We should add assertions to state transitions for GCParallelTask to check the original state is as expected and that we hold the helper thread lock when we do this.

Here's a patch to improve the assertions around GCParallelTask state management. I removed the argument to the cancel() method which was always passed GCParallelTask::CancelAndWait. I found one actual bug - the cancel field wasn't being initialised in the first constructor. This is only used in background alloc and decommit tasks though, which don't have to run to completion so it shouldn't have caused any actual problems.

Comment on attachment 8972852 [details] [diff] [review] bug1458839-parallel-task-assertions Review of attachment 8972852 [details] [diff] [review]: ----------------------------------------------------------------- Nice.

Pushed by jcoppeard@mozilla.com: https://hg.mozilla.org/integration/mozilla-inbound/rev/a35392007e5d Improve state assertions in GCParallelTask r=jandem