Mac WK2 EWS very frequently fails, claiming media/track/add-and-remove-track.html to be failing. But it's actually flaky. Quite obviously, it's caused by some other test doing Web Notification work. It is very likely that the same issue happens on testers too, but we botwatched it away by marking affected tests in TestExpectations. But due to the nature of this failure, it can clearly happen on other tests randomly, and can migrate to other tests with time. Thread 0 Crashed:: Dispatch queue: com.apple.main-thread 0 com.apple.JavaScriptCore 0x000000010ffdbc1e WTFCrash + 62 (Assertions.cpp:329) 1 WebKitTestRunnerInjectedBundle 0x0000000113e60703 WTF::CrashOnOverflow::overflowed() + 9 2 WebKitTestRunnerInjectedBundle 0x0000000113e632e4 WTR::InjectedBundle::page() const + 24 (InjectedBundle.cpp:134) 3 WebKitTestRunnerInjectedBundle 0x0000000113e568a1 WTR::TestRunner::removeAllWebNotificationPermissions() + 27 (InjectedBundlePage.h:41) 4 WebKitTestRunnerInjectedBundle 0x0000000113e63054 WTR::InjectedBundle::didReceiveMessage(OpaqueWKString const*, void const*) + 398 (Vector.h:610) 5 com.apple.WebKit 0x000000010ef97618 WebKit::InjectedBundleClient::didReceiveMessage(WebKit::InjectedBundle*, WTF::String const&, API::Object*) + 120 (WKSharedAPICast.h:130) 6 com.apple.WebKit 0x000000010f0f58e5 WebKit::WebProcess::postInjectedBundleMessage(IPC::DataReference const&) + 135 (RefPtr.h:55) 7 com.apple.WebKit 0x000000010f0fefd0 WebKit::WebProcess::didReceiveWebProcessMessage(IPC::Connection*, IPC::MessageDecoder&) + 2356 (HandleMessage.h:16) 8 com.apple.WebKit 0x000000010f0f41e9 WebKit::WebProcess::didReceiveMessage(IPC::Connection*, IPC::MessageDecoder&) + 175 (WebProcess.cpp:612) 9 com.apple.WebKit 0x000000010ef52400 IPC::Connection::dispatchMessage(std::__1::unique_ptr<IPC::MessageDecoder, std::__1::default_delete<IPC::MessageDecoder> >) + 94 (memory:2665)
media/media-fragments/TC0001.html is another test that's randomly affected.
I wonder if this is related to bug 122475. Apparently m_pages can be empty sometimes. Perhaps a race when terminating a process and creating a new one?
> I wonder if this is related to bug 122475. Yes it is! I could reproduce, looks like that fix was not sufficient for single-WebProcess case (as used on Mountain Lion).
Created attachment 235832 [details] proposed fix
Committed <http://trac.webkit.org/r171880>.