Inconsistent cross-browser behavior for mutating child <source>s of a <video>

[nicolo-ribaudo]

What is the issue with the HTML Standard?

I'm opening this issue here rather than in the browsers bug tracker because I'm not even sure about:

• what the HTML spec says
• what the intended behavior is

Consider a <video> element with two <source>s (source1 and source2) with a URL that returns an error. After that we try (and fail) loading both, consider these two (alternative) test cases:

• A: we add a third valid <source> at the end
• B: we set the .src of the first <source> to something valid, and then move (with .moveBefore()) it to the end of <video>
• C: we set the .src of the first <source> to something valid, and then move (with .moveBefore()) the second <source> before the first one

What I'm seeing in browsers is:

	A	B	C
Firefox	loads	loads	does not load
Chrome	does not load	does not load	does not load
WebKit	does not load	moveBefore not supported	moveBefore not supported

Test case (AI-generated based on my description): index.html

I think the behavior that the spec describes is:

• A: loads, because:
  • the new source is inserted between the two nodes that define the pointer (source2 and the original list end), so the pointer is updated to point between the second and the third source
  • the resource selection algorithm continues from step 22 of the "children" mode, and then jumps to step 13, 15, 16, 17, 2, 3, 4, 5, 6, 7, 8, 9.
• B: loads, because:
  • the source1 is moved between the two nodes that define the pointer (source2 and the original list end), so the pointer is updated to point between source2 and source1
  • the resource selection algorithm continues from step 22 of the "children" mode, and then jumps to step 13, 15, 16, 17, 2, 3, 4, 5, 6, 7, 8, 9.
• C: 🤷, I don't think the pointer is updated, but it's left in an invalid state because:
  • no node is inserted
  • the node that is moved (source2) is not moved between the ones that define the pointer (source2 and the list end)
  • no node is removed

I believe it's likely I'm interpreting the wording about inserting/removing/moving nodes inside the media element too literally (but I am doing it because it links to their definitions!), and C is intended to behave the same as B.

Probably the intended behavior is that all three test cases should load?

[andreubotella]

cc @whatwg/media

[nicolo-ribaudo]

An extra case: moving a source into an empty <video>

I believe this does not trigger the resource selection algorithm in the current spec, but both in Firefox and Chrome it causes the source to be loaded.

[nicolo-ribaudo]

And one more: after waiting for a new <source>, Firefox (the only one that will load it) transitions the networkState from NO_SOURCE to LOADING synchronously, without first waiting for a stable state (but then it waits for a stable state before reading the .src from that new <source>).