VideoFrame is missing from the "not origin-clean" definition for CanvasImageSource

[weinig]

What is the issue with the HTML Standard?

In the list of different CanvasImageSource image types when defining "not origin-clean", https://html.spec.whatwg.org/multipage/canvas.html#the-image-argument-is-not-origin-clean, a definition for VideoFrame is missing.

[Kaiido]

Though, can a VideoFrame actually be not origin clean?
It does throw when the constructor is called with a "dirty" source, and I don't think any other producer will accept dirty sources¹, also the VideoFrame interface doesn't have any safe-guard for e.g. the copyTo method, which would allow reading of the actual frame's content too. Maybe someone from web-codecs can confirm? (cc @padenot)

And if it is the case that they can't ever be "origin-dirty", shall we still include it somehow in that algorithm, maybe at least as a comment?

Footnotes

1. A MediaStreamTrackProcessor might accept producing VideoFrames from a MediaStream captured from an element (video or canvas) which becomes origin-dirty, but the MediaStream is muted in such a case. ↩

[annevk]

We should probably make it an actual algorithm that returns true or false. And given that we branch on type and call it for all types we have to handle VideoFrame one way or another. Currently it's undefined behavior which is bad.