759000 - Mac Peptest jobs intermittent hit an infinite loop of WARNING - PEP WARNING | MOZMILL fail {"message":"[JavaScript Error: \"TypeError: can't access dead object\" {file: \"resource://mozmill/stdlib/utils.js\" line: 102}]"}

Status: RESOLVED FIXED

(Testing Graveyard :: Peptest, defect)

Description

[Phil Ringnalda (:philor)]

See bug 758988 for the releng part of "something really should stop these eventually," but according to that, the Mac Peptest jobs that I'm seeing running for 12 or 24 or 48 hours really are running that long, looping on 


06:15:44     INFO -  PEP TEST-END   | test_openWindow.js | finished in: 234 ms
06:15:44  WARNING -  PEP WARNING    | MOZMILL fail {"message":"[JavaScript Error: \"TypeError: can't access dead object\" {file: \"resource://mozmill/stdlib/utils.js\" line: 102}]"}
06:15:44  WARNING -  PEP WARNING    | MOZMILL fail {"message":"[JavaScript Error: \"TypeError: can't access dead object\" {file: \"resource://mozmill/stdlib/utils.js\" line: 102}]"}

Comment 1

[Mark Côté [:mcote]]

I looked at the logs for 4 of these failures, and it looks it's always after the openwindow test that peptest (mozmill specifically) goes crazy. There are some other OS X problems too; I will investigate.

Comment 2

[cmtalbert]

Henrik, I think this is related to the changes that you want to make in the waitfor code that we discussed last week. Specifically, Mark, this will likely be fixed by the patch that is coming in bug https://bugzilla.mozilla.org/show_bug.cgi?id=754225

Comment 3

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Mark, where is this test located which is failing here? I would like to have a look first before making any assumptions.

Comment 4

[Mark Côté [:mcote]]

The test is part of peptest (mozilla-central/testing/peptest); specifically, it is mozilla-central/testing/peptest/tests/firefox/test_openWindow.js that seems to be triggering the problem.

test_resizeWindow.js is not working properly in OS X either, but I don't yet know if it's related or even an issue with mozmill. Sometimes it errors out with "TypeError: window is null"; other times it appears to work but gives "ReferenceError: test is not defined". These errors aren't present on other platforms (though Windows appears to be totally broken right now).

Comment 5

[Henrik Skupin [:whimboo][⌚️UTC+1]]

Just for reference the full URL is:
http://mxr.mozilla.org/mozilla-central/source/testing/peptest/tests/firefox/test_openWindow.js

So given that test and the lines of code in comment 0 shows me that we fail here:
http://mxr.mozilla.org/mozilla-central/source/testing/peptest/peptest/extension/resource/mozmill/stdlib/utils.js#94

Can someone point me to a full log? It would help me a lot with further investigation. But Clint seems to be right that my upcoming patch on bug 754225 could help here too.

Comment 6

[Mark Côté [:mcote]]

Here's a full log:

https://tbpl.mozilla.org/php/getParsedLog.php?id=12149179&tree=Mozilla-Inbound

Looks like this one was caught fairly early, but others have hundreds, if not thousands, of the same error message.

I did some more investigation, and there is a bug in a test_openWindow.js on OS X in which the main window is accidentally closed. So this is probably triggering the Mozmill bug.

Comment 7

[Henrik Skupin [:whimboo][⌚️UTC+1]]

So the problem seems to be this line:

 win.keypress("n", {"ctrlKey":true});

Ctrl doesn't have any effect on OS X. You should really change that to accelKey which will trigger a Cmd+N keypress. Otherwise as you have already stated the one and only open window will be closed by the last line of the test.

Comment 8

[Mark Côté [:mcote]]

Attachment: Switch to accelKey instead of ctrlKey in peptest tests

While the Mozmill errors themselves are probably a different bug, this should avoid triggering them. It will also fix other errors in peptests on OS X, since ctrlKey isn't support.

Comment 9

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/eb6f7f0f77d5