780674 - JSM reference to DOMWindow object leaks DOMWindow

Status: RESOLVED INVALID

(Core :: General, defect)

Description

[Myk Melez [:myk] [@mykmelez]]

Attachment: mochitest that demonstrates leak

Assigning an object created in a DOMWindow to a property of an object created in a JavaScript module causes the DOMWindow to leak.

The attached simple chrome mochitest demonstrates this by setting a property of XPCOMUtils to an object literal:

  Components.utils.import("resource://gre/modules/XPCOMUtils.jsm");
  XPCOMUtils.foo = {};

That shouldn't cause the DOMWindow associated with the mochitest to leak (the window itself holds no obvious reference to the object literal created in it).  But it does leak when I run the test via:

  make mochitest-chrome TEST_PATH=dom/tests/mochitest/chrome/test_object_reference_leak.xul


(Filing in General, as I'm unsure where specifically this should go.  Perhaps DOM or XPConnect?  The test itself is only in dom/ because it was convenient to put it there.  And the cc: list is just a pseudo-random collection of folks who might know something about this or be able to cc: the right person.)

Comment 1

[Boris Zbarsky [:bzbarsky]]

> (the window itself holds no obvious reference to the object literal created in it)

It's the other way around.  The object literal has a reference to its global, as all JS objects do.  This has nothing to do with the DOM; it's a pure Spidermonkey thing, and more or less required by the spec (in fact, upcoming things in ES5 are likely to allow retrieving the global from an object).

Comment 2

[Myk Melez [:myk] [@mykmelez]]

(In reply to Boris Zbarsky (:bz) [In and out Aug 1 - 10, out Aug 11-20] from comment #1)
> > (the window itself holds no obvious reference to the object literal created in it)
> 
> It's the other way around.  The object literal has a reference to its
> global, as all JS objects do.

Erm, indeed!

> This has nothing to do with the DOM; it's a
> pure Spidermonkey thing, and more or less required by the spec (in fact,
> upcoming things in ES5 are likely to allow retrieving the global from an
> object).

Hmm, ok.  That seems unfortunate, but I suppose them's our breaks.