Define the implicit root better in presence of adoption to a different browsing-context tree.

[emilio]

https://w3c.github.io/IntersectionObserver/#intersectionobserver-intersection-root says that the implicit root is the document node of the "top-level browsing context".

That makes sense, but if you're observing a node and adopt it into a new window, there are two potential candidates:

• The browsing context in which the intersection observer was created.
• The browsing context in which the target lives.

Both of those can have different viewports, as https://bugzilla.mozilla.org/attachment.cgi?id=9180905 shows. Which one should be chosen? If we choose the first, the target will presumably never intersect. If we choose the second, then we report intersections against the newly-opened window.

I think both of them would be reasonable. Gecko does something non-sensical in this case. WebKit seems to do the first, and Blink seems to do the second:

• https://webkit-search.igalia.com/webkit/rev/0cf28a346a4ad67f08eb1150c17fdb48ef6b29c7/Source/WebCore/dom/Document.cpp#7643-7645,7719
• https://source.chromium.org/chromium/chromium/src/+/master:third_party/blink/renderer/core/intersection_observer/intersection_geometry.cc;l=123;drc=5305c09e4169c2aa9f2c92cb8aace4a6ce232fc5;bpv=1;bpt=1

We should decide what is the right thing to do here and clarify the spec.

cc @szager-chromium

[emilio]

cc @smfr

[szager-chromium]

The comment in the WebKit code suggests that they intend to match chromium's behavior, which would be fine with me.

I think a case could be made for either behavior; but if I put myself in the shoes of the developer, I could imagine being surprised at not receiving an "intersecting" notification when I insert a target element into another window.

[emilio]

I think a case could be made for either behavior; but if I put myself in the shoes of the developer, I could imagine being surprised at not receiving an "intersecting" notification when I insert a target element into another window.

Yeah, on the other hand it seems weird/sketchy that only for the implicit case one IntersectionObserver can refer to multiple roots, depending on the observation target...

For now I aligned with Gecko with WebKit in https://bugzilla.mozilla.org/show_bug.cgi?id=1670327, because it was more on the lines of what the Gecko code was trying to do before that patch.