REGRESSION(313609@main): visionOS debug uid 1 == main thread invariant

d1483c0

WTF's Thread and WorkQueue classes share one monotonically-increasing UID namespace. The main work queue is hard-coded with mainThreadID == 1; sequence-safety assertions across IDB, DOMCacheEngine, and NativePromise resolution depend on the invariant that the main thread's Thread::uid() is also 1. Before this fix, UID assignment was purely positional — whichever thread constructed its Thread object first won uid 1.

Source/WTF/wtf/Threading.h

-    explicit Thread(SchedulingPolicy schedulingPolicy)
+    enum class IsMain : uint8_t { No, Yes, Unknown };
+
+    explicit Thread(SchedulingPolicy schedulingPolicy, IsMain isMain = IsMain::Unknown)
         : m_isRealtime(schedulingPolicy == SchedulingPolicy::Realtime)
+        , m_uid(isMain == IsMain::Yes ? 1 : ++s_uid)
     {
     }

Source/WTF/wtf/posix/ThreadingPOSIX.cpp

+#if PLATFORM(COCOA)
+    Ref thread = adoptRef(*new Thread(SchedulingPolicy::Other, pthread_main_np() ? IsMain::Yes : IsMain::No));
+#else
     Ref thread = adoptRef(*new Thread(SchedulingPolicy::Other));
+#endif

Source/WTF/wtf/MainThread.cpp

+ RELEASE_ASSERT(Thread::currentSingleton().uid() == 1);

On visionOS debug, the RefCountDebugger path inside CStringBuffer's constructor called Thread::currentSingleton() from the libxpc bootstrap queue before initializeMainThread() ran — stealing uid 1 and pushing the real main thread to uid 2. The fix introduces an IsMain enum, uses pthread_main_np() on Cocoa to detect the main thread reliably, and adds a RELEASE_ASSERT enforcing the invariant.

Significance

The uid 1 == main thread invariant is load-bearing across WebKit's entire concurrency model; a stolen uid silently misfires RELEASE_ASSERTs in unrelated code paths, making crashes appear in the wrong place and masking real threading bugs.

Audit directions

Aaa Aaaaaaaaaaaaaaaaa Aaaa Aa Aaaaaaa Aaa Aaaaaaa Aaaaa Aaaaa Aaaaaa Aa Aaaaaaa Aaaaaaaa a Aaa Aaaaaa Aaaa Aaaa Aaaaaaaaaa a Aaaaaaaa Aa a Aaaaaaaa Aaaaaa Aaaaaa Aaaaaaaaaaaaaaaaaaaaaaaa Aaaa Aaaaaaaaaa Aaa a Aaa Aaa Aaaaaaaaaaaaaaaa Aaaaa Aa Aaaaaaaa Aaa Aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa Aaaaaaaa Aaaaaaaa a Aaaaaaaaaaaa Aaa Aaaaaaaaaa Aaaa Aaaaaaaaaaaaaaaaa Aa Aaaaaaaaa Aaaa Aaaaaaaaaaaaaaaaaa Aaaaaaaaaaaaaaaa Aaaaaaa Aaaa Aa Aaaa Aaa Aa Aaaaaa Aaa Aaa Aaaaaa Aaaaaaaaa a Aaa Aaaa Aaaaaaaaa Aaa Aaaaaaaa Aaaaaaaaaaa Aaaaaaaaaa Aaaaaa Aaa Aaaaaaaaaaaaaaaaaaaa Aaaaa Aaaaaaa Aaa Aaa Aaaaaaaaaaaaa Aaaaa Aaaaaaa Aaa Aaaaaaaaa Aaaaaaaa Aaaaaaaaa Aaaaa Aaa Aaaaaaaa Aaaaaaaaaa

🔒

The non-Cocoa fallback path and WebThread identity edge cases in the new dispatcher logic are worth investigating.

Subscribe to read more