Many apptests appear to be flaky for the same reason: shutdown races.
In particular all the tests appear to run to completion just fine, but child processes may kill themselves prematurely and this can be interpreted as a failure.
WindowServerTest.ActivationNext in particular is failing much more often than the others since r379508 landed, enough that the bot often goes red for a cycle or two at a time.
I've done enough investigation to be convinced that the bindings patch isn't breaking this but only exposing symptoms of the problem for frequently -- maybe it's causing the shell connection to be terminated earlier with respect to other pipes and that leads to more frequent shutdown flake. The test always runs to completion, it just kills itself before signalling success.
Rather than needlessly revert the bindings patch again, I'm disabling this test to make the bot green while I investigate a proper solution to the flakiness.
Comment 1 by bugdroid1@chromium.org
, Mar 7 2016