986458 - browser_whitelist7.js leaked 1 docShell(s) until shutdown on all platforms in debug mode when run as a standalone directory

Status: RESOLVED FIXED

(Firefox for Android Graveyard :: Add-on Manager, defect)

Description

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

I am working to chunk the browser-chrome tests better, this requires us to run directory by directory instead of a big chunk of tests.  one side effect is that many tests fail and this is one of the lucky tests.

here is a try server run showing the failures:
https://tbpl.mozilla.org/?tree=Try&rev=29bcc82f7302

all platforms on debug we leak.

Comment 1

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

I am going to try removing this line:
http://dxr.mozilla.org/mozilla-central/source/toolkit/mozapps/extensions/test/xpinstall/browser_whitelist7.js#25

as this would make it identical to whitelist4.js.

Comment 2

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

here is the try server push:
https://tbpl.mozilla.org/?tree=Try&rev=99b14959adca

Comment 3

[Eugen Sawin [:esawin]]

Attachment: Add blocked add-on host default

The leaking test is testing an Android feature on the desktop path, since we are only concerned with the shared blocking mechanics. The desktop front-end, however, does not handle blocked direct requests well, so we need to default to |localhost| when no host is available there.

Lets test this patch. Should it fix the issue, we need to decide whether this is something we want to handle with an appropriate warning prompt on desktop, like we do it on Android.

Comment 4

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

I tested this patch and I still see it leaking:
https://tbpl.mozilla.org/?tree=Try&rev=99b14959adca

So quite possibly this is a leak in other tests in that directory that is getting blamed on whitelist7

Comment 5

[Ryan VanderMeulen [:RyanVM]]

We're seeing this on Cedar as well and it's one of the few remaining issues that's blocking us getting chunking enabled across all platforms. Any other ideas, Eugen?

https://tbpl.mozilla.org/php/getParsedLog.php?id=37281982&tree=Cedar

Comment 6

[Eugen Sawin [:esawin]]

The Try server push from comment 4 is the unpatched version, can we test this patch on the failing systems (Win 8 debug, OS X debug) to see if it helps? Here are the tests on the patched Cedar branch, but Try doesn't run the bc3 chunks on the failing systems: https://tbpl.mozilla.org/?tree=Try&rev=71acd51217d2.

Comment 7

[Eugen Sawin [:esawin]]

Just realized that the bc tests aren't chunked on those systems yet, queuing some more bc runs on the patched version: https://tbpl.mozilla.org/?tree=Try&rev=71acd51217d2.

Comment 8

[Ryan VanderMeulen [:RyanVM]]

Pushed to Cedar.
https://tbpl.mozilla.org/?tree=Cedar&jobname=mochitest.*chrome&rev=860811be1a5d

Comment 9

[Ryan VanderMeulen [:RyanVM]]

Looks like it's still leaking:
https://tbpl.mozilla.org/php/getParsedLog.php?id=37320967&tree=Cedar

Comment 10

[Ryan VanderMeulen [:RyanVM]]

FWIW, disabling browser_whitelist7.js does make the bc3 leaks go away, but we still do see other intermittent failures in the xpinstall tests like this:
https://tbpl.mozilla.org/php/getParsedLog.php?id=37334205&tree=Cedar

Comment 11

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

any other thoughts here?  Should we keep this disabled for now until?  We can keep trying patches on cedar.

Our goal is to have everything ready for the firefox31 branches by Friday.

Comment 12

[Eugen Sawin [:esawin]]

Let's keep this test disabled for now. It should be more convenient to test patches for it with chunking enabled across all platforms.

Comment 13

[Ryan VanderMeulen [:RyanVM]]

Disabled on Windows and OSX debug.
https://hg.mozilla.org/integration/fx-team/rev/6c260ce33c2a

Comment 14

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/6c260ce33c2a

Comment 15

[Eugen Sawin [:esawin]]

Does not leak when enabled on m-c: https://tbpl.mozilla.org/?tree=Try&rev=3dc2f2560d1b.
Should we enable the test but keep this bug open for now?

Comment 16

[Ryan VanderMeulen [:RyanVM]]

Yeah, that's worrying to me because chunk-by-dir was supposed to give us more consistent results.

Comment 17

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

chunk-by-dir will reduce the problems when we add/disable a test as they won't switch chunks- that same effect can still cause tests in the same chunk to have problems.

Comment 18

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/10e08248da4f

Comment 19

[Ryan VanderMeulen [:RyanVM]]

This appears to have gone away on its own. Re-enabling the test across the board.
https://hg.mozilla.org/integration/mozilla-inbound/rev/10e08248da4f

Comment 20

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/10e08248da4f