It seems to me that the algorithm fails even on very simple cases.

Given the following tree: div#a > div#b
and do the following steps:
1. call a.requestFullscreen()
2. call b.requestFullscreen()
3. call document.exitFullscreen()

The expected result is, #a, as well as the viewport, is still fullscreen. Is it correct?


However, it seems to me that the current spec will cause very strange result. When document.exitFullscreen() is called:

In step 4, the "collect ancestors" returns [document] to docs, and since |document| is the topLevelDoc, so in step 6, resize is set to true, which cause the viewport be restored to normal dimensions in step 8.

In step 9.1, the "collect ancestors" again returns [document] to docs, and finally in step 9.6, #b is removed from the fullscreen element stack of the document.

Finally, #a is still in fullscreen element stack, while the viewport is no longer fullscreen.


To fix this, I think we should add a step at the start of "collect ancestors": If fullscreen element stack of doc consists of more than one element, return an empty set.