• There is doubt whether the API is well designed.
• This may not be a high enough priority right now. Part of the team is responsible for implementing new "capabilities". Reaching outside the content area and contributing items to the browser's context menu is a new capability. yuzus, tkent and I are not on the capabilities part of the team; they are busy bringing web developers other new capabilities. They should be heavily involved in the decision to implement this but they can't right now.
• It is hard to prioritize for other reasons. For example, this feature only exists in Firefox, so we're not making interoperability better by implementing this.

Yeah, seems so — and that’s also what I had thought, but WebKit is the only UA to have implemented that spec change, and WebKit returns about:blank there. So it appears to be a bug in WebKit, caused by the Location code not having the Frame object that it should have.

Take a look at https://searchfox.org/wubkat/source/Source/WebCore/page/Location.cpp#53-54. That entire code assumes there’s a single frame() Frame object that it gets the location from. But for that test case, it’s null. So I guess changing the src value of that iframe either causes the childDoc to be prematurely completely disassociated from childFrame, or else that childFrame gets prematurely destroyed.

I don’t know — I need to trace through the rest of the code more. But code-wise, I see that Location code has access to a LocalDOMWindow window, along with access to that frame() Frame, and naïvely I wonder if it should be getting the document from that LocalDOMWindow object rather than from the Frame.

Anyway, if I can’t get it figured out, I’ll raise a bugs.webkit.org bug.

The failing test case is at https://wpt.fyi/results/html/webappapis/dynamic-markup-insertion/opening-the-input-stream/url.window.html

annevk: If I make a change to the Location-handling code in WebKit (specifically, to the Location::url() code), what WPT tests should I check for regressions?

I just locally ran all 926 tests in the html/webappapis and html/browsers subtrees. Are there WPT subtrees in addition to those that I should also run and look at results for?

emilio: I'll do some more investigation. Might take a while. If zcorpan has some notes somewhere from when he investigated all this for <input type=checkbox> that might be useful too.

Ideally I'd like to tackle this in stages, but there should be sufficient information for each stage and earlier design decisions shouldn't cause us to run into issues later on.