PermissionStatus's memory management is unclear

[smaug----]

Since permission may change at any point, PermissionStatus should stay alive (at least in its current form) as long as it has any change event listeners.
That would follow the model WebSocket for example has.
https://html.spec.whatwg.org/#garbage-collection-3
This all to avoid exposing gc behavior to the web.

[marcoscaceres]

@smaug----, this seems pretty straightforward to add (based on WebSockets in HTML).

I'm wondering, what are some of the interop implications we should watch for here? Anything observable and/or testable that would result from adding the memory management?... I'm assuming no, but what to make sure.

(maybe related to #162)

[smaug----]

Can wpt test changing permissions? If it can, this should be testable: create PermissionStatus, add event listener and drop any explicit references to the object. And then try to force a GC (allocate lots of memory or something) and then change permission.

[marcoscaceres]

Ah, got it! Yes, changing permissions should be testable through Web Driver (at least in Chrome, for now). I'd be reluctant to make any assurances about about CG actually occurring, but I'll throw in a bunch of iframes in the hope that creates enough garbage to cause CG.

In any case, just doing the setup in a block scope should be enough for to make an instance a CG target.