orca/tests/e2e/codex-composer-echo-latency...

316 lines
10 KiB
TypeScript

import type { Page } from '@stablyai/playwright-test'
export type CodexEchoLatencySample = {
index: number
char: string
/** keydown -> xterm finished parsing the echoed glyph (real echo latency). */
keyToParseMs: number
/** keydown -> xterm renderer painted the row carrying that glyph. */
keyToRenderMs: number | null
}
export type CodexDataLatencySample = {
index: number
kind: 'text' | 'ime'
/** `event.code`: `event.key` is the literal string 'Process' for every Pinyin/Cangjie keydown. */
code: string
/** keydown -> xterm emitted that keystroke's bytes on `onData` (composer-vs-onData delta). */
keyToDataMs: number
data: string
}
export type CodexEchoProbeReport = {
samples: CodexEchoLatencySample[]
/** Per-keystroke keydown->onData deltas, sampled DURING typing rather than after quiescence. */
dataSamples: CodexDataLatencySample[]
keysObserved: number
imeKeysObserved: number
parseEvents: number
dataEvents: number
/** `onData` emissions with no keydown to charge them to (paste, PTY-driven replies). */
unattributedDataEvents: number
renderEvents: number
cols: number
rows: number
}
declare global {
// oxlint-disable-next-line typescript-eslint/consistent-type-definitions -- declaration merging requires interface
interface Window {
__codexEchoProbe?: {
report(): CodexEchoProbeReport
dispose(): void
}
}
}
/**
* Installs an in-renderer latency recorder on the active terminal pane, over two
* independent arms:
* - echo: keydown -> `onWriteParsed` -> `onRender`, i.e. what the screen shows.
* - input: keydown -> `onData`, i.e. how fast the keystroke leaves the composer.
* They answer different questions and neither substitutes for the other.
*
* Why in-page: polling a serialized buffer over CDP adds serialize + IPC +
* poll-granularity cost to every sample, which swamped the signal it measured.
* Timestamps here are taken inside the renderer with performance.now(), so the
* measured window contains no cross-process work at all.
*/
export async function installCodexEchoLatencyProbe(page: Page, target: string): Promise<void> {
await page.evaluate((target) => {
type PendingSample = {
index: number
char: string
expected: string
startedAt: number
parsedAt: number | null
}
type KeyStamp = {
index: number
kind: 'text' | 'ime'
code: string
startedAt: number
charged: boolean
}
const state = window.__store?.getState()
const worktreeId = state?.activeWorktreeId
const tabId =
state?.activeTabType === 'terminal'
? state.activeTabId
: worktreeId
? (state?.activeTabIdByWorktree?.[worktreeId] ?? null)
: null
const manager = tabId ? window.__paneManagers?.get(tabId) : null
const pane = manager?.getActivePane?.() ?? manager?.getPanes?.()[0] ?? null
if (!pane) {
throw new Error('Codex echo probe: no active terminal pane')
}
const terminal = pane.terminal
if (typeof terminal.onWriteParsed !== 'function') {
throw new Error('Codex echo probe: xterm build has no onWriteParsed')
}
if (typeof terminal.onData !== 'function') {
throw new Error('Codex echo probe: xterm build has no onData')
}
const samples: CodexEchoLatencySample[] = []
const dataSamples: CodexDataLatencySample[] = []
const awaitingRender: {
sample: CodexEchoLatencySample
startedAt: number
}[] = []
// Why a queue, not one slot: a slow echo can still be outstanding when the
// next key is pressed, and a single slot silently discards that sample.
const pending: PendingSample[] = []
let lastKey: KeyStamp | null = null
let keysObserved = 0
let imeKeysObserved = 0
let parseEvents = 0
let dataEvents = 0
let unattributedDataEvents = 0
let renderEvents = 0
// Why concatenated without a separator: a composer line that wraps splits the
// token across rows, and trailing-trimmed rows rejoin exactly at the break.
const viewportText = (): string => {
const buffer = terminal.buffer.active
let text = ''
for (let row = 0; row < terminal.rows; row += 1) {
text += buffer.getLine(buffer.viewportY + row)?.translateToString(true) ?? ''
}
return text
}
const observeParse = (): void => {
parseEvents += 1
if (pending.length === 0) {
return
}
const text = viewportText()
// Why drain in order: one parse can land several queued keystrokes at
// once, and each still gets credited against its own keydown timestamp.
while (pending.length > 0 && text.includes(pending[0].expected)) {
const entry = pending.shift()
if (!entry) {
break
}
entry.parsedAt = performance.now()
const sample: CodexEchoLatencySample = {
index: entry.index,
char: entry.char,
keyToParseMs: entry.parsedAt - entry.startedAt,
keyToRenderMs: null
}
samples.push(sample)
awaitingRender.push({ sample, startedAt: entry.startedAt })
}
}
// Why charge to the LATEST keydown and not a FIFO head: composing jamo emit no
// `onData` at all, so a queue would credit a committing keystroke's bytes to the
// first keydown of the syllable and report the whole composition as its latency.
// `charged` keeps only the first emission per key, since a commit that also opens
// the next preedit fires twice for one press.
const observeData = (data: string): void => {
dataEvents += 1
if (!lastKey || lastKey.charged) {
unattributedDataEvents += 1
return
}
lastKey.charged = true
dataSamples.push({
index: lastKey.index,
kind: lastKey.kind,
code: lastKey.code,
keyToDataMs: performance.now() - lastKey.startedAt,
data
})
}
const observeRender = (): void => {
renderEvents += 1
const paintedAt = performance.now()
for (const entry of awaitingRender.splice(0, awaitingRender.length)) {
entry.sample.keyToRenderMs = paintedAt - entry.startedAt
}
}
// Why not `key.length === 1`: an IME keydown carries no character. Pinyin and
// Cangjie report `key: 'Process'` (length 7) on every press, so a length filter
// drops 100% of those runs; the same shape the app itself branches on in
// `src/renderer/src/lib/ime-composition-keyboard-event.ts`.
const isImeKeyDown = (event: KeyboardEvent): boolean =>
event.key === 'Process' || event.keyCode === 229 || event.isComposing
// Why window capture: a listener on an ancestor in the capture phase is
// guaranteed to run before xterm's own keydown handler forwards to the PTY,
// so t0 is stamped before any of the work being measured starts.
const onKeyDown = (event: KeyboardEvent): void => {
const ime = isImeKeyDown(event)
if (!ime && event.key.length !== 1) {
return
}
const startedAt = performance.now()
// Why IME keys stay out of `pending`: the echo arm matches a prefix of `target`,
// and N jamo presses produce one syllable, so per-keystroke prefix matching
// would charge each syllable to the wrong keydown and never drain the queue.
if (ime) {
lastKey = {
index: imeKeysObserved,
kind: 'ime',
code: event.code,
startedAt,
charged: false
}
imeKeysObserved += 1
return
}
if (keysObserved >= target.length) {
return
}
const index = keysObserved
keysObserved += 1
lastKey = {
index,
kind: 'text',
code: event.code,
startedAt,
charged: false
}
pending.push({
index,
char: target[index],
expected: target.slice(0, index + 1),
startedAt,
parsedAt: null
})
}
window.addEventListener('keydown', onKeyDown, { capture: true })
const parsedDisposable = terminal.onWriteParsed(observeParse)
const dataDisposable = terminal.onData(observeData)
const renderDisposable = terminal.onRender(observeRender)
window.__codexEchoProbe = {
report: () => ({
samples: [...samples],
dataSamples: [...dataSamples],
keysObserved,
imeKeysObserved,
parseEvents,
dataEvents,
unattributedDataEvents,
renderEvents,
cols: terminal.cols,
rows: terminal.rows
}),
dispose: () => {
window.removeEventListener('keydown', onKeyDown, { capture: true })
parsedDisposable.dispose()
dataDisposable.dispose()
renderDisposable.dispose()
}
}
}, target)
}
/** Drains every recorded sample in a single round-trip once typing has finished. */
export async function collectCodexEchoLatencyReport(page: Page): Promise<CodexEchoProbeReport> {
return page.evaluate(() => {
const probe = window.__codexEchoProbe
if (!probe) {
throw new Error('Codex echo probe was never installed')
}
const report = probe.report()
probe.dispose()
return report
})
}
export type LatencyDistribution = {
count: number
p50: number
p95: number
max: number
}
function percentile(sorted: number[], quantile: number): number {
if (sorted.length === 0) {
return 0
}
const rank = Math.min(sorted.length - 1, Math.ceil(quantile * sorted.length) - 1)
return sorted[Math.max(0, rank)]
}
/**
* Why this throws instead of returning zeros: an empty sample set used to summarise as
* `p50/p95/max = 0`, i.e. *perfect* latency, so every `toBeLessThan` threshold passed and a run
* that measured nothing looked like the best run ever recorded. The existing consumer is safe only
* because a separate line asserts the sample count. Refusing here makes that guard unnecessary
* rather than load-bearing.
*/
export function summarizeLatencies(values: number[]): LatencyDistribution {
if (values.length === 0) {
throw new Error(
'summarizeLatencies received 0 samples — a distribution over no data would report 0ms ' +
'and pass every latency threshold. Assert the sample count before summarising.'
)
}
const sorted = [...values].sort((a, b) => a - b)
return {
count: sorted.length,
p50: percentile(sorted, 0.5),
p95: percentile(sorted, 0.95),
max: sorted.at(-1) ?? 0
}
}
export function formatDistribution(label: string, distribution: LatencyDistribution): string {
return (
`${label} n=${distribution.count} p50=${distribution.p50.toFixed(1)}ms ` +
`p95=${distribution.p95.toFixed(1)}ms max=${distribution.max.toFixed(1)}ms`
)
}