471647 - Some of the dom-level2-html tests trigger a 484 bytes leak, with Java SE 6 Update 1x NPAPI-based plugin

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect)

Description

[Serge Gautherie (:sgautherie)]

I wanted to narrow down bug 471505 (on mozilla-1.9.1),
"instead" I found this one (on mozilla-central).

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.2a1pre) Gecko/20081230 SeaMonkey/2.0a3pre] (home, optim default) (W2Ksp4)
(http://hg.mozilla.org/mozilla-central/rev/6201c0669e15
 +http://hg.mozilla.org/comm-central/rev/2854a16a867f)

{
TEST-UNEXPECTED-FAIL | runtests-leaks | leaked 484 bytes during test execution (threshold set at 8 bytes)
TEST-UNEXPECTED-FAIL | runtests-leaks | leaked 1 instance of nsComponentManagerImpl with size 276 bytes
TEST-UNEXPECTED-FAIL | runtests-leaks | leaked 2 instances of nsLocalFile with size 88 bytes each (176 bytes total)
TEST-UNEXPECTED-FAIL | runtests-leaks | leaked 3 instances of nsStringBuffer with size 8 bytes each (32 bytes total)
TEST-UNEXPECTED-FAIL | runtests-leaks | leaked 2 instances of nsTArray_base with size 4 bytes each (8 bytes total)
}

In tests\dom\tests\mochitest\dom-level2-html,
any of test_HTMLBodyElement07.html to test_HTMLBodyElement12.html triggers this;
and some/all of test_HTMLDocument01.html to test_HTMLDocument27.html do too.

Comment 1

[Bernd]

Does this leak also on ff or is this seamonkey only?

Comment 2

[Serge Gautherie (:sgautherie)]

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.2a1pre) Gecko/20090101 Minefield/3.2a1pre] (home, optim default) (W2Ksp4)
(http://hg.mozilla.org/mozilla-central/rev/e807ec425ad7)

Same leak.

Why didn't the Firefox tinderboxes catch this ? :-/

Comment 3

[Serge Gautherie (:sgautherie)]

(In reply to comment #2)
> Why didn't the Firefox tinderboxes catch this ? :-/

Odd: "WINNT 5.2 mozilla-central unit test" (and Linux and MacOSX) reports "no leaks detected!" before/after/since bug 427878 checkin.

My build is on Windows 2000 and has --disable-vista-sdk-requirements...

Comment 4

[Serge Gautherie (:sgautherie)]

Fwiw, I see this leak when running
|include ../../layout/base/crashtests/crashtests.list|
too.

Comment 5

[Serge Gautherie (:sgautherie)]

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b5pre) Gecko/20090424 SeaMonkey/2.0b1pre] (home, optim default) (W2Ksp4)
(http://hg.mozilla.org/releases/mozilla-1.9.1/rev/d77750c8a7b8
 +http://hg.mozilla.org/comm-central/rev/b7fd81f51c6e)

I reproduce with 1.9.1 too.

Comment 6

[Serge Gautherie (:sgautherie)]

http://tinderbox.mozilla.org/showlog.cgi?log=Firefox3.5/1244579133.1244583196.14718.gz
WINNT 5.2 mozilla-1.9.1 unit test on 2009/06/09 13:25:33
http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1244582652.1244588660.31973.gz
WINNT 5.2 mozilla-central unit test on 2009/06/09 14:24:12
crashtest + mochitest-plain

This started to happen on FF 3.5/3.6 box(es), with no checkins:
obviously related to the maintenance:
{
The following will be landed:
https://bugzilla.mozilla.org/show_bug.cgi?id=491298 and https://bugzilla.mozilla.org/show_bug.cgi?id=474572 - OPSI deployment on win32 slaves
https://bugzilla.mozilla.org/show_bug.cgi?id=495533 - Java update on win32 slaves
https://bugzilla.mozilla.org/show_bug.cgi?id=494971 - Add 10 new slaves to Talos Try
https://bugzilla.mozilla.org/show_bug.cgi?id=495610 - whitespace changes to some of our Buildbot code
}

Bug 495533 is the most probable candidate,
and I'm also using the new "Java(TM) Platform SE 6 U13" plugin atm!

Comment 7

[[e-mail] (:bhearsum)]

Yeah, the Java update definitely tripped this. The leak appears to be in XPCOM - cc'ing :bs and :dougt since they're authorities in that component. If either of you has a few minutes, can you see if anything sticks out to you?

If we can't fix the leak we need to back out the java update, or set a threshold.

Comment 8

[Doug Turner (:dougt)]

if you changed a npapi plugin, and we started leaking, i would start the blame in the new plugin.

Comment 9

[[e-mail] (:bhearsum)]

Is there anyway we can confirm that? Would a backtrace help?

Comment 10

[[e-mail] (:bhearsum)]

Doug made the excellent suggestion of removing the Java plugin and testing again. And indeed, without the java plugin we don't leak. I'm going to take the rest of this back to bug 495533, but it looks like we'll need a leak threshold to deal with the orange.

I'm working on a patch to do that so we can fix this perma-orange state.

Comment 11

[[e-mail] (:bhearsum)]

Josh volunteered to track down this leak.

Comment 12

[Nick Thomas [:nthomas] (UTC+12)]

The Java upgrade on the Firefox builders is also hitting a 484 byte leak running crashtest. eg http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1244617102.1244625811.23152.gz - looks the same as comment #0.

Comment 13

[Serge Gautherie (:sgautherie)]

(In reply to comment #12)
> - looks the same as comment #0.

Reported in comment 4 and comment 6 indeed!

Comment 14

[Mike Beltzner [:beltzner, not reading bugmail]]

This only leaks when you instantiate the plugin, right? So when you actually run some DOM manipluation code that requires Java?

Don't think we'll block Firefox 3.5 on a 400 byte leak that only happens under those circumstances, please feel free to renominate if I'm misinterpreting.

Comment 15

[Serge Gautherie (:sgautherie)]

(In reply to comment #14)

(Fwiw, I think I have a real life JavaWebStart case (on a private "bank" site) where I see a continuous leak increase of SeaMonkey process "and" a shutdown hang of the external Java process (only).
But I haven't tried to confirm/pinpoint these yet.
Mentioning it here, just in case someone else would run into something like that.)

Comment 16

[Serge Gautherie (:sgautherie)]

Jesse, why abuse bug 438871? This is not intermittent.

Comment 17

[Jesse Ruderman]

Sorry, I saw orange attributed to an old bug and assumed it was intermittent.  I guess it's actually tied to the version of Java on each machine?

Comment 18

[[e-mail] (:bhearsum)]

(In reply to comment #17)
> Sorry, I saw orange attributed to an old bug and assumed it was intermittent. 
> I guess it's actually tied to the version of Java on each machine?

For the record, all of the slaves running m-c, 1.9.1, and tracemonkey builds have Java 6 on them. Only the try slaves haven't been updated yet.

Comment 19

[Serge Gautherie (:sgautherie)]

1.9.2 trunk:


http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1246020805.1246027759.11985.gz&fulltext=1
WINNT 5.2 mozilla-central unit test on 2009/06/26 05:53:25

Still leaks.


http://tinderbox.mozilla.org/showlog.cgi?log=Firefox/1246028465.1246036567.16139.gz&fulltext=1
WINNT 5.2 mozilla-central unit test on 2009/06/26 08:01:05

Doesn't leak anymore.

Fix timeframe:
http://hg.mozilla.org/mozilla-central/pushloghtml?startdate=2009-06-26+05%3A50%3A24&enddate=2009-06-26+07%3A58%3A06
A few changesets, related to (Java) Plugin!


[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.2a1pre) Gecko/20090711 SeaMonkey/2.1a1pre] (home, optim default) (W2Ksp4)
(http://hg.mozilla.org/mozilla-central/rev/47bfcd275ede
 +http://hg.mozilla.org/comm-central/rev/291cbe3374b9)

No more leak either.


*****

1.9.1 branch:


http://tinderbox.mozilla.org/showlog.cgi?log=Firefox3.5/1247342838.1247348243.6059.gz&fulltext=1
WINNT 5.2 mozilla-1.9.1 unit test on 2009/07/11 13:07:18

Still leaks.


http://tinderbox.mozilla.org/showlog.cgi?log=SeaMonkey2.0/1247314133.1247321343.27351.gz&fulltext=1
WINNT 5.2 comm-1.9.1 unit test on 2009/07/11 05:08:53

Does not leak.
KaiRo, which Java plugin do the Windows SeaMonkey boxes use? (Were they upgraded to v6r14?)

Comment 20

[Robert Kaiser]

no upgrade to v6r14 was done, we still have build 1.5.0_10-b03 right now.

Comment 21

[Serge Gautherie (:sgautherie)]

(In reply to comment #20)
> no upgrade to v6r14 was done

Now done in bug 503507 and leaking (on m-1.9.1).