Forums

I filed a ticket a while ago (http://yuilibrary.com/projects/yui3/ticket/2532270), but since I haven't received any response, I think it's time to try to involve the larger community.

The problem is that Y.Node has a static member called _instances which stores references to all instances of Y.Node. The only way to remove an instance from _instances is to explicitly call destroy().

This forces everybody back to the programming model of IE6 and IE7.

Y.Node._instances provides significant value, however, since it ensures that only one Y.Node instance is created for each DOM element -- within each sandbox, of course.

I therefore propose an alternative to the global _instances map: On each DOM element, store a mapping from sandbox id's to Y.Node instances.

Unfortunately, this does mean one sandbox can in principle access Y.Node instances from another sandbox -- though it does require extra effort. But sandboxes have never been completely isolated, since DOM elements are global and Y.Node itself avoids instanceof in order to allow Y.Node instances to be passed between sandboxes.

My patch is here:

https://github.com/jafl/yui3/commit/498 ... cf3ce539c1

I would appreciate feedback on whether this is an acceptable replacement for Y.Node._instances.

I take back what I said. Good WeakMap shims are reliant on ES5 features (setting enumerability to false).

Well, YUI still needs to support those browsers, so I don't think it belongs in the core library. Is it monkey-patchable?

Sorry about not being responsive to this topic. I did some testing last year on various caching approaches, and the DOM-based caching in general seemed the best alternative for all but IE <=8.

Its not quite as straightforward as hanging the instance(s) on the DOM node for all browsers however, as this compounds the memory problem in IE, so we will likely need to continue the off-DOM cache for those browsers.

Look for something to land in 3.7.0 to address this issue. Ideally I can just merge in John's patch, however there will likely need to be a forked implementation for IE <= 8.

I'll take a stab at integrating an IE version into the patch.