In  Issue 605349  a flaky failure in V8's semaphore implementation was fixed by picking up a new version of Clang. I don't fully understand the failure mode -- I think it was that a memory address against which atomic operations were performed was 4-byte rather than 8-byte aligned -- but to prevent this sort of failure from happening again a targeted test is urgently needed. hpayer@ and ulan@ can provide concrete details of the failure.

Is anyone available to reduce this to a test case that can be run at least during Clang rolls?