377899 - Cycle between nsXULTooltipListener and nsGlobalWindows causes leaks

Status: RESOLVED FIXED

(Core :: XUL, defect)

Description

[Peter Van der Beken [:peterv]]

Attachment: Testcase

See bug 376802, comment 11 and others. In the attached testcase, hover over the button and the browser will quit, leaking nsGlobalWindows (the first time you'll need to allow UniversalXPConnect privileges). It shows one way to keep a nsGlobalWindow alive, I'll also attach part of an ownership graph.

Comment 1

[Peter Van der Beken [:peterv]]

Attachment: Onwership graph

The missing edge to nsXULTooltipListener comes from nsLayoutStatics (which keeps it because there are live nsGlobalWindows).

Comment 2

[[e-mail]]

The tooltiplistener keeps a strong reference to the tooltip as long as the tooltip is shown, so waiting for a popuphiding event that never comes. I assume before bug 376802 the tooltip listener would have been destroyed when the source element was unbound from the tree, thus releasing the reference to the tooltip?

Comment 3

[Olli Pettay [:smaug][[e-mail]]]

Attachment: using cycle collector, v1

Comment 4

[Peter Van der Beken [:peterv]]

Comment on attachment 262751 [details] [diff] [review]
using cycle collector, v1

That should work, yes.

Comment 5

[[e-mail]]

Comment on attachment 262751 [details] [diff] [review]
using cycle collector, v1

>-
>-  // nsISupports
>-  NS_DECL_ISUPPORTS
>+  NS_DECL_CYCLE_COLLECTING_ISUPPORTS
>+  NS_DECL_CYCLE_COLLECTION_CLASS_AMBIGUOUS(nsXULTooltipListener,
>+                                           nsIDOMMouseListener)
> 
>   // nsIDOMMouseListener
Nit: any reason why you removed the nsISupports comment?

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

Because NS_DECL_CYCLE_COLLECTING_*** does something else too, not only
declares nsISupports.