fix(app): reuse terminal cells during serialization (#46763)

This commit is contained in:
Luke Parker 2026-09-02 18:04:21 +10:00 committed by GitHub
parent dfe3052bb6
commit 499e22bf52
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
9 changed files with 577 additions and 6 deletions

View file

@ -0,0 +1,44 @@
# Native Terminal Benchmark
Manual Windows benchmark. Run only in an isolated development worktree. It does
not connect to an OpenCode service, user profile, or database.
Build from `packages/app` with
`bun x vite build --config e2e/performance/terminals/vite.config.ts`, then freeze
`dist` outside the repository. Set `PLAYWRIGHT_BUILD=1`, `PLAYWRIGHT_BASE_URL` to
an unused loopback URL, `TERMINAL_BUILD` to the frozen build,
`TERMINAL_ARTIFACTS` to an existing external directory, and `TERMINAL_RESULTS`
to an external result directory. Run:
```sh
bun x playwright test --config e2e/performance/terminals/playwright.config.ts --repeat-each=20
```
The runner owns its preview server and each test owns a PowerShell ConPTY process.
Session metadata is deterministic. Native output is forwarded through Playwright's
WebSocket fixture into the real production `Terminal`, writer, Ghostty WASM/canvas,
and serializer. No output is dropped, paused, or delayed. This is native terminal
plus production renderer evidence, not the production PTY backend or Electron IPC.
The workload is 12,000 colored build/test log lines with file paths, durations, and
result descriptions. Cases separate visible output, the same output while hidden,
and closing the session tab after filling the configured scrollback. Ghostty
converts the app's 10,000-line setting to bytes at its initial 80-column width;
resizing reduces the effective row capacity. The report records actual retained
rows and the first retained fixture record rather than assuming 10,000 rows. Completion
requires the final marker in Ghostty and completion of its write callbacks, not
just WebSocket delivery. Teardown requires Home readiness and the final serialized
snapshot. Input, focus, resizing, and native process survival are checked.
`probe.ts` is included only by this benchmark build. It observes actual writes,
renderer calls, and serialization. Chrome `TaskDuration` measures renderer task
time, not total process CPU or RAM. For attribution, set
`OPENCODE_PERFORMANCE_TRACE_DIR`; keep traced runs separate from clean timing.
`TERMINAL_DRAW_PROBE=1` separately counts actual canvas draws to verify hidden
rendering; do not mix these instrumented samples with clean timing.
Use `TERMINAL_REVISION` and `TERMINAL_BUNDLE` to identify frozen artifacts.
`TERMINAL_SCREENSHOTS` captures the visible result after timing.
The benchmark has no machine-dependent performance thresholds. Keep raw logs,
snapshots, traces, and screenshots outside Git. Run heavy work through the
coordinator's exclusive gate when participating in a shared performance wave.

View file

@ -0,0 +1,20 @@
import { defineConfig } from "@playwright/test"
import config from "../../../playwright.config"
export default defineConfig({
...config,
testDir: ".",
testIgnore: [],
testMatch: "terminal-benchmark.spec.ts",
workers: 1,
retries: 0,
timeout: 120_000,
outputDir: process.env.TERMINAL_RESULTS,
reporter: [["line"]],
webServer: {
command: `bun x vite preview --host 127.0.0.1 --port ${new URL(process.env.PLAYWRIGHT_BASE_URL!).port} --strictPort --outDir ${process.env.TERMINAL_BUILD}`,
url: process.env.PLAYWRIGHT_BASE_URL,
reuseExistingServer: false,
},
use: { ...config.use, viewport: { width: 1440, height: 900 }, trace: "off", video: "off", serviceWorkers: "block" },
})

View file

@ -0,0 +1,74 @@
import { Terminal } from "ghostty-web"
import { SerializeAddon } from "../../../src/session/terminal/serialize"
export type TerminalProbe = {
term?: Terminal
writes: number
pending: number
bytes: number
renders: number
hiddenRenders: number
draws: number
hiddenDraws: number
serialized: { ms: number; bytes: number; value: string }[]
}
declare global {
interface Window {
terminalProbe: TerminalProbe
}
}
const probe: TerminalProbe = {
writes: 0,
pending: 0,
bytes: 0,
renders: 0,
hiddenRenders: 0,
draws: 0,
hiddenDraws: 0,
serialized: [],
}
window.terminalProbe = probe
const open = Terminal.prototype.open
Terminal.prototype.open = function (element) {
probe.term = this
open.call(this, element)
// Ghostty does not expose render events. This benchmark-only wrapper observes its
// actual renderer; it does not alter scheduling, parsing, or drawing.
const renderer = (this as unknown as { renderer: { render: (...args: unknown[]) => void } }).renderer
const render = renderer.render
let hidden = false
renderer.render = function (...args) {
probe.renders++
hidden = !element.checkVisibility()
if (hidden) probe.hiddenRenders++
return render.apply(this, args)
}
if (new URL(location.href).searchParams.has("terminalDrawProbe")) {
const context = element.querySelector("canvas")!.getContext("2d")!
const draw = context.drawImage
context.drawImage = function (...args: unknown[]) {
probe.draws++
if (hidden) probe.hiddenDraws++
Reflect.apply(draw, this, args)
}
}
}
const write = Terminal.prototype.write
Terminal.prototype.write = function (data, done) {
probe.writes++
probe.pending++
probe.bytes += typeof data === "string" ? new TextEncoder().encode(data).byteLength : data.byteLength
return write.call(this, data, () => {
probe.pending--
done?.()
})
}
const serialize = SerializeAddon.prototype.serialize
SerializeAddon.prototype.serialize = function (options) {
const start = performance.now()
const value = serialize.call(this, options)
probe.serialized.push({ ms: performance.now() - start, bytes: new TextEncoder().encode(value).byteLength, value })
return value
}

View file

@ -0,0 +1,13 @@
param([Parameter(Mandatory = $true)][string]$Fixture)
$ErrorActionPreference = 'Stop'
[Console]::WriteLine('TERMINAL_FIXTURE_READY')
while ($null -ne ($command = [Console]::ReadLine())) {
if ($command -eq 'exit') { exit 0 }
if ($command -eq 'run') {
foreach ($line in [System.IO.File]::ReadLines($Fixture)) {
[Console]::WriteLine($line)
}
[Console]::WriteLine('TERMINAL_WORKLOAD_DONE')
}
if ($command -eq 'ping') { [Console]::WriteLine('TERMINAL_PROCESS_ALIVE') }
}

View file

@ -0,0 +1,352 @@
import { createRequire } from "node:module"
import { mkdtemp, writeFile, rm } from "node:fs/promises"
import { tmpdir } from "node:os"
import path from "node:path"
import { fileURLToPath } from "node:url"
import type { Page } from "@playwright/test"
import { benchmark, benchmarkDiagnostics, expect } from "../benchmark"
import { mockOpenCodeServer } from "../../utils/mock-server"
import { expectSessionTitle } from "../../utils/waits"
import type {} from "./probe"
// Use the same installed native PTY package as Core, with a fixture-owned process.
const native = createRequire(new URL("../../../../core/package.json", import.meta.url))("@lydell/node-pty") as {
spawn: (
file: string,
args: string[],
options: { cols: number; rows: number; cwd: string },
) => {
pid: number
write: (data: string) => void
resize: (cols: number, rows: number) => void
kill: () => void
onData: (handler: (data: string) => void) => { dispose: () => void }
onExit: (handler: () => void) => { dispose: () => void }
}
}
const sessionID = "ses_terminal_benchmark"
const ptyID = "pty_terminal_benchmark"
const title = "Terminal build output"
const server = process.env.PLAYWRIGHT_BASE_URL!
const href = `/server/${Buffer.from(server).toString("base64url")}/session/${sessionID}`
const lines = Array.from({ length: 12_000 }, (_, i) => {
const unit = ["session/history", "session/runner", "project/discovery", "tool/shell", "provider/stream"][i % 5]
return `\x1b[32mPASS\x1b[0m packages/core/test/${unit}-${String(i).padStart(5, "0")}.test.ts \x1b[2m[${10 + (i % 237)}ms]\x1b[0m validates ordered output and durable recovery`
}).join("\r\n")
benchmark.use({ traceScope: "interaction", viewport: { width: 1440, height: 900 } })
for (const scenario of ["visible-output", "hidden-output", "full-scrollback-teardown"] as const) {
benchmark(scenario, async ({ page, report }, info) => {
const dir = await mkdtemp(path.join(process.env.TERMINAL_ARTIFACTS ?? tmpdir(), "terminal-fixture-"))
await writeFile(path.join(dir, "build.log"), lines)
const pty = native.spawn(
"pwsh.exe",
[
"-NoLogo",
"-NoProfile",
"-NonInteractive",
"-File",
fileURLToPath(new URL("./shell.ps1", import.meta.url)),
"-Fixture",
path.join(dir, "build.log"),
],
{ cols: 120, rows: 24, cwd: dir },
)
const exited = new Promise<void>((resolve) => pty.onExit(resolve))
let output = ""
let connected = 0
let closed = 0
let send: ((data: string) => void) | undefined
const listener = pty.onData((data) => {
output += data
send?.(data)
})
const sizes: { cols: number; rows: number }[] = []
const removals: string[] = []
try {
if (process.env.TERMINAL_DRAW_PROBE) {
await page.addInitScript(() => {
const fill = CanvasRenderingContext2D.prototype.fillText
CanvasRenderingContext2D.prototype.fillText = function (...args: Parameters<typeof fill>) {
if (this.canvas instanceof HTMLCanvasElement && this.canvas.closest('[data-component="terminal"]')) {
window.terminalProbe.draws++
if (!this.canvas.checkVisibility()) window.terminalProbe.hiddenDraws++
}
Reflect.apply(fill, this, args)
}
})
}
const location = { directory: dir, project: { id: "proj_terminal_benchmark", directory: dir } }
const data = {
id: ptyID,
title: "Terminal 1",
command: "pwsh.exe",
args: [],
cwd: dir,
status: "running",
pid: pty.pid,
}
await mockOpenCodeServer(page, {
directory: dir,
project: {
id: location.project.id,
worktree: dir,
vcs: "git",
name: "terminal-benchmark",
time: { created: 1, updated: 1 },
sandboxes: [],
},
provider: {
all: [
{
id: "opencode",
name: "OpenCode",
models: { test: { id: "test", name: "Test", limit: { context: 200_000 } } },
},
],
connected: ["opencode"],
default: { providerID: "opencode", modelID: "test" },
},
sessions: [
{
id: sessionID,
slug: sessionID,
projectID: location.project.id,
directory: dir,
title,
version: "dev",
time: { created: 1700000000000, updated: 1700000000000 },
},
],
pageMessages: () => ({ items: [] }),
})
await page.route("**/api/pty**", async (route) => {
if (route.request().method() === "DELETE") removals.push(route.request().url())
const body = route.request().postDataJSON()
if (body?.size) {
sizes.push(body.size)
pty.resize(body.size.cols, body.size.rows)
}
return route.fulfill({
status: 200,
contentType: "application/json",
body: JSON.stringify({
location,
data: route.request().url().includes("connect-token") ? { ticket: "fixture", expires_in: 60 } : data,
}),
})
})
await page.routeWebSocket(new RegExp(`/api/pty/${ptyID}/connect`), (socket) => {
connected++
send = (data) => socket.send(data)
socket.send(output.slice(Number(new URL(socket.url()).searchParams.get("cursor") ?? 0)))
socket.onMessage((data) => pty.write(String(data)))
socket.onClose(() => {
closed++
send = undefined
})
})
await page.addInitScript(
({ server, sessionID }) => {
localStorage.setItem("settings.v3", JSON.stringify({ general: { terminalPlacement: "bottom" } }))
localStorage.setItem(
"opencode.window.browser.dat:tabs",
JSON.stringify([{ type: "session", server, sessionId: sessionID }]),
)
},
{ server, sessionID },
)
await page.goto(
`${href}${process.env.TERMINAL_DRAW_PROBE && scenario !== "full-scrollback-teardown" ? "?terminalDrawProbe" : ""}`,
)
await expectSessionTitle(page, title)
await page.keyboard.press("Control+Backquote")
await waitForText(page, "TERMINAL_FIXTURE_READY")
const terminal = page.locator('[data-component="terminal"]')
await expect(terminal).toBeVisible()
await page.evaluate(() => document.fonts.ready.then(() => undefined))
await expect
.poll(async () => {
const size = await page.evaluate(() => ({
cols: window.terminalProbe.term!.cols,
rows: window.terminalProbe.term!.rows,
}))
return sizes.at(-1)?.cols === size.cols && sizes.at(-1)?.rows === size.rows
})
.toBe(true)
if (scenario === "hidden-output") {
await page.keyboard.press("Control+Backquote")
await expect(terminal).toBeHidden()
}
const cdp = await page.context().newCDPSession(page)
await cdp.send("Performance.enable")
const before = await cdp.send("Performance.getMetrics")
const start = await page.evaluate(() => {
window.terminalProbe.renders = 0
window.terminalProbe.hiddenRenders = 0
window.terminalProbe.draws = 0
window.terminalProbe.hiddenDraws = 0
return performance.now()
})
await benchmarkDiagnostics(page).startTrace()
// The producer is not throttled. The visible and hidden cases receive the same bytes.
pty.write("run\r")
await waitForText(page, "TERMINAL_WORKLOAD_DONE")
const produced = await page.evaluate(
(start) => ({
ms: performance.now() - start,
renders: window.terminalProbe.renders,
hiddenRenders: window.terminalProbe.hiddenRenders,
draws: window.terminalProbe.draws,
hiddenDraws: window.terminalProbe.hiddenDraws,
bytes: window.terminalProbe.bytes,
scrollback: window.terminalProbe.term!.getScrollbackLength(),
cols: window.terminalProbe.term!.cols,
rows: window.terminalProbe.term!.rows,
firstRecord: Number(
window.terminalProbe
.term!.buffer.normal.getLine(0)
?.translateToString(true)
.match(/-(\d{5})\.test\.ts/)?.[1],
),
}),
start,
)
const after = await cdp.send("Performance.getMetrics")
const cpuMs =
(after.metrics.find((x) => x.name === "TaskDuration")!.value -
before.metrics.find((x) => x.name === "TaskDuration")!.value) *
1000
let interaction: Record<string, unknown> = {}
if (scenario === "hidden-output") {
const start = await page.evaluate(() => performance.now())
await page.keyboard.press("Control+Backquote")
await expect(terminal).toBeVisible()
await waitForText(page, "TERMINAL_WORKLOAD_DONE")
interaction = { returnMs: await page.evaluate((start) => performance.now() - start, start) }
}
if (scenario === "full-scrollback-teardown") {
// Ghostty converts the configured line limit to bytes at the initial
// 80-column size. Resizing changes the effective retained row count.
expect(produced.scrollback).toBeGreaterThan(0)
expect(produced.firstRecord).toBeGreaterThan(0)
expect(produced.firstRecord).toBeLessThan(11_999)
const close = page.locator(`[data-titlebar-tab-slot]:has(a[href="${href}"]) [data-component="icon-button-v2"]`)
await expect(close).toBeVisible()
const cpuBefore = await cdp.send("Performance.getMetrics")
const start = await page.evaluate(() => performance.now())
await close.click()
await expect(page).toHaveURL("/")
await expect(page.locator('[data-component="home-session-search"]')).toBeVisible()
await expect(page.locator('[data-component="home-session-search"] input')).toBeEditable()
await expect.poll(() => page.evaluate(() => window.terminalProbe.serialized.length)).toBe(1)
interaction = await page.evaluate(
(start) => ({
homeReadyMs: performance.now() - start,
serializeMs: window.terminalProbe.serialized[0].ms,
serializedBytes: window.terminalProbe.serialized[0].bytes,
}),
start,
)
const cpuAfter = await cdp.send("Performance.getMetrics")
interaction.teardownCpuMs =
(cpuAfter.metrics.find((x) => x.name === "TaskDuration")!.value -
cpuBefore.metrics.find((x) => x.name === "TaskDuration")!.value) *
1000
const snapshot = await page.evaluate(() => window.terminalProbe.serialized[0].value)
expect(Array.from(snapshot.matchAll(/-(\d{5})\.test\.ts/g), (match) => Number(match[1]))).toEqual(
Array.from({ length: 12_000 - produced.firstRecord }, (_, index) => produced.firstRecord + index),
)
expect(snapshot).toContain("TERMINAL_WORKLOAD_DONE")
await writeFile(
path.join(
process.env.TERMINAL_ARTIFACTS ?? tmpdir(),
`${process.env.TERMINAL_BUNDLE}-${info.repeatEachIndex}.ansi`,
),
snapshot,
)
await expect(terminal).toHaveCount(0)
expect(closed).toBe(1)
// UI teardown must not terminate the native process.
pty.write("ping\r")
await expect.poll(() => output.includes("TERMINAL_PROCESS_ALIVE")).toBe(true)
}
await benchmarkDiagnostics(page).stop()
expect(connected).toBe(1)
expect(removals).toEqual([])
expect(sizes.length).toBeGreaterThan(0)
report(
{ ...produced, cpuMs, ...interaction },
{
revision: process.env.TERMINAL_REVISION,
bundle: process.env.TERMINAL_BUNDLE,
fixtureBytes: Buffer.byteLength(lines),
fixtureLines: 12_000,
transport: "Windows ConPTY -> Playwright WebSocket fixture -> production Terminal/writer/Ghostty",
scope: "Chromium renderer; not Electron total RAM or production backend IPC",
},
)
if (scenario !== "full-scrollback-teardown") {
// Validate input, focus, and resize after both visible and hidden output.
await terminal.click()
await expect(terminal.locator("textarea")).toBeFocused()
await page.keyboard.type("ping")
await page.keyboard.press("Enter")
await waitForText(page, "TERMINAL_PROCESS_ALIVE")
const columns = await page.evaluate(() => window.terminalProbe.term!.cols)
await page.setViewportSize({ width: 1100, height: 800 })
await expect.poll(() => page.evaluate(() => window.terminalProbe.term!.cols)).not.toBe(columns)
await expect
.poll(async () => sizes.at(-1)?.cols === (await page.evaluate(() => window.terminalProbe.term!.cols)))
.toBe(true)
expect(closed).toBe(0)
}
if (process.env.TERMINAL_SCREENSHOTS && scenario !== "full-scrollback-teardown") {
await page.screenshot({
path: path.join(process.env.TERMINAL_SCREENSHOTS, `${scenario}-${info.repeatEachIndex}.png`),
})
}
} finally {
listener.dispose()
try {
await benchmarkDiagnostics(page).stop()
// Stop fixture request handlers before killing their native resource. The
// app debounces PTY resize requests independently of the canvas resize.
await page.unrouteAll({ behavior: "wait" })
await page.close()
} finally {
pty.kill()
await exited
await writeFile(
path.join(
process.env.TERMINAL_ARTIFACTS ?? tmpdir(),
`${process.env.TERMINAL_BUNDLE}-${scenario}-${info.repeatEachIndex}.native.log`,
),
output,
)
await rm(dir, { recursive: true, force: true })
}
}
})
}
async function waitForText(page: Page, text: string) {
await expect
.poll(() =>
page.evaluate((text) => {
const probe = window.terminalProbe
const term = probe?.term
if (!term || probe.pending !== 0) return false
const buffer = term.buffer.active
return Array.from(
{ length: term.rows },
(_, i) => buffer.getLine(buffer.length - term.rows + i)?.translateToString(true) ?? "",
)
.join("\n")
.includes(text)
}, text),
)
.toBe(true)
}

View file

@ -0,0 +1,17 @@
import { mergeConfig } from "vite"
import config from "../../../vite.config"
// The probe is included only in this manual benchmark build, never in the app build.
export default mergeConfig(config, {
plugins: [
{
name: "terminal-benchmark-probe",
transformIndexHtml: {
order: "pre",
handler: () => [
{ tag: "script", attrs: { type: "module", src: "/e2e/performance/terminals/probe.ts" }, injectTo: "head" },
],
},
},
],
})

View file

@ -1,4 +1,4 @@
import { describe, test, expect, beforeAll, afterEach } from "bun:test"
import { describe, test, expect, beforeAll, afterEach, spyOn } from "bun:test"
import { Terminal, Ghostty } from "ghostty-web"
import { SerializeAddon } from "./serialize"
@ -37,6 +37,32 @@ function writeAndWait(term: Terminal, data: string): Promise<void> {
}
describe("SerializeAddon", () => {
test("reuses cells across style changes without rereading their rows", async () => {
const { term, addon } = createTerminal(20, 5)
await writeAndWait(term, "\x1b[31mred\x1b[32mgreen\x1b[0m\r\n\x1b[1;44mbold\x1b[0m")
const reads = spyOn(term.buffer.normal, "getLine")
const serialized = addon.serialize({ range: { start: 0, end: 1 } })
expect(reads.mock.calls.map((args) => args[0])).toEqual([0, 1, 1])
reads.mockRestore()
const restored = createTerminal(20, 5)
await writeAndWait(restored.term, serialized)
for (let row = 0; row < 2; row++) {
const original = term.buffer.normal.getLine(row)!
const replayed = restored.term.buffer.normal.getLine(row)!
expect(replayed.translateToString()).toBe(original.translateToString())
for (let col = 0; col < 20; col++) {
const before = original.getCell(col)!
const after = replayed.getCell(col)!
expect([after.getFgColor(), after.getBgColor(), after.isBold()]).toEqual([
before.getFgColor(),
before.getBgColor(),
before.isBold(),
])
}
}
})
test("preserves color scheme reporting mode", async () => {
const { term, addon } = createTerminal()
await writeAndWait(term, "\x1b[?2031h")

View file

@ -442,11 +442,8 @@ class StringSerializeHandler extends BaseSerializeHandler {
this._currentRow += `\u001b[${sgrSeq.join(";")}m`
const line = this._buffer.getLine(row)
const cellFromLine = line?.getCell(col)
if (cellFromLine) {
this._cursorStyle = cellFromLine
}
// Ghostty cells are snapshots; rereading this cell copies the whole row again.
this._cursorStyle = cell
}
if (isEmptyCell) {

View file

@ -61,4 +61,32 @@ describe("terminalWriter", () => {
done?.()
expect(settled).toBe(true)
})
test("final flush preserves output queued behind an in-flight write", () => {
const scheduled: VoidFunction[] = []
const completions: VoidFunction[] = []
const events: string[] = []
const writer = terminalWriter(
(data, done) => {
events.push(data)
if (done) completions.push(done)
},
(flush) => scheduled.push(flush),
)
writer.push("build started\r\n")
scheduled.shift()?.()
writer.push("\x1b[32mPASS\x1b[0m ")
writer.push("session/history.test.ts\r\n")
writer.flush(() => events.push("persist and dispose"))
expect(events).toEqual(["build started\r\n"])
completions.shift()?.()
scheduled.shift()?.()
expect(events).toEqual(["build started\r\n", "\x1b[32mPASS\x1b[0m session/history.test.ts\r\n"])
completions.shift()?.()
expect(events).toEqual([
"build started\r\n",
"\x1b[32mPASS\x1b[0m session/history.test.ts\r\n",
"persist and dispose",
])
})
})