I'm not sure how complicated this will be.
Ok, I think this test is wrong.
The event should dispatch to the front-most element under that mouse location. (I'm sure there is a spec which says this.)
In this case, that's the SVG <rect> element which is semi-transparent. Since pointer-events defaults to "visiblePainted" the <rect> element can actually accept events, and thus the hit testing stops. An event is dispatched up the tree to the <rect> element, through the <object> and <svg> elements, but NOT touching the <html:a> elements which are not in the ancestor tree for the <rect>. Then the event bubbles back down through the same tree, again, never touching the <html:a> elements.
If the <rect> had pointer-events="none" specified, then it would fail to hit-test and the <html:a> element should actually be clickable.
We still fail this test with pointer-events="none", not sure why. That might be real bug on our part, or I might be misinterpreting how hit testing should actually work.
Ok, I've looked into this more w/ our implementation. Even when the <svg> declines hit-testing via pointer-events="none", the <object> itself is still hit. Which makes some sense. At least it would make sense for an absolutely positioned empty <object> to block any mouse events from reaching elements below it. But perhaps in this case, since the <object> is just a container, if its kids fail to hit test, it should also fail.
To "fix" this would simply require overriding nodeAtPoint in RenderPartObject, RenderPart, or RenderReplaced (probably RenderPart) to only hit-or-not depending on what its child render objects say, except in the case of fallback content, where it should probably participate in hit-testing like normal.
I CC'd hyatt so that he can comment on if he feels <object> should be accepting the click event itself when its kids refuse it.
What do the specs say? I seem to recall SVG has some sort of rule about events always hitting the "canvas" and going to the root element or something. Anyway CSS doesn't define hit-testing, we should really come up with a spec to define this sometime.
AFAICT the clicks should always hit the <object> element. Until this is specified otherwise, I think this should be closed as invalid.