trajectory-virtualization.e2e.ts 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347
  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 { createChatScrollFixture } from './chat-scroll-fixture.ts'
  14. import {
  15. captureStableAria,
  16. compareOrRefreshGolden,
  17. launchWebScaffold,
  18. seedSession,
  19. watchConsole,
  20. webSnapshotMode,
  21. type WebScaffold,
  22. } from './scaffold.ts'
  23. import { newEnglishPage, saveFailureShot } from './support.ts'
  24. const MODE = webSnapshotMode()
  25. const LOAD_MORE_EXPECTED = fileURLToPath(new URL(
  26. './expected/trajectory-virtualization/load-more.expected.md',
  27. import.meta.url,
  28. ))
  29. const SESSION_ID = 'trajectory-virtualization-e2e'
  30. const FIXTURE = createChatScrollFixture({
  31. markerPrefix: 'TRAJECTORY_VIRTUAL',
  32. title: 'TRAJECTORY_VIRTUAL long ledger',
  33. turns: 88,
  34. })
  35. const MAX_MOUNTED_ROWS = 160
  36. const GEOMETRY_TOLERANCE = 2
  37. const STREAM_MARKER = 'TRAJECTORY_VIRTUAL_STREAM_FINISHED'
  38. const STREAM_TEXT = Array.from(
  39. { length: 80 },
  40. (_, index) => `stream fragment ${String(index + 1).padStart(2, '0')} `,
  41. ).join('') + STREAM_MARKER
  42. const STREAM_CHUNKS: StreamChunk[] = [
  43. { type: 'block-start', index: 0, blockType: 'text' },
  44. ...Array.from({ length: 80 }, (_, index): StreamChunk => ({
  45. type: 'text-delta',
  46. index: 0,
  47. text: `stream fragment ${String(index + 1).padStart(2, '0')} `,
  48. })),
  49. { type: 'text-delta', index: 0, text: STREAM_MARKER },
  50. { type: 'block-end', index: 0, block: { type: 'text', text: STREAM_TEXT } },
  51. { type: 'usage', usage: { inputTokens: 2_700, outputTokens: 240 } },
  52. { type: 'finish', reason: { kind: 'stop' } },
  53. ]
  54. interface ScrollGeometry {
  55. readonly clientHeight: number
  56. readonly scrollHeight: number
  57. readonly scrollTop: number
  58. }
  59. interface RowAnchor {
  60. readonly key: string
  61. readonly top: number
  62. }
  63. async function openSeed(page: Page): Promise<void> {
  64. // Search collapsed into a header action; expand it before filling.
  65. const searchButton = page.getByRole('button', { name: 'Search sessions' })
  66. if (await searchButton.getAttribute('aria-expanded') !== 'true') await searchButton.click()
  67. const search = page.getByRole('textbox', { name: 'Search sessions...', exact: true })
  68. await search.fill(FIXTURE.markers.user(1))
  69. const result = page.getByRole('tree', { name: 'Search results' }).getByRole('treeitem')
  70. await expect.poll(() => result.count(), { timeout: 60_000 }).toBe(1)
  71. await result.click()
  72. await page.getByRole('tab', { name: 'Trajectory', exact: true }).waitFor({ timeout: 30_000 })
  73. await page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false })
  74. .last()
  75. .waitFor({ timeout: 30_000 })
  76. }
  77. async function openTrajectory(page: Page): Promise<void> {
  78. await page.getByRole('tab', { name: 'Trajectory', exact: true }).click()
  79. const pane = page.locator('[data-trajectory-scroll]')
  80. await pane.waitFor({ timeout: 30_000 })
  81. await page.locator('[data-trajectory-scroll] table[data-scroll-ready="true"]')
  82. .waitFor({ timeout: 30_000 })
  83. }
  84. async function logicalRows(page: Page): Promise<number> {
  85. const raw = await page.locator('[data-trajectory-scroll] table').getAttribute('aria-rowcount')
  86. if (raw === null || !/^\d+$/.test(raw)) {
  87. throw new Error(`trajectory table has invalid aria-rowcount ${JSON.stringify(raw)}`)
  88. }
  89. return Number(raw)
  90. }
  91. async function mountedRows(page: Page): Promise<number> {
  92. return page.locator('[data-trajectory-scroll] tr[data-trajectory-row-key]').count()
  93. }
  94. async function geometry(page: Page): Promise<ScrollGeometry> {
  95. return page.locator('[data-trajectory-scroll]').evaluate(host => ({
  96. clientHeight: host.clientHeight,
  97. scrollHeight: host.scrollHeight,
  98. scrollTop: host.scrollTop,
  99. }))
  100. }
  101. async function nextPaint(page: Page): Promise<void> {
  102. await page.evaluate(() => new Promise<void>((resolve) => {
  103. requestAnimationFrame(() => requestAnimationFrame(() => { resolve() }))
  104. }))
  105. }
  106. async function scrollToRatio(page: Page, ratio: number): Promise<void> {
  107. await page.locator('[data-trajectory-scroll]').evaluate((host, value) => {
  108. const maximum = Math.max(0, host.scrollHeight - host.clientHeight)
  109. host.scrollTop = Math.round(maximum * value)
  110. host.dispatchEvent(new Event('scroll'))
  111. }, ratio)
  112. await nextPaint(page)
  113. }
  114. async function firstVisibleRow(page: Page): Promise<RowAnchor> {
  115. return page.locator('[data-trajectory-scroll]').evaluate((host) => {
  116. const hostBox = host.getBoundingClientRect()
  117. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  118. const row = rows.find((candidate) => {
  119. const box = candidate.getBoundingClientRect()
  120. return candidate.dataset.requestOnly !== 'true'
  121. && box.bottom > hostBox.top
  122. && box.top < hostBox.bottom
  123. })
  124. const key = row?.dataset.trajectoryRowKey
  125. if (row === undefined || key === undefined) {
  126. throw new Error('trajectory scrollport has no visible semantic row')
  127. }
  128. return { key, top: row.getBoundingClientRect().top - hostBox.top }
  129. })
  130. }
  131. async function rowTop(page: Page, key: string): Promise<number | null> {
  132. return page.locator('[data-trajectory-scroll]').evaluate((host, targetKey) => {
  133. const rows = [...host.querySelectorAll<HTMLElement>('tr[data-trajectory-row-key]')]
  134. const row = rows.find(candidate => candidate.dataset.trajectoryRowKey === targetKey)
  135. return row === undefined
  136. ? null
  137. : row.getBoundingClientRect().top - host.getBoundingClientRect().top
  138. }, key)
  139. }
  140. async function loadToFirstTurn(page: Page): Promise<void> {
  141. const marker = FIXTURE.markers.user(1)
  142. for (let attempt = 0; attempt < 12; attempt += 1) {
  143. await scrollToRatio(page, 0)
  144. if (await page.getByText(marker, { exact: false }).count() > 0) return
  145. const before = await logicalRows(page)
  146. const anchor = await firstVisibleRow(page)
  147. await expect.poll(async () => ({
  148. marker: await page.getByText(marker, { exact: false }).count() > 0,
  149. rows: await logicalRows(page),
  150. }), { timeout: 30_000 }).not.toEqual({ marker: false, rows: before })
  151. await nextPaint(page)
  152. await expect.poll(async () => {
  153. const top = await rowTop(page, anchor.key)
  154. return top === null ? Number.POSITIVE_INFINITY : Math.abs(top - anchor.top)
  155. }, { timeout: 15_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  156. }
  157. throw new Error('trajectory did not reach the first turn after twelve older-page requests')
  158. }
  159. describe('web e2e: Trajectory virtualization over tail-paged history', () => {
  160. let scaffold: WebScaffold
  161. let browser: Browser
  162. let page: Page
  163. let tripwire: ReturnType<typeof watchConsole>
  164. let replayDir: string
  165. beforeAll(async () => {
  166. replayDir = await mkdtemp(join(tmpdir(), 'dsh-trajectory-virtualization-'))
  167. const replayFixture = join(replayDir, 'session.jsonl')
  168. const replayOverride = join(replayDir, 'replay.override.json')
  169. await writeFile(replayFixture, FIXTURE.log)
  170. await writeFile(replayOverride, JSON.stringify([{
  171. kind: 'chunks',
  172. chunks: STREAM_CHUNKS,
  173. } satisfies ReplayEntry]))
  174. scaffold = await launchWebScaffold({
  175. paceMs: 10,
  176. replayFixture,
  177. replayOverride,
  178. })
  179. await seedSession(scaffold, FIXTURE.log, SESSION_ID)
  180. browser = await chromium.launch()
  181. page = await newEnglishPage(browser, 900)
  182. tripwire = watchConsole(page)
  183. await page.goto(scaffold.baseUrl, { waitUntil: 'load' })
  184. await page.waitForSelector('[class*="frame"]', { timeout: 30_000 })
  185. // The compact layout dropped group session counts; the seeded baseline is
  186. // the Ungrouped bucket once cold summaries load.
  187. await page.getByText('Ungrouped', { exact: true }).waitFor({ timeout: 30_000 })
  188. }, 120_000)
  189. afterAll(async () => {
  190. await browser?.close()
  191. await scaffold?.close()
  192. await rm(replayDir, { recursive: true, force: true })
  193. })
  194. it.skipIf(MODE === 'record')('retains identity on prepend and reaches the bounded virtual range', async () => {
  195. onTestFailed(() => saveFailureShot(page, 'web-e2e-trajectory-virtualization'))
  196. await openSeed(page)
  197. let held = false
  198. let releaseHistory: () => void = () => {}
  199. let finishHeldRequest: () => void = () => {}
  200. const gate = new Promise<void>((resolve) => { releaseHistory = resolve })
  201. const heldRequestFinished = new Promise<void>((resolve) => { finishHeldRequest = resolve })
  202. await page.route('**/api/session/page', async (route) => {
  203. const request = route.request().postDataJSON() as {
  204. method?: string
  205. payload?: { args?: { request?: { beforeSeq?: number } } }
  206. }
  207. if (!held && request.method === 'session/page'
  208. && request.payload?.args?.request?.beforeSeq !== undefined) {
  209. held = true
  210. await gate
  211. try {
  212. await route.continue()
  213. } finally {
  214. finishHeldRequest()
  215. }
  216. return
  217. }
  218. await route.continue()
  219. })
  220. try {
  221. await openTrajectory(page)
  222. const initialRows = await logicalRows(page)
  223. expect(initialRows).toBeGreaterThan(0)
  224. expect(await page.getByText('Initial System Prompt', { exact: true }).count()).toBe(0)
  225. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  226. const loadMore = page.locator('[data-history-load] button')
  227. await loadMore.waitFor({ timeout: 15_000 })
  228. expect(await loadMore.textContent()).toBe('Load earlier history')
  229. const loadMoreSnapshot = await captureStableAria(
  230. page,
  231. '[data-history-load]',
  232. scaffold.workspaceCwd,
  233. )
  234. await compareOrRefreshGolden(LOAD_MORE_EXPECTED, loadMoreSnapshot, MODE)
  235. // Avoid Playwright scrolling the offscreen first row into the automatic-load threshold.
  236. await loadMore.evaluate((button: HTMLButtonElement) => { button.click() })
  237. await expect.poll(() => held, { timeout: 15_000 }).toBe(true)
  238. await expect.poll(async () => ({
  239. disabled: await loadMore.isDisabled(),
  240. label: await loadMore.getAttribute('aria-label'),
  241. }), { timeout: 15_000 }).toEqual({
  242. disabled: true,
  243. label: 'Loading earlier history…',
  244. })
  245. await scrollToRatio(page, 0)
  246. const anchor = await firstVisibleRow(page)
  247. const selectedRow = page.locator(
  248. `[data-trajectory-scroll] tr[data-trajectory-row-key=${JSON.stringify(anchor.key)}]`,
  249. )
  250. await selectedRow.click()
  251. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  252. .toBe('true')
  253. releaseHistory()
  254. await expect.poll(() => logicalRows(page), { timeout: 60_000 }).toBeGreaterThan(initialRows)
  255. await nextPaint(page)
  256. await expect.poll(async () => {
  257. const top = await rowTop(page, anchor.key)
  258. return top === null ? Number.POSITIVE_INFINITY : Math.abs(top - anchor.top)
  259. }, { timeout: 15_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  260. await expect.poll(() => selectedRow.getAttribute('aria-selected'), { timeout: 10_000 })
  261. .toBe('true')
  262. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  263. await loadToFirstTurn(page)
  264. await expect.poll(
  265. () => page.getByText(FIXTURE.markers.user(1), { exact: false }).count(),
  266. { timeout: 10_000 },
  267. ).toBeGreaterThan(0)
  268. const fullRows = await logicalRows(page)
  269. await scrollToRatio(page, 0.5)
  270. const middle = await geometry(page)
  271. const maximum = middle.scrollHeight - middle.clientHeight
  272. expect(middle.scrollTop).toBeGreaterThan(maximum * 0.25)
  273. expect(middle.scrollTop).toBeLessThan(maximum * 0.75)
  274. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  275. expect(await mountedRows(page)).toBeLessThan(fullRows)
  276. await scrollToRatio(page, 1)
  277. await expect.poll(async () => {
  278. const value = await geometry(page)
  279. return value.scrollHeight - value.clientHeight - value.scrollTop
  280. }, { timeout: 10_000 }).toBeLessThanOrEqual(GEOMETRY_TOLERANCE)
  281. await expect.poll(
  282. () => page.getByText(FIXTURE.markers.assistant(FIXTURE.turns), { exact: false }).count(),
  283. { timeout: 10_000 },
  284. ).toBeGreaterThan(0)
  285. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  286. const trajectoryScroll = page.locator('[data-trajectory-scroll]')
  287. await trajectoryScroll.evaluate((host) => {
  288. const measuredWindow = window as Window & { __trajectoryScrollCalls?: number }
  289. measuredWindow.__trajectoryScrollCalls = 0
  290. const original = host.scrollTo.bind(host)
  291. const trackedScrollTo = (...args: [ScrollToOptions?] | [number, number]) => {
  292. measuredWindow.__trajectoryScrollCalls = (measuredWindow.__trajectoryScrollCalls ?? 0) + 1
  293. Reflect.apply(original, host, args)
  294. }
  295. host.scrollTo = trackedScrollTo as typeof host.scrollTo
  296. })
  297. const settled = scaffold.whenTurnSettled()
  298. const input = page.locator('textarea').first()
  299. await input.fill('Stream one deterministic response while Trajectory remains visible.')
  300. await input.press('Enter')
  301. await settled
  302. await page.getByText('stream fragment 01', { exact: false }).waitFor({ timeout: 30_000 })
  303. await nextPaint(page)
  304. const streamingScrollCalls = await trajectoryScroll.evaluate(() => {
  305. return (window as Window & { __trajectoryScrollCalls?: number })
  306. .__trajectoryScrollCalls ?? 0
  307. })
  308. expect(streamingScrollCalls).toBeLessThanOrEqual(5)
  309. expect(await mountedRows(page)).toBeLessThanOrEqual(MAX_MOUNTED_ROWS)
  310. expect({
  311. pageErrors: tripwire.pageErrors,
  312. warnings: tripwire.warnings,
  313. }).toEqual({ pageErrors: [], warnings: [] })
  314. } finally {
  315. releaseHistory()
  316. if (held) await heldRequestFinished
  317. await page.unroute('**/api/session/page')
  318. }
  319. }, 180_000)
  320. })