Windows 64-bit opt xpcshell is currently permafailing since the uplift with harness timeouts like the log below: https://treeherder.mozilla.org/logviewer.html#?job_id=106399916&repo=mozilla-beta&lineNumber=6234 There's a common theme of lots of rmtree() failures and eventually a Python "thread.error: can't start new thread" error followed by the harness timing out and being force killed. Not sure what's at fault at this point, so filing under the XPCShell harness for now. I think this needs to block gtb for 55b1, though, since it's preventing us from getting any win64 xpcshell results. Wes did a set of uplift simulations early Sunday morning and this issue wasn't visible then, so there's at least a decent possibility that this was a recent regression. Below is a pushlog between the last set of simulations and what was eventually uplifted to Beta: https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=799d43edb324&tochange=f9605772a0c9 Bug 1371548 looks like a possibility, but I'm not sure how that code interacts with code run during xpcshell. It's at least worth a Try push, though. As Ehsan pointed out on IRC, the line below looks like something that could be broken on Win64: https://hg.mozilla.org/mozilla-central/rev/f7987c98de3e#l1.12
Try push of current Beta tip: https://treeherder.mozilla.org/#/jobs?repo=try&revision=854963773a811e828babdbb0946bb51598f23bfe Try push of current Beta tip and bug 1371548 backed out: https://treeherder.mozilla.org/#/jobs?repo=try&revision=737c09d206997f2467b23e3405c6578eb284a7a7
(In reply to Ryan VanderMeulen [:RyanVM] from comment #0) > Bug 1371548 looks like a possibility, but I'm not sure how that code > interacts with code run during xpcshell. It's at least worth a Try push, > though. As Ehsan pointed out on IRC, the line below looks like something > that could be broken on Win64: > https://hg.mozilla.org/mozilla-central/rev/f7987c98de3e#l1.12 Nope. The only way that code could possibly be relevant is if we were seeing the failure when running subprocess tests. But even if that code didn't work on win64 (which, incidentally, is the platform it was written for and tested on), we'd just fall back to the old error that change was intended to fix.
Turns out this was permafail from bug 1158387 that got misstarred and eventually merged around just before the uplift to Beta. FML.