Created attachment 297035 [details] Crash log Crash in com.apple.WebKit:IPC::Connection::platformCanSendOutgoingMessages() const + 0 Thread 1 Crashed:: Dispatch queue: com.apple.IPC.ReceiveQueue 0 com.apple.WebKit 0x0000000108ebc32c IPC::Connection::platformCanSendOutgoingMessages() const + 0 (ConnectionMac.mm:290) 1 ??? 0x0000000100000000 0 + 4294967296 2 com.apple.WebKit 0x0000000108ebc4c7 IPC::Connection::sendOutgoingMessage(std::__1::unique_ptr<IPC::Encoder, std::__1::default_delete<IPC::Encoder> >) + 393 (ConnectionMac.mm:379) 3 com.apple.WebKit 0x0000000108eb8340 IPC::Connection::sendOutgoingMessages() + 82 (Connection.cpp:825) 4 libdispatch.dylib 0x00007fff8770293d _dispatch_call_block_and_release + 12 5 libdispatch.dylib 0x00007fff876f740b _dispatch_client_callout + 8 6 libdispatch.dylib 0x00007fff876fc03b _dispatch_queue_drain + 754 7 libdispatch.dylib 0x00007fff87702707 _dispatch_queue_invoke + 549 8 libdispatch.dylib 0x00007fff876fad53 _dispatch_root_queue_drain + 538 9 libdispatch.dylib 0x00007fff876fab00 _dispatch_worker_thread3 + 91 10 libsystem_pthread.dylib 0x00007fff8c31f4de _pthread_wqthread + 1129 11 libsystem_pthread.dylib 0x00007fff8c31d341 start_wqthread + 13 https://build.webkit.org/results/Apple%20El%20Capitan%20Release%20WK2%20(Tests)/r209759%20(11768)/results.html
<rdar://problem/29626731>
Application Specific Information: Unhandled error code 1000000a, message 'WebInspectorProxy:CreateInspectorPage:' CRASHING TEST: /xmlhttprequest/access-control-repeated-failed-preflight-crash.html https://build.webkit.org/results/Apple%20El%20Capitan%20Release%20WK2%20(Tests)/r209881%20(11850)/http/tests/xmlhttprequest/access-control-repeated-failed-preflight-crash-crash-log.txt
Created attachment 297474 [details] Patch
Committed r209995: <http://trac.webkit.org/changeset/209995>
Comment on attachment 297474 [details] Patch Any way to cover this with a regression test?
(In reply to comment #5) > Comment on attachment 297474 [details] > Patch > > Any way to cover this with a regression test? Not easily - it was timing dependent (and relied on rescheduling the message because the mach queue was full).
Existing regression tests were hitting this somewhat frequently (which is how we learned about the issue).