71623 - leaks caused by ownership cycle through event listeners and JS

Status: RESOLVED WORKSFORME

(Core :: DOM: UI Events & Focus Handling, defect)

Description

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

DESCRIPTION:  Without the fix for bug 31847, I'm seeing leaks loading
http://www.people.fas.harvard.edu/~dbaron/dom/test/two-events/EventTarget
and pressing the button.

The cause of the leaks seems to be (I think!):
 * nsJSDOMEventListener objects own an nsJSContext (mContext)
 * the nsJSContext has a rooted script object (why?)
 * this root prevents the document from being GCed
 * the document owns the nsEventListenerManager
 * the nsEventListenerManager owns the nsJSDOMEventListener (cycle)

The fix to bug 31847 obviously entrains a lot more stuff in this leak since that
fix essentially adds ownership where we needed ownership to prevent a crash.

Comment 1

[joki (gone)]

So when are you seeing the leaks?  Of the 7 listeners on the page I can see the 
3 that get removed getting destroyed.  If I switch pages a couple times I can 
see the other 4 get removed when the GC eventually collects the script objects.  
Running WinNT with a few local changes though I don't think offhand they'd 
affect leakage.

On a side note on the test case, the capturers will start working with the fix 
in my tree now.  Not sure if you just haven't gotten around to updating stuff 
but 'currentNode' was changed to 'currentTarget'.  Bubbles and cancelable work 
also, at least locally for me.

Comment 2

[clayton]

David, could you confirm that this still happens?  I wonder if GC just hadn't 
yet fired.  Also wonder if Tom's fix has solved it.

Comment 3

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

This does still happen.  joki's right that it isn't really a problem though,
except that it makes leak finding harder.  That jscontext_savedroot gets
unrooted if I go to a different page before exiting.

Comment 4

[clayton]

Futuring since we can live with this one and Joki is swamped.

Comment 5

[Madhur Bhatia]

QA contact updated

Comment 6

[Pham]

.

Comment 7

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Maybe duplicate of bug 241518 if the problem still exists.

Comment 8

[Robert Sayre]

We don't leak on this test anymore, marking WFM.