Generating a crash log takes ~2 minutes on my machine, which is longer than what run-webkit-tests waits before deciding that the tool crashed. Looking at bots, crashes are mis-reported as timeouts too. See also: <rdar://problem/18701447>.
Created attachment 243727 [details] proposed fix
Comment on attachment 243727 [details] proposed fix View in context: https://bugs.webkit.org/attachment.cgi?id=243727&action=review > Tools/DumpRenderTree/mac/DumpRenderTree.mm:1141 > NSString *identifier = [[NSBundle mainBundle] bundleIdentifier]; > const char *stdinPath = [[NSString stringWithFormat:@"/tmp/%@_IN", identifier] UTF8String]; > const char *stdoutPath = [[NSString stringWithFormat:@"/tmp/%@_OUT", identifier] UTF8String]; This code (lines 1139 thru 1149 in the proposed patch) is specific to the iOS LayoutTestRelay infrastructure (*) and hence should be guarded with PLATFORM(IOS). That is, we should only move the SIGNAL(3) calls outside the PLATFORM(IOS)-guard (lines 1151 thru 1154 in the proposed patch). (*) On iOS we indirectly launch DumpRenderTree.app/WebKitTestRunnerApp.app using a helper app, LayoutTestRelay. We use FIFOs (see MKFIFO(1)) as the communication medium between LayoutTestRelay and DumpRenderTree.app/WebKitTestRunnerApp.app instead of using pipes because of limitations with respect to launching a process in the iOS Simulator.
Ah, so that's what this sorcery is about. Thanks! Committed <http://trac.webkit.org/r177722>.