ManagedBrowser SEGV_ACCERR crash with --disable-features=OptimizationHints in --headless=new (macOS arm64)

[wwtearl]

Bug

ManagedBrowser's hardcoded --disable-features=OptimizationHints,MediaRouter,DialMediaRouteProvider flag crashes Chrome-for-Testing with a SEGV_ACCERR segfault immediately after the CDP endpoint comes up, when launched with --headless=new. crawl4ai reports this generically as "CDP endpoint not ready after startup", which gives no indication a crash (rather than a slow startup) occurred.

Environment

• crawl4ai version: 0.9.3
• OS: macOS 26.6.2 (Darwin, arm64 / Apple Silicon)
• Browser: Chrome-for-Testing 151.0.7922.34
• Launch config: BrowserConfig(headless=True, use_managed_browser=True, user_data_dir=<persistent profile>) — i.e. the ManagedBrowser code path (raw subprocess + CDP connect), not the Playwright-managed BrowserManager path.

Steps to reproduce

1. Launch Chrome-for-Testing directly with the exact flag list ManagedBrowser.build_browser_flags constructs, including --headless=new and --disable-features=OptimizationHints,MediaRouter,DialMediaRouteProvider.
2. Redirect stdout/stderr to a log file and poll http://localhost:<port>/json/version for CDP readiness.
3. Observe: the log shows DevTools listening on ws://... (CDP genuinely came up), immediately followed by Received signal 11 SEGV_ACCERR and a crash stack trace. The process then dies, so the 5-attempt exponential-backoff readiness poll in _verify_cdp_ready() eventually times out and crawl4ai surfaces only "CDP endpoint not ready after startup".

Isolating the cause

I bisected the ~20-flag launch argument list by binary search (removing half the flags, reproducing/not-reproducing, repeating) using direct manual Chrome launches outside of crawl4ai. This isolated the crash to a single flag:

• --disable-features=OptimizationHints,MediaRouter,DialMediaRouteProvider → crashes
• --disable-features=MediaRouter,DialMediaRouteProvider (i.e. OptimizationHints removed) → does not crash
• MediaRouter and DialMediaRouteProvider are safe both individually and combined; OptimizationHints alone in the disable-list reproduces the crash by itself.

This was not a stale-lock-file or --disable-gpu-related issue — both of those were ruled out first (removed Singleton{Lock,Cookie,Socket}, confirmed no stale process held them via ps -p <pid>; removed --disable-gpu and still reproduced the crash). The only thing that changed the outcome was the presence/absence of OptimizationHints in --disable-features.

Fix applied locally

Removed OptimizationHints from both --disable-features flag lists in crawl4ai/browser_manager.py (ManagedBrowser.build_browser_flags, and a sibling list further down the same file), leaving MediaRouter,DialMediaRouteProvider. This fully resolves the crash on this machine with no other observed regressions.

Possibly related but distinct

#1445 ("Identity based browsing fails") looks superficially similar (CDP/managed-browser startup failure) but its root cause there was diagnosed as a CDP-readiness-check timing race (asyncio.sleep waits too short before Chrome fully initializes), not a crash — the browser process in that report doesn't appear to actually die. This issue is a genuine SEGV crash triggered by one specific Chromium feature-disable flag, reproducible outside of crawl4ai entirely via a direct Chrome launch, so I believe it's a separate root cause worth tracking on its own even though the surface symptom (CDP never becomes ready) looks similar.

Suggested fix

Drop OptimizationHints from the --disable-features list in ManagedBrowser.build_browser_flags (and the sibling list), or make the flag list configurable so users hitting Chromium-version/platform-specific crashes on a particular feature flag can override it without patching the package.

[ntohidi]

@wwtearl thanks fo your contribution. ur fix is in the develop branchnow. It would be great if you could check the fix