We try to make sure all features in noVNC work on all current browsers, so it's annoying that Firefox isn't ready yet. :/ Is the new API the only method? Have you looked at using the older WebRTC API?

WebRTC is not a good fit for this as it does not expose an API for decoding video at the required level of abstraction.

The WebCodecs definitely looks like a better fit. No arguments there. But if WebCodecs isn't fully supported yet, the question is if it's possible to build a more complex solution around WebRTC that works on more browsers.

The WebCodecs definitely looks like a better fit. No arguments there. But if WebCodecs isn't fully supported yet, the question is if it's possible to build a more complex solution around WebRTC that works on more browsers.

I suppose that it might be possible to create two peers within the browser, connect them together and stream video from one to the other which would render into an offscreen video element that gets duplicated onto the canvas. I have not explored this option. It's just hypothetical. It would be very hacky.

In any case, I don't think that this would be worth the effort as WebCodec support is very close to being complete in Firefox.

That does sound very complex. I was hoping it was a smaller step, like having to frame the data better, or some extra steps to get the data out (like creating some <video> element). Let's hope Firefox is ready soon then.

The requested changes have been made.

I added a couple of basic tests for H264Parser and H264Context, but I intend to add more. I hope it's OK to export those. I don't see any other way to test them.

I hardcoded the colour space because it seems that Firefox and Chrome have different defaults. Colour spaces were never specified for the Open H.264 extension as far as I can see. The protocol is missing a way to tell the client what colour space was used for encoding the video. Maybe we can decide a colour space and just say, in the protocol, that the video should always be encoded using that?

I added a couple of basic tests for H264Parser and H264Context, but I intend to add more. I hope it's OK to export those. I don't see any other way to test them.

They should not require explicit testing, as they are not part of the interface. Can't we feed the data to H264Decoder? That's the interface we want a stable behaviour out of.

I hardcoded the colour space because it seems that Firefox and Chrome have different defaults. Colour spaces were never specified for the Open H.264 extension as far as I can see. The protocol is missing a way to tell the client what colour space was used for encoding the video. Maybe we can decide a colour space and just say, in the protocol, that the video should always be encoded using that?

VNC doesn't explicitly define color space anywhere. But given its nature, sRGB would be the base assumption.

VNC doesn't explicitly define color space anywhere. But given its nature, sRGB would be the base assumption.

That is, assuming the data stream doesn't specify an explicit colour space. In which case, I'd say anything goes. It's up to the client at that point to match things up.

VNC doesn't explicitly define color space anywhere. But given its nature, sRGB would be the base assumption.

That is, assuming the data stream doesn't specify an explicit colour space. In which case, I'd say anything goes. It's up to the client at that point to match things up.

Well, it's YCbCr encoded, so I don't think sRGB applies.

I added a couple of basic tests for H264Parser and H264Context, but I intend to add more. I hope it's OK to export those. I don't see any other way to test them.

They should not require explicit testing, as they are not part of the interface. Can't we feed the data to H264Decoder? That's the interface we want a stable behaviour out of.

Sure, we can do that too, but we get more thorough test coverage by testing the individual components.

If you see no value in test coverage and just want the end-to-end test, I can throw them out. It's up to you.

I don't think we have any plans of reusing those components elsewhere, so it's quite sufficient to test only that which we might see from an actual VNC server.

I don't think we have any plans of reusing those components elsewhere, so it's quite sufficient to test only that which we might see from an actual VNC server.

I think that we have very different ideas about what unit tests are supposed to achieve, but it's up to you.

I've added some more tests and I've fixed some bugs that I found when writing them.

Generating good test data was actually a bit of a headache and I was unable to get ffmpeg to do colour conversions correctly so I ended up writing a C program to do it using libavcodec and libswscale. I looked deeper into colour spaces along the way and did my best to set them up correctly in other projects.

I wrote some lower level unit tests to help me figure out some things and I've included them, but I can throw them out if you don't want them. They did help me to find some bugs and to understand what was going wrong with the higher level tests.

The tests don't run on all the CIs, due to lack of WebCodecs support. Do you know if a test file can be skipped depending on whether a feature exists or not?

The tests don't run on all the CIs, due to lack of WebCodecs support. Do you know if a test file can be skipped depending on whether a feature exists or not?

Yes, you can do runtime checks and skip tests. An example here:

Thanks, the test skipping works.

Now it appears that I've exposed a bug in Safari. It doesn't scale from limited range up to full range like it should when rendering the frame to canvas. How should that be handled?

Well, I managed to make the tests pass on all browsers by carefully crafting the test data. It's now BT.709 full range.

Ideally, the browsers should get this right, but I can't be chasing down bugs in all the browsers.

Ping?