Is the complexity of cross-frame observation worth it?

[smfr]

Allowing an intersection observer to observe an element in a subframe adds a lot of additional complexity, and risk to the implementation:

• You have to track layouts and scrolling in subframes, but then "Run the Update Intersection Observations Steps" in some ancestor frame, but then ensure that all the subframes which contain target elements hit by that algorithm are laid out.
• Lifetimes of IntersectionObservers are tied to lifetimes of elements in a different document

Is there a major use case for cross-document observation?

[szager-chromium]

Is the complexity any greater than the in the implicit root case (i.e., root is the top-level document, target is in a subframe)?

In the chromium implementation, lifetime management was indeed tricky to get right, but wasn't any trickier for the case you cite than for the implicit root case.

Same comment about ensuring that layout is clean in both ancestor and descendant -- this is just as true for the implicit root case.

[miketaylr]

@smfr any thoughts? Or should we close this issue?