Broken out from bug 782633. The failure there has now been made so that it won't turn the run orange, but we should still fix this. (In reply to Chris Jones [:cjones] [:warhammer] from comment #160) > The problem here is that we don't "join" all content processes on shutdown, > i.e. wait for them to terminate. And, not surprisingly, sometimes they > don't shut down before the parent process does. (In reply to Ted Mielczarek [:ted.mielczarek] from comment #161) > Can we do that for debug builds only? That's the only place we're doing leak > checks anyway. (In reply to Chris Jones [:cjones] [:warhammer] from comment #162) > Oh, golly, yes. I didn't mean to even suggest anything otherwise :). (In reply to Chris Jones [:cjones] [:warhammer] from comment #255) > Luckily the mechanism we need from this is coming, so hopefully in the near > future. (In reply to Ed Morley [:edmorley UTC+0] from comment #283) > Now that bug 821192 has landed, could I ask a massive favour and get you to > finish off the necessary work here? :-) (In reply to Chris Jones [:cjones] [:warhammer] from comment #290) > Bug 821192 ended up not providing the fix for this, unfortunately :/. Once we've fixed this bug and ensured that this warning isn't shown when we PROCESS-CRASH (per bug 782633 comment 385), we can s/WARNING/TEST-UNEXPECTED-FAIL/ again.

Here's the trivial test harness part of this.

This should also print out the pid of the process that is missing the leak log, as I don't know how you can possibly debug it otherwise.

This new version prints out the name of the log, which contains the pid, and ignores it when the GMP process fails to produce a log, because it never does.

This seems to actually just work, so I'm going to close this...

I landed some other code to require leak logs in another bug I'm not certain of. It is only enabled for some kinds of processes on some platforms, but it is easy to enable as they are fixed.