'Clear goal' }).click()
  335. await expect.poll(() => page.locator('[data-goal-bar]').count(), { timeout: 10_000 }).toBe(0)
  336. await page.getByRole('button', { name: 'Stop generating' }).click()
  337. await settled
  338. expect(turnEndReasons(sessionEvents)).toEqual(['aborted'])
  339. expect(tripwire.pageErrors).toEqual([])
  340. expect(tripwire.warnings).toEqual([])
  341. }, 120_000)
  342. it.skipIf(MODE === 'record')('keeps its snapshot inventory closed', async () => {
  343. await assertFixtureInventory(
  344. SNAPSHOT_DIR,
  345. [
  346. 'collapsed.expected.md', 'editing.expected.md', 'layout.expected.md',
  347. 'preserved.expected.md', 'preserved-expanded.expected.md', 'ui.expected.md',
  348. 'sending.expected.md', 'failed.expected.md',
  349. ],
  350. )
  351. })
  352. })