867745 - leaking docshell mochitest-chrome test causes many other tests to fail, pass, etc...

Status: RESOLVED WORKSFORME

(Core :: DOM: Navigation, defect)

Description

[Joel Maher ( :jmaher ) (UTC -8)]

While debugging why dom/tests/mochitest/chrome/test_fullscreen.xul fails on ubuntu VM's, I found that I couldn't reproduce the problem by running just the dom subdirectory of the mochitest-chrome tests.  

Upon further investigation I had removed the docshell/* tests from the chrome folder and my tests pass.  What is more disturbing is new tests fail:
6333 ERROR TEST-UNEXPECTED-FAIL | chrome://mochitests/content/chrome/dom/tests/mochitest/localstorage/test_localStorageQuotaPrivateBrowsing_perwindowpb.html | Test timed out.
8715 ERROR TEST-UNEXPECTED-FAIL | chrome://mochitests/content/chrome/layout/forms/test/test_bug536567_perwindowpb.html | uncaught exception - TypeError: can't access dead object at chrome://mochitests/content/chrome/layout/forms/test/test_bug536567_perwindowpb.html:120
12689 ERROR TEST-UNEXPECTED-FAIL | chrome://mochitests/content/chrome/toolkit/components/passwordmgr/test/test_privbrowsing_perwindowpb.html | Checking submitted password - got , expected notifyp1


So something in docshell is causing 3 private browsing tests to pass and causing test_fullscreen.xul to fail.

Comment 1

[Joel Maher ( :jmaher ) (UTC -8)]

seems to be in test_bug[1-3]*.

Comment 2

[Joel Maher ( :jmaher ) (UTC -8)]

appears to be in test_bug36* and test_bug39*.  There are 4 files and if all are missing I pass the fullscreen test consistently.  

As for the private browsing errors, those tests are not related to it.

Comment 3

[Joel Maher ( :jmaher ) (UTC -8)]

test_bug364461.xul is the culprit for the fullscreen.xul errors, could it be we don't call finish here?  I would assume if we messed up the call to SimpleTest.finish() that it would affect one of the dozen or so other test cases between it and fullscreen.xul:
http://mxr.mozilla.org/mozilla-central/source/docshell/test/chrome/bug364461_window.xul#74

Comment 4

[(not currently active) Ted Mielczarek]

This test looks like it could use the new SimpleTest.registerCleanupFunction from bug 866641!

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

yes, but what would we clean up?

Comment 6

[(not currently active) Ted Mielczarek]

The stuff that finish is doing?

Comment 7

[Joel Maher ( :jmaher ) (UTC -8)]

oh yeah...lets make sure when this test case is fixed that it get retrofitted for some new goodness.  No point in fixing this if the test case will be disabled soon (i.e. the original developers cannot fix this).

Comment 8

[Joel Maher ( :jmaher ) (UTC -8)]

ok, it also looks test_bug301397.xul (like http://mxr.mozilla.org/mozilla-central/source/docshell/test/chrome/bug301397_window.xul) is also causing this to display as well.

I am not sure what to do here.  

removing only:
test_bug364461.xul

or removing only:
test_bug301397.xul

allows for test_fullscreen.xul to work.  What is even more strange is that we can remove just one of them and the other doesn't cause problems.  Removing both allows us to work just fine.  

With all the problems these docshell tests seem to be causing I am really tempted to turm them all off.  I can't do that because all kinds of other chrome tests (i.e. private browsing related) will time out.

speaking of private browsing, removing:
test_bug303267.xul 

seems to make all the private browser tests fail (by timeout).  So far 3 out of 29 of these tests are causing problems in other tests.

Comment 9

[Joel Maher ( :jmaher ) (UTC -8)]

cc'ing folks who have all written tests mentioned in this bug to date, if you are an original author, please look at this bug and help figure out what is going on.

Comment 10

[Joel Maher ( :jmaher ) (UTC -8)]

jgriffin, if you are going to look at this on your loaner ubuntu slave, remember to remove the mochitest/tests.jar file so you can use the tests from the mochitest/chrome/ folder.

Comment 11

[Joel Maher ( :jmaher ) (UTC -8)]

Attachment: disable one of the known leaking tests for linux (1.0)

we can disable one test for now since the original test authors haven't come up with a fix for the last week.  Once this is landed and we are running mochitest-other on ubuntu, this will be easy to test on try server.

Comment 12

[Armen [:armenzg]]

Comment on attachment 747467 [details] [diff] [review]
disable one of the known leaking tests for linux (1.0)

Review of attachment 747467 [details] [diff] [review]:
-----------------------------------------------------------------

Not that I know much about these type of patches but you don't seem to do anything scary! :P

Comment 13

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/351f669dce96

Comment 14

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/351f669dce96

Comment 16

[Joel Maher ( :jmaher ) (UTC -8)]

in debug mode these 3 tests cause test_fullscreen.xul to fail:
browser/components/feeds/test/chrome/test_423060.xul
docshell/test/chrome/test_bug301397.xul
dom/tests/mochitest/chrome/test_callback_wrapping.xul

Comment 17

[Joel Maher ( :jmaher ) (UTC -8)]

Attachment: disable fullscreen.xul for linux (1.0)

re-enable the docshell test and disable the test_fullscreen.xul.  After finding 3 other tests which cause interference, I am going to disable the one test which is in our way.

Comment 18

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/a9ce98271a7c

Comment 19

[Andrew McCreight [:mccr8]]

This test eventually became test_fullscreen.xhtml. It was accidentally re-enabled in bug 1798274, and I don't see any intermittent failures on file for it. I filed bug 1818000 for re-enabling the other test that was accidentally disabled at the same time, and deleting the annotation here.