trajectory-virtualization.e2e.ts 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349
  1. // Browser contract for the tail-paged, virtualized Trajectory ledger. The
  2. // scenario proves that semantic row identity survives an older-page prepend,
  3. // DOM mounting stays bounded, and every scroll range remains reachable.
  4. import { mkdtemp, rm, writeFile } from 'node:fs/promises'
  5. import { tmpdir } from 'node:os'
  6. import { fileURLToPath } from 'node:url'
  7. import { join } from 'node:path'
  8. import type { Browser, Page } from 'playwright'
  9. import { chromium } from 'playwright'
  10. import { afterAll, beforeAll, describe, expect, it, onTestFailed } from 'vitest'
  11. import type { StreamChunk } from '@deepseek-ai/dsh-llm'
  12. import type { ReplayEntry } from '@deepseek-ai/dsh-llm-replay'
  13. import { sessionFixtureName } from '@deepseek-ai/dsh-session-snapshot'
  14. import { SESSION_FORMAT_VERSION } from '@deepseek-ai/dsh-session'
  15. import { createChatScrollFixture } from './chat-scroll-fixture.ts'
  16. import {
  17. captureStableAria,
  18. compareOrRefreshGolden,
  19. launchWebScaffold,
  20. seedSession,
  21. watchConsole,
  22. webSnapshotMode,
  23. type WebScaffold,
  24. } from './scaffold.ts'
  25. import { newEnglishPage, saveFailureShot } from './support.ts'
  26. const MODE = webSnapshotMode()
  27. const LOAD_MORE_EXPECTED = fileURLToPath(new URL(
  28. './expected/trajectory-virtualization/load-more.expected.md',
  29. import.meta.url,
  30. ))
  31. const SESSION_ID = 'trajectory-virtualization-e2e'
  32. const FIXTURE = createChatScrollFixture({
  33. markerPrefix: 'TRAJECTORY_VIRTUAL',
  34. title: 'TRAJECTORY_VIRTUAL long ledger',
  35. turns: 88,
  36. })
  37. const MAX_MOUNTED_ROWS = 160
  38. const GEOMETRY_TOLERANCE = 2
  39. const STREAM_MARKER = 'TRAJECTORY_VIRTUAL_STREAM_FINISHED'
  40. const STREAM_TEXT = Array.from(
  41. { length: 80 },
  42. (_, index) => `stream fragment ${String(index + 1).padStart(2, '0')} `,
  43. ).join('') + STREAM_MARKER
  44. const STREAM_CHUNKS: StreamChunk[] = [
  45. { type: 'block-start', index: 0, blockType: 'text' },
  46. ...Array.from({ length: 80 }, (_, index): StreamChunk => ({
  47. type: 'text-delta',
  48. index: 0,
  49. text: `stream fragment ${String(index + 1).padStart(2, '0')} `,
  50. })),
  51. { type: 'text-delta', index: 0, text: STREAM_MARKER },
  52. { type: 'block-end', index: 0, block: { type: 'text', text: STREAM_TEXT } },
  53. { type: 'usage', usage: { inputTokens: 2_700, outputTokens: 240 } },
  54. { type: 'finish', reason: { kind: 'stop' } },
  55. ]
  56. interface ScrollGeometry {
  57. readonly clientHeight: number
  58. readonly scrollHeight: number
  59. readonly scrollTop: number
  60. }
  61. interface RowAnchor {
  62. readonly key: string
  63. readonly top: number
  64. }
  65. async function openSeed(page: Page): Promise<void> {
  66. // Search collapsed into a header action; expand it before filling.
  67. const searchButton = page.getByRole('button', { name: 'Search sessions' })
  68. if (await searchButton.getAttribute('aria-expanded') !== 'true') await searchButton.click()
  69. const search = page.getByRole('textbox', { name: 'Search sessions...', exact: true })
  70. await search.fill(FIXTURE.markers.user(1))
  71. const result = page.getByRole('tree', { name: 'Search results' }).getByRole('treeitem')
  72. await expect.poll(() => result.count(), { timeout: 60_000 }).toBe(1)
  73. await result.click()
  74. await page.getByRole('tab', { name: 'Trajectory', exact: true }).waitFor({ timeout: 30_000 })
  75. await page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false })
  76. .last()
  77. .waitFor({ timeout: 30_000 })
  78. }
  79. async function openTrajectory(page: Page): Promise<void> {
  80. await page.getByRole('tab', { name: 'Trajectory', exact: true }).click()
  81. const pane = page.locator('[data-trajectory-scroll]')
  82. await pane.waitFor({ timeout: 30_000 })
  83. await page.locator('[data-trajectory-scroll] table[data-scroll-ready="true"]')
  84. .waitFor({ timeout: 30_000 })
  85. }
  86. async function logicalRows(page: Page): Promise<number> {
  87. const raw = await page.locator('[data-trajectory-scroll] table').getAttribute('aria-rowcount')
  88. if (raw === null || !/^\d+$/.test(raw)) {
  89. throw new Error(`trajectory table has invalid aria-rowcount ${JSON.stringify(raw)}`)
  90. }
  91. return Number(raw)
  92. }
  93. async function mountedRows(page: Page): Promise<number> {
  94. return page.locator('[data-trajectory-scroll] tr[data-trajectory-row-key]').count()
  95. }
  96. async function geometry(page: Page): Promise<ScrollGeometry> {
  97. return page.locator('[data-trajectory-scroll]').evaluate(host => ({
  98. clientHeight: host.clientHeight,
  99. scrollHeight: host.scrollHeight,
  100. scrollTop: host.scrollTop,
  101. }))
  102. }
  103. async function nextPaint(page: Page): Promise<void> {
  104. await page.evaluate(() => new Promise<void>((resolve) => {
  105. requestAnimationFrame(() => requestAnimationFrame(() => { resolve() }))
  106. }))
  107. }
  108. async function scrollToRatio(page: Page, ratio: number): Promise<void> {
  109. await page.locator('[data-trajectory-scroll]').evaluate((host, value) => {
  110. const maximum = Math.max(0, host.scrollHeight - host.clientHeight)
  111. host.scrollTop = Math.round(maximum * value)
  112. host.dispatchEvent(new Event('scroll'))
  113. }, ratio)
  114. await nextPaint(page)
  115. }
  116. async function firstVisibleRow(page: Page): Promise<RowAnchor> {
  117. return page.locator('[data-trajectory-scroll]').evaluate((host) => {
  118. const hostBox = host.getBoundingClientRect()
  119. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  120. const row = rows.find((candidate) => {
  121. const box = candidate.getBoundingClientRect()
  122. return candidate.dataset.requestOnly !== 'true'
  123. && box.bottom > hostBox.top
  124. && box.top < hostBox.bottom
  125. })
  126. const key = row?.dataset.trajectoryRowKey
  127. if (row === undefined || key === undefined) {
  128. throw new Error('trajectory scrollport has no visible semantic row')
  129. }
  130. return { key, top: row.getBoundingClientRect().top - hostBox.top }
  131. })
  132. }
  133. async function rowTop(page: Page, key: string): Promise<number | null> {
  134. return page.locator('[data-trajectory-scroll]').evaluate((host, targetKey) => {
  135. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  136. const row = rows.find(candidate => candidate.dataset.trajectoryRowKey === targetKey)
  137. return row === undefined
  138. ? null
  139. : row.getBoundingClientRect().top - host.getBoundingClientRect().top
  140. }, key)
  141. }
  142. async function loadToFirstTurn(page: Page): Promise<void> {
  143. const marker = FIXTURE.markers.user(1)
  144. for (let attempt = 0; attempt < 12; attempt += 1) {
  145. await scrollToRatio(page, 0)
  146. if (await page.getByText(marker, { exact: false }).count() > 0) return
  147. const before = await logicalRows(page)
  148. const anchor = await firstVisibleRow(page)
  149. await expect.poll(async () => ({
  150. marker: await page.getByText(marker, { exact: false }).count() > 0,
  151. rows: await logicalRows(page),
  152. }), { timeout: 30_000 }).not.toEqual({ marker: false, rows: before })
  153. await nextPaint(page)
  154. await expect.poll(async () => {
  155. const top = await rowTop(page, anchor.key)
  156. return top === null ? Number.POSITIVE_INFINITY : Math.abs(top - anchor.top)
  157. }, { timeout: 15_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  158. }
  159. throw new Error('trajectory did not reach the first turn after twelve older-page requests')
  160. }
  161. describe('web e2e: Trajectory virtualization over tail-paged history', () => {
  162. let scaffold: WebScaffold
  163. let browser: Browser
  164. let page: Page
  165. let tripwire: ReturnType<typeof watchConsole>
  166. let replayDir: string
  167. beforeAll(async () => {
  168. replayDir = await mkdtemp(join(tmpdir(), 'dsh-trajectory-virtualization-'))
  169. const replayFixture = join(replayDir, sessionFixtureName(0, SESSION_FORMAT_VERSION))
  170. const replayOverride = join(replayDir, 'replay.override.json')
  171. await writeFile(replayFixture, FIXTURE.log)
  172. await writeFile(replayOverride, JSON.stringify([{
  173. kind: 'chunks',
  174. chunks: STREAM_CHUNKS,
  175. } satisfies ReplayEntry]))
  176. scaffold = await launchWebScaffold({
  177. paceMs: 10,
  178. replayFixture,
  179. replayOverride,
  180. })
  181. await seedSession(scaffold, FIXTURE.log, SESSION_ID)
  182. browser = await chromium.launch()
  183. page = await newEnglishPage(browser, 900)
  184. tripwire = watchConsole(page)
  185. await page.goto(scaffold.authenticatedUrl, { waitUntil: 'load' })
  186. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  187. // The compact layout dropped group session counts; the seeded baseline is
  188. // the Ungrouped bucket once cold summaries load.
  189. await page.getByText('Ungrouped', { exact: true }).waitFor({ timeout: 30_000 })
  190. }, 120_000)
  191. afterAll(async () => {
  192. await browser?.close()
  193. await scaffold?.close()
  194. await rm(replayDir, { recursive: true, force: true })
  195. })
  196. it.skipIf(MODE === 'record')('retains identity on prepend and reaches the bounded virtual range', async () => {
  197. onTestFailed(() => saveFailureShot(page, 'web-e2e-trajectory-virtualization'))
  198. await openSeed(page)
  199. let held = false
  200. let releaseHistory: () => void = () => {}
  201. let finishHeldRequest: () => void = () => {}
  202. const gate = new Promise<void>((resolve) => { releaseHistory = resolve })
  203. const heldRequestFinished = new Promise<void>((resolve) => { finishHeldRequest = resolve })
  204. await page.route('**/api/session/page', async (route) => {
  205. const request = route.request().postDataJSON() as {
  206. method?: string
  207. payload?: { args?: { request?: { beforeSeq?: number } } }
  208. }
  209. if (!held && request.method === 'session/page'
  210. && request.payload?.args?.request?.beforeSeq !== undefined) {
  211. held = true
  212. await gate
  213. try {
  214. await route.continue()
  215. } finally {
  216. finishHeldRequest()
  217. }
  218. return
  219. }
  220. await route.continue()
  221. })
  222. try {
  223. await openTrajectory(page)
  224. const initialRows = await logicalRows(page)
  225. expect(initialRows).toBeGreaterThan(0)
  226. expect(await page.getByText('Initial System Prompt', { exact: true }).count()).toBe(0)
  227. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  228. const loadMore = page.locator('[data-history-load] button')
  229. await loadMore.waitFor({ timeout: 15_000 })
  230. expect(await loadMore.textContent()).toBe('Load earlier history')
  231. const loadMoreSnapshot = await captureStableAria(
  232. page,
  233. '[data-history-load]',
  234. scaffold.workspaceCwd,
  235. )
  236. await compareOrRefreshGolden(LOAD_MORE_EXPECTED, loadMoreSnapshot, MODE)
  237. // Avoid Playwright scrolling the offscreen first row into the automatic-load threshold.
  238. await loadMore.evaluate((button: HTMLButtonElement) => { button.click() })
  239. await expect.poll(() => held, { timeout: 15_000 }).toBe(true)
  240. await expect.poll(async () => ({
  241. disabled: await loadMore.isDisabled(),
  242. label: await loadMore.getAttribute('aria-label'),
  243. }), { timeout: 15_000 }).toEqual({
  244. disabled: true,
  245. label: 'Loading earlier history…',
  246. })
  247. await scrollToRatio(page, 0)
  248. const anchor = await firstVisibleRow(page)
  249. const selectedRow = page.locator(
  250. `[data-trajectory-scroll] tr[data-trajectory-row-key=${JSON.stringify(anchor.key)}]`,
  251. )
  252. await selectedRow.click()
  253. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  254. .toBe('true')
  255. releaseHistory()
  256. await expect.poll(() => logicalRows(page), { timeout: 60_000 }).toBeGreaterThan(initialRows)
  257. await nextPaint(page)
  258. await expect.poll(async () => {
  259. const top = await rowTop(page, anchor.key)
  260. return top === null ? Number.POSITIVE_INFINITY : Math.abs(top - anchor.top)
  261. }, { timeout: 15_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  262. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  263. .toBe('true')
  264. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  265. await loadToFirstTurn(page)
  266. await expect.poll(
  267. () => page.getByText(FIXTURE.markers.user(1), { exact: false }).count(),
  268. { timeout: 10_000 },
  269. ).toBeGreaterThan(0)
  270. const fullRows = await logicalRows(page)
  271. await scrollToRatio(page, 0.5)
  272. const middle = await geometry(page)
  273. const maximum = middle.scrollHeight - middle.clientHeight
  274. expect(middle.scrollTop).toBeGreaterThan(maximum * 0.25)
  275. expect(middle.scrollTop).toBeLessThan(maximum * 0.75)
  276. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  277. expect(await mountedRows(page)).toBeLessThan(fullRows)
  278. await scrollToRatio(page, 1)
  279. await expect.poll(async () => {
  280. const value = await geometry(page)
  281. return value.scrollHeight - value.clientHeight - value.scrollTop
  282. }, { timeout: 10_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  283. await expect.poll(
  284. () => page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false }).count(),
  285. { timeout: 10_000 },
  286. ).toBeGreaterThan(0)
  287. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  288. const trajectoryScroll = page.locator('[data-trajectory-scroll]')
  289. await trajectoryScroll.evaluate((host) => {
  290. const measuredWindow = window as Window & { __trajectoryScrollCalls?: number }
  291. measuredWindow.__trajectoryScrollCalls = 0
  292. const original = host.scrollTo.bind(host)
  293. const trackedScrollTo = (...args: [ScrollToOptions?] | [number, number]) => {
  294. measuredWindow.__trajectoryScrollCalls = (measuredWindow.__trajectoryScrollCalls ?? 0) + 1
  295. Reflect.apply(original, host, args)
  296. }
  297. host.scrollTo = trackedScrollTo as typeof host.scrollTo
  298. })
  299. const settled = scaffold.whenTurnSettled()
  300. const input = page.locator('[data-composer-input]').first()
  301. await input.fill('Stream one deterministic response while Trajectory remains visible.')
  302. await input.press('Enter')
  303. await settled
  304. await page.getByText('stream fragment 01', { exact: false }).waitFor({ timeout: 30_000 })
  305. await nextPaint(page)
  306. const streamingScrollCalls = await trajectoryScroll.evaluate(() => {
  307. return (window as Window & { __trajectoryScrollCalls?: number })
  308. .__trajectoryScrollCalls ?? 0
  309. })
  310. expect(streamingScrollCalls).toBeLessThanOrEqual(5)
  311. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  312. expect({
  313. pageErrors: tripwire.pageErrors,
  314. warnings: tripwire.warnings,
  315. }).toEqual({ pageErrors: [], warnings: [] })
  316. } finally {
  317. releaseHistory()
  318. if (held) await heldRequestFinished
  319. await page.unroute('**/api/session/page')
  320. }
  321. }, 180_000)
  322. })