rnxsim 0.1.407 → 0.1.409

package/cli/drivers/playwright-sim-host.ts

@@ -0,0 +1,501 @@
+// the detached browser host, as CJS source.
+//
+// it resolves playwright from the absolute path the driver hands it, launches
+// Chrome for Testing, opens the target url, then keeps itself (and therefore
+// the browser) alive until the browser disconnects. all inputs arrive through
+// the environment.
+//
+// it stays a source string rather than a module of its own: playwright is
+// resolved from the user's project at runtime and is not a dependency here, so
+// there are no types for any of this, and a typed rewrite would be `unknown`
+// casts end to end. cli/internal-child.ts runs it through node:vm with a real
+// require, in a child process the CLI spawns as itself.
+import { DISPOSABLE_BROWSER_CACHE_DIR_NAMES } from '../../src/browser-cache'
+
+export const PLAYWRIGHT_SIM_HOST = `
+const modulePath = process.env.SOOTSIM_PW_MODULE;
+const url = process.env.SOOTSIM_PW_URL;
+// the WS bridge port the sim page must register with, injected below as
+// window.__sootsimBridgePort so the page skips its shell-port heuristic (which
+// assumes the default offset and breaks when the dev bridge drifted).
+const bridgePort = process.env.SOOTSIM_PW_BRIDGE_PORT || '';
+const headless = process.env.SOOTSIM_PW_HEADLESS === '1';
+const callerUserDataDir = process.env.SOOTSIM_PW_USERDATADIR || '';
+// pid of the session that launched us (its topmost non-init ancestor). when
+// that process exits we tear Chrome down — see startOwnerWatch below. it is a
+// required lifetime bound: an invalid value aborts before Chrome launches.
+// SOOTSIM_PW_OWNER_POLL_MS is a test-only override of the 5s poll interval;
+// production never sets it.
+const ownerPid = Number(process.env.SOOTSIM_PW_OWNER_PID || 0);
+const ownerPollMs = Number(process.env.SOOTSIM_PW_OWNER_POLL_MS || 5000);
+// optional CDP port — exposes Chrome's remote-debugging endpoint so a
+// bridge-driven sim can also be cpu-profiled (rnx perf cpu --cdp-port).
+// playwright keeps its own pipe connection; the port is an additional endpoint.
+const cdpPort = process.env.SOOTSIM_PW_CDP_PORT || '';
+// extra chromium launch args, space-separated. a linux GPU runner injects
+// '--use-angle=vulkan --enable-features=Vulkan' here so chrome binds the
+// real GPU instead of silently falling back to swiftshader software
+// rendering (which pegs cores and produces ~1fps canvas capture).
+const extraChromeArgs = (process.env.SOOTSIM_PW_CHROME_ARGS || '')
+  .split(' ')
+  .filter(Boolean);
+// Local CanvasKit work must never fall back to Playwright's software
+// renderer. CI may use the bundled browser for non-interactive coverage, but
+// local test drives fail closed unless the hardware-backed test browser is
+// proven after launch.
+const hardwareGpuRequired = process.env.CI !== '1' && process.env.CI !== 'true';
+const softwareRendererPattern = /swiftshader|software(?: rasterizer)?|llvmpipe|lavapipe|softpipe/i;
+// page viewport as '<w>x<h>'. the caller sizes it from the device profile;
+// the fallback is phone-shaped because the page hosts a phone sim — the
+// playwright default (1280x720 desktop) letterboxes the device frame.
+const viewportSpec = /^(\\d+)x(\\d+)$/.exec(process.env.SOOTSIM_PW_VIEWPORT || '');
+const viewport = viewportSpec
+  ? { width: Number(viewportSpec[1]), height: Number(viewportSpec[2]) }
+  : { width: 600, height: 900 };
+// device pixels per css pixel. playwright's default is 1, which makes the
+// engine allocate its canvas backing store at HALF the resolution a normal
+// retina-mac chrome would (dpr 2) — the sim looks visibly soft. default to 2
+// so a driver-launched sim matches what the user sees in their own browser.
+const deviceScaleFactor = Number(process.env.SOOTSIM_PW_DSF || 2);
+const closeTimeoutMs = Number(process.env.SOOTSIM_PW_CLOSE_TIMEOUT_MS || 2500);
+const connectAckFile = process.env.SOOTSIM_PW_CONNECTED_ACK_FILE || '';
+const disposableBrowserCacheDirs = new Set(${JSON.stringify(
+  DISPOSABLE_BROWSER_CACHE_DIR_NAMES,
+)});
+const { existsSync, mkdtempSync, readdirSync, rmSync, unlinkSync } = require('fs');
+const { execSync } = require('child_process');
+const { tmpdir } = require('os');
+const { join } = require('path');
+// chrome tree reaper — playwright's browser.close() and even SIGKILLing the
+// root chrome process do NOT reliably take down the helper tree (gpu, renderer,
+// network/storage utilities). they reparent to launchd/init and survive, each
+// one busy enough to peg CPU at 100%+. close() only reaches the root.
+//
+// the prior approach was \`pgrep -f ms-playwright/[c]hromium\`. it never matched
+// system Chrome, so the reaper was a no-op in the common case and every
+// aborted/closed \`rnx open\` leaked a renderer + gpu helper burning 100%+
+// CPU until manually killed (multiple incidents 2026-06).
+//
+// fix: identify our chrome tree by its UNIQUE user-data-dir. every helper a
+// chrome root spawns inherits \`--user-data-dir=<dir>\` in its cmdline, so
+// \`pgrep -f <profileDir>\` finds exactly the tree we launched and nothing else
+// — Chrome for Testing and bundled playwright chromium alike. as a side benefit we
+// no longer need the launch / launchPersistentContext fork: we ALWAYS use a
+// profile dir (mkdtemp'd if the caller didn't pass one, then removed at exit).
+const reapEnabled = process.platform !== 'win32';
+let profileDir = callerUserDataDir;
+let ephemeralProfile = false;
+if (!profileDir) {
+  profileDir = mkdtempSync(join(tmpdir(), 'rnx-playwright-host-'));
+  ephemeralProfile = true;
+}
+// browser HTTP caches are acceleration data, not profile identity. give each
+// host a bounded temporary cache so persistent named profiles retain cookies
+// and storage without multiplying a private cache for every profile.
+const diskCacheDir = mkdtempSync(join(tmpdir(), 'rnx-playwright-cache-'));
+const chromePidsByProfile = () => {
+  if (!reapEnabled) return [];
+  try {
+    return execSync('pgrep -f ' + profileDir + ' || true', { stdio: ['ignore', 'pipe', 'ignore'] })
+      .toString().split('\\n').filter(Boolean).map((p) => Number(p))
+      .filter((p) => p && p !== process.pid);
+  } catch { return []; }
+};
+const chromeCommandsByProfile = () => {
+  if (!reapEnabled) return [];
+  const pids = chromePidsByProfile();
+  if (pids.length === 0) return [];
+  try {
+    return execSync('ps -p ' + pids.join(',') + ' -o command=', {
+      stdio: ['ignore', 'pipe', 'ignore'],
+    }).toString().split('\\n').filter(Boolean);
+  } catch { return []; }
+};
+const assertNoUnsafeSwiftShaderFlag = () => {
+  if (!hardwareGpuRequired) return;
+  const unsafe = chromeCommandsByProfile().find((command) =>
+    command.includes('--enable-unsafe-swiftshader') ||
+    /--use-angle=(?:swiftshader|swiftshader-webgl)(?:\\s|$)/i.test(command)
+  );
+  if (unsafe) {
+    throw new Error(
+      '[playwright-driver] refused software-rendering Chrome: launched process contains a SwiftShader flag'
+    );
+  }
+};
+const killTree = (pid) => {
+  let kids = [];
+  try {
+    kids = execSync('pgrep -P ' + pid + ' || true', { stdio: ['ignore', 'pipe', 'ignore'] })
+      .toString().split('\\n').filter(Boolean);
+  } catch {}
+  for (const k of kids) killTree(Number(k));
+  try { process.kill(pid, 'SIGKILL'); } catch {}
+};
+const reapChrome = () => { for (const pid of chromePidsByProfile()) killTree(pid); };
+const removeDisposableCaches = (root) => {
+  let children = [];
+  try { children = readdirSync(root, { withFileTypes: true }); } catch { return; }
+  for (const child of children) {
+    if (!child.isDirectory() || child.isSymbolicLink()) continue;
+    const childPath = join(root, child.name);
+    if (disposableBrowserCacheDirs.has(child.name)) {
+      try { rmSync(childPath, { recursive: true, force: true }); } catch {}
+      continue;
+    }
+    removeDisposableCaches(childPath);
+  }
+};
+const cleanupProfile = () => {
+  try { rmSync(diskCacheDir, { recursive: true, force: true }); } catch {}
+  if (ephemeralProfile) {
+    try { rmSync(profileDir, { recursive: true, force: true }); } catch {}
+    return;
+  }
+  removeDisposableCaches(profileDir);
+};
+(async () => {
+  const pw = require(modulePath);
+  let context;
+  let keepAlive = null;
+  let connectTimer = null;
+  let connectAckPoll = null;
+  let ownerWatch = null;
+  let connected = false;
+  let shuttingDown = false;
+  const clearConnectTimer = () => {
+    if (connectTimer) {
+      clearTimeout(connectTimer);
+      connectTimer = null;
+    }
+    if (connectAckPoll) {
+      clearInterval(connectAckPoll);
+      connectAckPoll = null;
+    }
+  };
+  const markConnected = () => {
+    connected = true;
+    clearConnectTimer();
+  };
+  const ackFileExists = () => {
+    if (!connectAckFile) return false;
+    try { return existsSync(connectAckFile); } catch { return false; }
+  };
+  const bounded = async (label, work, timeoutMs) => {
+    let timer = null;
+    try {
+      const timeout = new Promise((resolve) => {
+        timer = setTimeout(() => resolve({ timedOut: true }), timeoutMs);
+      });
+      if (typeof timer.unref === 'function') timer.unref();
+      const result = await Promise.race([
+        Promise.resolve().then(work).then(
+          () => ({ timedOut: false }),
+          (error) => ({ timedOut: false, error }),
+        ),
+        timeout,
+      ]);
+      if (result && result.timedOut) {
+        process.stderr.write('[shutdown-timeout] ' + label + ' exceeded ' + timeoutMs + 'ms\\n');
+        return false;
+      }
+      if (result && result.error) {
+        process.stderr.write('[shutdown-error] ' + label + ': ' + String((result.error && result.error.message) || result.error) + '\\n');
+        return false;
+      }
+      return true;
+    } finally {
+      if (timer) clearTimeout(timer);
+    }
+  };
+  const browserProcess = () => {
+    if (context && typeof context.browser === 'function') {
+      try {
+        const b = context.browser();
+        if (b && typeof b.process === 'function') return b.process();
+      } catch {}
+    }
+    return null;
+  };
+  const processIsAlive = (proc) => {
+    if (!proc || !proc.pid) return false;
+    try {
+      process.kill(proc.pid, 0);
+      return true;
+    } catch (err) {
+      return err && err.code === 'EPERM';
+    }
+  };
+  const terminateBrowserProcess = async () => {
+    const proc = browserProcess();
+    if (!proc || !proc.pid || !processIsAlive(proc)) return;
+    try { proc.kill('SIGTERM'); } catch {}
+    await bounded('browser process exit after SIGTERM', async () => {
+      while (processIsAlive(proc)) {
+        await new Promise((resolve) => setTimeout(resolve, 50));
+      }
+    }, 600);
+    if (!processIsAlive(proc)) return;
+    try { proc.kill('SIGKILL'); } catch {}
+  };
+  const shutdown = async (code = 0) => {
+    if (shuttingDown) return;
+    shuttingDown = true;
+    clearConnectTimer();
+    if (keepAlive) clearInterval(keepAlive);
+    if (ownerWatch) clearInterval(ownerWatch);
+    const timeoutMs = Number.isFinite(closeTimeoutMs) && closeTimeoutMs > 0 ? closeTimeoutMs : 2500;
+    if (context) await bounded('context.close', () => context.close(), timeoutMs);
+    await terminateBrowserProcess();
+    // close() + root-SIGKILL above handle the graceful case; the helper tree
+    // (gpu/renderer/utilities) reparents and survives every one of those, so
+    // we always sweep by our unique profile dir as the load-bearing teardown.
+    reapChrome();
+    cleanupProfile();
+    if (connectAckFile) {
+      try { unlinkSync(connectAckFile); } catch {}
+    }
+    process.exit(code);
+  };
+  process.once('SIGTERM', () => { void shutdown(0); });
+  process.once('SIGINT', () => { void shutdown(0); });
+  // the launch-failure path at the bottom of this script exits without going
+  // through shutdown(); the exit hook (sync sweep) still reaps the tree.
+  process.on('exit', () => { reapChrome(); cleanupProfile(); });
+  // owner-liveness watchdog — the load-bearing fix for leaked Chrome hosts.
+  // we are spawned detached + unref'd so \`rnx open\` returns immediately,
+  // which means nothing in the OS process tree ties Chrome's lifetime to the
+  // session that asked for it. once the sim connects, the connect-timeout
+  // safety net is cleared, so absent an explicit \`rnx close\` (or the
+  // daemon's idle reaper closing the sim socket) the host would hold Chrome
+  // open forever after its owner exits. poll the owning session's pid and tear
+  // down when it's gone. ESRCH (kill throws) = owner exited; EPERM = alive but
+  // not signalable by us, which still counts as alive.
+  const startOwnerWatch = () => {
+    if (!Number.isInteger(ownerPid) || ownerPid <= 1) {
+      throw new Error('[playwright-driver] missing valid owning process; refusing unsupervised browser host');
+    }
+    const intervalMs = Number.isFinite(ownerPollMs) && ownerPollMs > 0 ? ownerPollMs : 5000;
+    ownerWatch = setInterval(() => {
+      try {
+        process.kill(ownerPid, 0);
+      } catch (err) {
+        if (err && err.code === 'EPERM') return;
+        process.stderr.write('[owner-gone] owning process ' + ownerPid + ' exited; closing playwright host\\n');
+        void shutdown(0);
+      }
+    }, intervalMs);
+    if (typeof ownerWatch.unref === 'function') ownerWatch.unref();
+  };
+  startOwnerWatch();
+  // Local rendering uses Playwright's full Chrome for Testing app. On macOS its
+  // com.google.chrome.for.testing identity cannot intercept open/quit events for
+  // the user's com.google.Chrome browser. CI may use chromium_headless_shell;
+  // local rendering fails closed because that shell has no GPU and silently
+  // chooses SwiftShader. NOTE: this is NOT enough to make \`rnx perf cpu\`
+  // work — that profiles the tenant *worker*, and the JS Self-Profiler API
+  // cannot be constructed in a dedicated worker (see
+  // worker-handlers/cpu-profile.ts). cpu-profile needs a CDP-based profiler
+  // instead.
+  if (
+    hardwareGpuRequired &&
+    extraChromeArgs.some((arg) =>
+      arg === '--enable-unsafe-swiftshader' ||
+      /^--use-angle=(?:swiftshader|swiftshader-webgl)$/i.test(arg)
+    )
+  ) {
+    throw new Error('[playwright-driver] refused SwiftShader launch arguments');
+  }
+  // one browser path per environment: the full, separately identified test app
+  // locally; chromium_headless_shell on CI. never launch a branded user browser.
+  const channelsToTry = hardwareGpuRequired ? ['chromium'] : [null];
+  const launchWith = async (channel) => {
+    const opts = channel
+      ? { headless, channel, viewport, deviceScaleFactor }
+      : { headless, viewport, deviceScaleFactor };
+    // playwright injects --force-color-profile=srgb by default, which reads
+    // visibly washed-out next to a normal chrome window on a wide-gamut mac
+    // display. strip it so driver-launched sims color-match the user's own
+    // browser (real incident 2026-07-09: "low res and whitewashed" sim).
+    opts.ignoreDefaultArgs = ['--force-color-profile=srgb'];
+    // headless chromium backgrounds/occludes the renderer and throttles rAF +
+    // timers to ~1fps, which silently starves combined/video frame capture
+    // (was 47 frames over 63s -> frozen-looking mp4). keep the render loop at
+    // full rate so the recorder captures real motion. document.hidden stays
+    // false in headless, so assertTabVisibleForCapture never caught this — it
+    // has to be fixed at the launch flags.
+    opts.args = [
+      // the compositor renders through skia graphite on webgpu, so a launch
+      // without navigator.gpu boots to a fatal instead of a sim.
+      '--enable-unsafe-webgpu',
+      '--disk-cache-dir=' + diskCacheDir,
+      '--disk-cache-size=67108864',
+      '--disable-background-timer-throttling',
+      '--disable-backgrounding-occluded-windows',
+      '--disable-renderer-backgrounding',
+    ];
+    if (channel) {
+      // the full test browser can bind the real platform GPU — force the native
+      // ANGLE backend and strip playwright's --enable-unsafe-swiftshader
+      // default arg. without both, headless chrome silently falls back to
+      // swiftshader software rendering for the CanvasKit/WebGL canvas and
+      // pegs every core. SOOTSIM_PW_CHROME_ARGS is appended later, so an
+      // env-injected --use-angle=... still overrides (last flag wins).
+      const angle = process.platform === 'darwin' ? 'metal'
+        : process.platform === 'linux' ? 'vulkan' : 'default';
+      opts.args.unshift(
+        '--use-angle=' + angle,
+        '--enable-gpu',
+        '--enable-gpu-rasterization',
+        '--enable-zero-copy',
+        '--ignore-gpu-blocklist',
+      );
+      opts.ignoreDefaultArgs.push('--enable-unsafe-swiftshader');
+    } else {
+      // playwright's bundled software chromium exposes no navigator.gpu until
+      // Vulkan is named explicitly, and webgpu is the only backend there is.
+      opts.args.push('--enable-features=Vulkan');
+    }
+    if (cdpPort) opts.args.push('--remote-debugging-port=' + cdpPort);
+    if (extraChromeArgs.length) opts.args.push(...extraChromeArgs);
+    // headed: size the OS window to the (phone-shaped) viewport too, plus
+    // room for chrome's own toolbar, so the window isn't desktop-shaped
+    // around a letterboxed page.
+    if (!headless) opts.args.push('--window-size=' + viewport.width + ',' + (viewport.height + 88));
+    // ALWAYS persistent: every chrome process inherits --user-data-dir=<dir>
+    // in its cmdline, which is how reapChrome() finds the whole tree (root +
+    // helpers) on shutdown. without a unique dir there is no way to attribute
+    // an orphaned helper back to us, which is the leak we just fixed.
+    return await pw.chromium.launchPersistentContext(profileDir, opts);
+  };
+  let launchErr = null;
+  let launchedChannel = null;
+  for (const channel of channelsToTry) {
+    try {
+      context = await launchWith(channel);
+      launchedChannel = channel || 'chromium';
+      launchErr = null;
+      break;
+    } catch (err) {
+      launchErr = err;
+      const msg = err && err.message ? String(err.message) : '';
+      const isMissingChannel =
+        /not found at/i.test(msg) ||
+        /not installed/i.test(msg) ||
+        /Chromium distribution/i.test(msg);
+      if (!isMissingChannel) throw err; // unexpected failure: bubble up
+      if (channel) {
+        const next = hardwareGpuRequired
+          ? '; bundled Chromium is disabled for local GPU work\\n'
+          : ', falling back to bundled chromium\\n';
+        process.stderr.write('[playwright-driver] channel "' + channel + '" not available' + next);
+      }
+    }
+  }
+  if (launchErr) throw launchErr;
+  if (launchedChannel) {
+    process.stderr.write('[playwright-driver] launched ' + launchedChannel + '\\n');
+  }
+  // grant camera up front so the sim's Camera app reaches the host webcam
+  // through getUserMedia, mirroring a real iOS simulator where the camera just
+  // works. without this the unhandled chrome permission prompt auto-denies and
+  // CameraApp renders its "Camera Access Required" placeholder.
+  await context.grantPermissions(['camera']);
+  const hostMeta = {
+    sootsimHostDriver: 'playwright',
+    sootsimHostPid: process.pid,
+    sootsimHostOwnerPid: ownerPid,
+    sootsimHostStartedAt: Date.now(),
+  };
+  // inject the resolved bridge port (numeric literal) BEFORE the page runs, so
+  // the shell/engine talk to the daemon this driver owns. addInitScript runs
+  // ahead of every page script, and the vite dev plugin only fills the value in
+  // when it is still unset, so this one wins. plain assignment, NOT
+  // defineProperty: other injectors also assign to it, and a non-writable
+  // property would throw at that assignment and crash boot.
+  const bridgePortInject = bridgePort
+    ? 'window.__sootsimBridgePort = ' + Number(bridgePort) + ';'
+    : '';
+  await context.addInitScript({
+    content: 'Object.defineProperty(window, "__sootsimHostMeta", { value: ' + JSON.stringify(hostMeta) + ', configurable: true });' + bridgePortInject,
+  });
+  await context.exposeBinding('__sootsimHostClose', async () => {
+    await shutdown(0);
+  });
+  const connectTimeoutMs = Number(process.env.SOOTSIM_PW_CONNECT_TIMEOUT_MS || 120000);
+  if (connectAckFile) {
+    connectAckPoll = setInterval(() => {
+      if (ackFileExists()) markConnected();
+    }, 500);
+    if (typeof connectAckPoll.unref === 'function') connectAckPoll.unref();
+  }
+  connectTimer = setTimeout(() => {
+    if (ackFileExists()) {
+      markConnected();
+      return;
+    }
+    if (connected) return;
+    process.stderr.write('[connect-timeout] sim did not register with the bridge within ' + connectTimeoutMs + 'ms; closing playwright host\\n');
+    void shutdown(2);
+  }, Number.isFinite(connectTimeoutMs) && connectTimeoutMs > 0 ? connectTimeoutMs : 120000);
+  if (typeof connectTimer.unref === 'function') connectTimer.unref();
+  const page = await context.newPage();
+  if (hardwareGpuRequired) {
+    assertNoUnsafeSwiftShaderFlag();
+    const gpu = await page.evaluate(() => {
+      const canvas = document.createElement('canvas');
+      const gl =
+        canvas.getContext('webgl2', { powerPreference: 'high-performance' }) ||
+        canvas.getContext('webgl', { powerPreference: 'high-performance' });
+      if (!gl) return { renderer: null, vendor: null };
+      const info = gl.getExtension('WEBGL_debug_renderer_info');
+      return {
+        renderer: info ? String(gl.getParameter(info.UNMASKED_RENDERER_WEBGL)) : null,
+        vendor: info ? String(gl.getParameter(info.UNMASKED_VENDOR_WEBGL)) : null,
+      };
+    });
+    if (!gpu.renderer) {
+      throw new Error(
+        '[playwright-driver] could not prove a hardware WebGL renderer; refusing local render-heavy launch'
+      );
+    }
+    if (softwareRendererPattern.test(gpu.renderer)) {
+      throw new Error('[playwright-driver] refused software WebGL renderer: ' + gpu.renderer);
+    }
+    assertNoUnsafeSwiftShaderFlag();
+    process.stderr.write(
+      '[playwright-driver] hardware WebGL renderer: ' + gpu.renderer +
+      (gpu.vendor ? ' (' + gpu.vendor + ')' : '') + '\\n'
+    );
+  }
+  // surface page-side failures (engine boot crash, failed WS registration)
+  // to stderr so a sim that never connects produces a real diagnostic
+  // instead of a silent "timed out waiting for opened sim to connect".
+  page.on('pageerror', (e) => {
+    process.stderr.write('[pageerror] ' + ((e && e.stack) || e) + '\\n');
+  });
+  page.on('console', (m) => {
+    if (m.type() === 'error') {
+      process.stderr.write('[console.error] ' + m.text() + '\\n');
+    }
+  });
+  page.on('crash', () => process.stderr.write('[page] renderer crashed\\n'));
+  try {
+    await page.goto(url, { waitUntil: 'domcontentloaded', timeout: 60000 });
+  } catch (err) {
+    process.stderr.write('[goto] ' + String((err && err.message) || err) + '\\n');
+  }
+  keepAlive = setInterval(() => {}, 1 << 30);
+  // browser disconnect (e.g. user killed chrome from outside) → tear down.
+  // we are persistent-only now, so .browser() is the only handle to it.
+  try {
+    const b = context.browser && context.browser();
+    if (b && typeof b.on === 'function') b.on('disconnected', () => { void shutdown(0); });
+  } catch {}
+  context.on('close', () => { void shutdown(0); });
+  page.on('close', () => { void shutdown(0); });
+})().catch((err) => {
+  process.stderr.write(String((err && err.stack) || err) + '\\n');
+  process.exit(1);
+});
+`