trajectory-virtualization.e2e.ts 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314
  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 { join } from 'node:path'
  7. import type { Browser, Page } from 'playwright'
  8. import { chromium } from 'playwright'
  9. import { afterAll, beforeAll, describe, expect, it, onTestFailed } from 'vitest'
  10. import type { StreamChunk } from '@deepseek-ai/dsh-llm'
  11. import type { ReplayEntry } from '@deepseek-ai/dsh-llm-replay'
  12. import { createChatScrollFixture } from './chat-scroll-fixture.ts'
  13. import {
  14. launchWebScaffold,
  15. seedSession,
  16. watchConsole,
  17. webSnapshotMode,
  18. type WebScaffold,
  19. } from './scaffold.ts'
  20. import { newEnglishPage, saveFailureShot } from './support.ts'
  21. const MODE = webSnapshotMode()
  22. const SESSION_ID = 'trajectory-virtualization-e2e'
  23. const FIXTURE = createChatScrollFixture({
  24. markerPrefix: 'TRAJECTORY_VIRTUAL',
  25. title: 'TRAJECTORY_VIRTUAL long ledger',
  26. turns: 88,
  27. })
  28. const MAX_MOUNTED_ROWS = 160
  29. const GEOMETRY_TOLERANCE = 2
  30. const STREAM_MARKER = 'TRAJECTORY_VIRTUAL_STREAM_FINISHED'
  31. const STREAM_TEXT = Array.from(
  32. { length: 80 },
  33. (_, index) => `stream fragment ${String(index + 1).padStart(2, '0')} `,
  34. ).join('') + STREAM_MARKER
  35. const STREAM_CHUNKS: StreamChunk[] = [
  36. { type: 'block-start', index: 0, blockType: 'text' },
  37. ...Array.from({ length: 80 }, (_, index): StreamChunk => ({
  38. type: 'text-delta',
  39. index: 0,
  40. text: `stream fragment ${String(index + 1).padStart(2, '0')} `,
  41. })),
  42. { type: 'text-delta', index: 0, text: STREAM_MARKER },
  43. { type: 'block-end', index: 0, block: { type: 'text', text: STREAM_TEXT } },
  44. { type: 'usage', usage: { inputTokens: 2_700, outputTokens: 240 } },
  45. { type: 'finish', reason: { kind: 'stop' } },
  46. ]
  47. interface ScrollGeometry {
  48. readonly clientHeight: number
  49. readonly scrollHeight: number
  50. readonly scrollTop: number
  51. }
  52. interface RowAnchor {
  53. readonly key: string
  54. readonly top: number
  55. }
  56. async function openSeed(page: Page): Promise<void> {
  57. // Search collapsed into a header action; expand it before filling.
  58. const searchButton = page.getByRole('button', { name: 'Search sessions' })
  59. if (await searchButton.getAttribute('aria-expanded') !== 'true') await searchButton.click()
  60. const search = page.getByRole('textbox', { name: 'Search sessions...', exact: true })
  61. await search.fill(FIXTURE.markers.user(1))
  62. const result = page.getByRole('tree', { name: 'Search results' }).getByRole('treeitem')
  63. await expect.poll(() => result.count(), { timeout: 60_000 }).toBe(1)
  64. await result.click()
  65. await page.getByRole('tab', { name: 'Trajectory', exact: true }).waitFor({ timeout: 30_000 })
  66. await page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false })
  67. .last()
  68. .waitFor({ timeout: 30_000 })
  69. }
  70. async function openTrajectory(page: Page): Promise<void> {
  71. await page.getByRole('tab', { name: 'Trajectory', exact: true }).click()
  72. const pane = page.locator('[data-trajectory-scroll]')
  73. await pane.waitFor({ timeout: 30_000 })
  74. await page.locator('[data-trajectory-scroll] table[data-scroll-ready="true"]')
  75. .waitFor({ timeout: 30_000 })
  76. }
  77. async function logicalRows(page: Page): Promise<number> {
  78. const raw = await page.locator('[data-trajectory-scroll] table').getAttribute('aria-rowcount')
  79. if (raw === null || !/^\d+$/.test(raw)) {
  80. throw new Error(`trajectory table has invalid aria-rowcount ${JSON.stringify(raw)}`)
  81. }
  82. return Number(raw)
  83. }
  84. async function mountedRows(page: Page): Promise<number> {
  85. return page.locator('[data-trajectory-scroll] tr[data-trajectory-row-key]').count()
  86. }
  87. async function geometry(page: Page): Promise<ScrollGeometry> {
  88. return page.locator('[data-trajectory-scroll]').evaluate(host => ({
  89. clientHeight: host.clientHeight,
  90. scrollHeight: host.scrollHeight,
  91. scrollTop: host.scrollTop,
  92. }))
  93. }
  94. async function nextPaint(page: Page): Promise<void> {
  95. await page.evaluate(() => new Promise<void>((resolve) => {
  96. requestAnimationFrame(() => requestAnimationFrame(() => { resolve() }))
  97. }))
  98. }
  99. async function scrollToRatio(page: Page, ratio: number): Promise<void> {
  100. await page.locator('[data-trajectory-scroll]').evaluate((host, value) => {
  101. const maximum = Math.max(0, host.scrollHeight - host.clientHeight)
  102. host.scrollTop = Math.round(maximum * value)
  103. host.dispatchEvent(new Event('scroll'))
  104. }, ratio)
  105. await nextPaint(page)
  106. }
  107. async function firstVisibleRow(page: Page): Promise<RowAnchor> {
  108. return page.locator('[data-trajectory-scroll]').evaluate((host) => {
  109. const hostBox = host.getBoundingClientRect()
  110. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  111. const row = rows.find((candidate) => {
  112. const box = candidate.getBoundingClientRect()
  113. return candidate.dataset.requestOnly !== 'true'
  114. && box.bottom > hostBox.top
  115. && box.top < hostBox.bottom
  116. })
  117. const key = row?.dataset.trajectoryRowKey
  118. if (row === undefined || key === undefined) {
  119. throw new Error('trajectory scrollport has no visible semantic row')
  120. }
  121. return { key, top: row.getBoundingClientRect().top - hostBox.top }
  122. })
  123. }
  124. async function rowTop(page: Page, key: string): Promise<number | null> {
  125. return page.locator('[data-trajectory-scroll]').evaluate((host, targetKey) => {
  126. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  127. const row = rows.find(candidate => candidate.dataset.trajectoryRowKey === targetKey)
  128. return row === undefined
  129. ? null
  130. : row.getBoundingClientRect().top - host.getBoundingClientRect().top
  131. }, key)
  132. }
  133. async function loadToFirstTurn(page: Page): Promise<void> {
  134. const marker = FIXTURE.markers.user(1)
  135. for (let attempt = 0; attempt < 12; attempt += 1) {
  136. await scrollToRatio(page, 0)
  137. if (await page.getByText(marker, { exact: false }).count() > 0) return
  138. const before = await logicalRows(page)
  139. await expect.poll(async () => ({
  140. marker: await page.getByText(marker, { exact: false }).count() > 0,
  141. rows: await logicalRows(page),
  142. }), { timeout: 30_000 }).not.toEqual({ marker: false, rows: before })
  143. }
  144. throw new Error('trajectory did not reach the first turn after twelve older-page requests')
  145. }
  146. describe('web e2e: Trajectory virtualization over tail-paged history', () => {
  147. let scaffold: WebScaffold
  148. let browser: Browser
  149. let page: Page
  150. let tripwire: ReturnType<typeof watchConsole>
  151. let replayDir: string
  152. beforeAll(async () => {
  153. replayDir = await mkdtemp(join(tmpdir(), 'dsh-trajectory-virtualization-'))
  154. const replayFixture = join(replayDir, 'session.jsonl')
  155. const replayOverride = join(replayDir, 'replay.override.json')
  156. await writeFile(replayFixture, FIXTURE.log)
  157. await writeFile(replayOverride, JSON.stringify([{
  158. kind: 'chunks',
  159. chunks: STREAM_CHUNKS,
  160. } satisfies ReplayEntry]))
  161. scaffold = await launchWebScaffold({
  162. paceMs: 10,
  163. replayFixture,
  164. replayOverride,
  165. })
  166. await seedSession(scaffold, FIXTURE.log, SESSION_ID)
  167. browser = await chromium.launch()
  168. page = await newEnglishPage(browser, 900)
  169. tripwire = watchConsole(page)
  170. await page.goto(scaffold.baseUrl, { waitUntil: 'load' })
  171. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  172. // The compact layout dropped group session counts; the seeded baseline is
  173. // the Ungrouped bucket once cold summaries load.
  174. await page.getByText('Ungrouped', { exact: true }).waitFor({ timeout: 30_000 })
  175. }, 120_000)
  176. afterAll(async () => {
  177. await browser?.close()
  178. await scaffold?.close()
  179. await rm(replayDir, { recursive: true, force: true })
  180. })
  181. it.skipIf(MODE === 'record')('retains identity on prepend and reaches the bounded virtual range', async () => {
  182. onTestFailed(() => saveFailureShot(page, 'web-e2e-trajectory-virtualization'))
  183. await openSeed(page)
  184. let held = false
  185. let releaseHistory: () => void = () => {}
  186. let finishHeldRequest: () => void = () => {}
  187. const gate = new Promise<void>((resolve) => { releaseHistory = resolve })
  188. const heldRequestFinished = new Promise<void>((resolve) => { finishHeldRequest = resolve })
  189. await page.route('**/api/session.history', async (route) => {
  190. const request = route.request().postDataJSON() as {
  191. method?: string
  192. payload?: { beforeSeq?: number }
  193. }
  194. if (!held && request.method === 'session.history' && request.payload?.beforeSeq !== undefined) {
  195. held = true
  196. await gate
  197. try {
  198. await route.continue()
  199. } finally {
  200. finishHeldRequest()
  201. }
  202. return
  203. }
  204. await route.continue()
  205. })
  206. try {
  207. await openTrajectory(page)
  208. const initialRows = await logicalRows(page)
  209. expect(initialRows).toBeGreaterThan(0)
  210. expect(await page.getByText('Initial System Prompt', { exact: true }).count()).toBe(0)
  211. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  212. await scrollToRatio(page, 0)
  213. await expect.poll(() => held, { timeout: 15_000 }).toBe(true)
  214. const anchor = await firstVisibleRow(page)
  215. const selectedRow = page.locator(
  216. `[data-trajectory-scroll] tr[data-trajectory-row-key=${JSON.stringify(anchor.key)}]`,
  217. )
  218. await selectedRow.click()
  219. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  220. .toBe('true')
  221. releaseHistory()
  222. await expect.poll(() => logicalRows(page), { timeout: 60_000 }).toBeGreaterThan(initialRows)
  223. await nextPaint(page)
  224. await expect.poll(async () => {
  225. const top = await rowTop(page, anchor.key)
  226. return top === null ? Number.POSITIVE_INFINITY : Math.abs(top - anchor.top)
  227. }, { timeout: 15_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  228. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  229. .toBe('true')
  230. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  231. await loadToFirstTurn(page)
  232. await expect.poll(
  233. () => page.getByText(FIXTURE.markers.user(1), { exact: false }).count(),
  234. { timeout: 10_000 },
  235. ).toBeGreaterThan(0)
  236. const fullRows = await logicalRows(page)
  237. await scrollToRatio(page, 0.5)
  238. const middle = await geometry(page)
  239. const maximum = middle.scrollHeight - middle.clientHeight
  240. expect(middle.scrollTop).toBeGreaterThan(maximum * 0.25)
  241. expect(middle.scrollTop).toBeLessThan(maximum * 0.75)
  242. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  243. expect(await mountedRows(page)).toBeLessThan(fullRows)
  244. await scrollToRatio(page, 1)
  245. await expect.poll(async () => {
  246. const value = await geometry(page)
  247. return value.scrollHeight - value.clientHeight - value.scrollTop
  248. }, { timeout: 10_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  249. await expect.poll(
  250. () => page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false }).count(),
  251. { timeout: 10_000 },
  252. ).toBeGreaterThan(0)
  253. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  254. const trajectoryScroll = page.locator('[data-trajectory-scroll]')
  255. await trajectoryScroll.evaluate((host) => {
  256. const measuredWindow = window as Window & { __trajectoryScrollCalls?: number }
  257. measuredWindow.__trajectoryScrollCalls = 0
  258. const original = host.scrollTo.bind(host)
  259. const trackedScrollTo = (...args: [ScrollToOptions?] | [number, number]) => {
  260. measuredWindow.__trajectoryScrollCalls = (measuredWindow.__trajectoryScrollCalls ?? 0) + 1
  261. Reflect.apply(original, host, args)
  262. }
  263. host.scrollTo = trackedScrollTo as typeof host.scrollTo
  264. })
  265. const settled = scaffold.whenTurnSettled()
  266. const input = page.locator('textarea').first()
  267. await input.fill('Stream one deterministic response while Trajectory remains visible.')
  268. await input.press('Enter')
  269. await settled
  270. await page.getByText('stream fragment 01', { exact: false }).waitFor({ timeout: 30_000 })
  271. await nextPaint(page)
  272. const streamingScrollCalls = await trajectoryScroll.evaluate(() => {
  273. return (window as Window & { __trajectoryScrollCalls?: number })
  274. .__trajectoryScrollCalls ?? 0
  275. })
  276. expect(streamingScrollCalls).toBeLessThanOrEqual(5)
  277. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  278. expect({
  279. pageErrors: tripwire.pageErrors,
  280. warnings: tripwire.warnings,
  281. }).toEqual({ pageErrors: [], warnings: [] })
  282. } finally {
  283. releaseHistory()
  284. if (held) await heldRequestFinished
  285. await page.unroute('**/api/session.history')
  286. }
  287. }, 180_000)
  288. })