This is maybe expected, but when I experimented with making the WebContent process run [NSRunLoop run] instead of [NSApplication run], the timer latency increased because the priority of the WebContent process became lower. This increased the layout test flakiness. Ideally, the layout test results should not depend too much on timer latency. Perhaps we should drain the message queue between each test?
This sounds like a great way to make flakiness more reproducible! I’m not sure if doing something between the tests is the answer though. We already do many things intended to ensure that there is a strong boundary between tests.