Processing of intersection observers doesn't match implementations

[emilio]

https://w3c.github.io/IntersectionObserver/#update-intersection-observations-algo says:

Let observer list be a list of all IntersectionObservers whose root is in the DOM tree of document. For the top-level browsing context, this includes implicit root observers.

This means that, per spec, if you're in a cross-origin iframe, a new IntersectionObserver() without a root node, is supposed to be processed in the top browsing context.

I don't think that matches what engines do (it's not implementable in Gecko at least).

What gecko does is, it uses the IntersectionObservers current document on construction as the document from where it's processed. That seems to be what WebKit is doing too here.

@szager-chromium do you know what Blink is doing? Should we spec the current / constructor document behavior or was the current language intentional?

[emilio]

cc @smaug----

[emilio]

This was discussed during TPAC. Tentative plan is to adopt WebKit / Gecko behavior and write tests. cc @szager-chromium @rniwa