feat: route audio from CEF #1517
Conversation
|
Is there any progress on this feature? |
|
@qu1rk3y I did this prototype implementation a while back because I was curious on how hard it would be, but haven't done as much testing as I think it needs before it can be merged. If anyone wants to give it a try, there is a build based on top of the master branch at the time of writing at https://builds.julusian.dev/casparcg/casparcg-server-b52edd40855b659798c9b78643e66d1d8e6aebdc-windows.zip |
I tried to test this but I was unable to run this server, it was complaining about an asio io_service. I don't know if the error comes from that or the diagnostics exception.
This happens in b52edd4 as well as latest master build e4e9ed2 |
| @@ -302,7 +369,10 @@ class html_client | |||
| { | |||
| std::lock_guard<std::mutex> lock(frames_mutex_); | |||
|
|
|||
| frames_.push(std::make_pair(now(), core::draw_frame(std::move(frame)))); | |||
| core::draw_frame new_frame = core::draw_frame(std::move(frame)); | |||
| last_generated_frame_ = new_frame; | |||
There was a problem hiding this comment.
@Julusian Could you talk me through how the move semantics works here? As far as I can tell, the draw_frame assignment operator moves the content of new_frame to last_generated_frame_ which would leave new_frame invalid.
There was a problem hiding this comment.
As far as I can tell, the draw_frame assignment operator moves the content of new_frame to last_generated_frame_ which would leave new_frame invalid.
The parameter of the assignment operator is not a reference, so will be a copy of new_frame making it safe to mutate. You have made me question this now, but this same pattern has been in place for draw_frame for years, with the core::frame_producer doing it with any draw_frame that gets produced by every producer.
There was a problem hiding this comment.
Ok, I see it now. I missed the implicit call to the copy constructor during the assignment. It looks OK.
| auto audio = audioResampler_->convert(samples, reinterpret_cast<const void**>(data)); | ||
| auto audio_frame = core::mutable_frame(this, {}, std::move(audio), core::pixel_format_desc()); | ||
|
|
||
| { |
There was a problem hiding this comment.
This crashes on Linux. If i comment out this entire block, it runs.
There was a problem hiding this comment.
Perhaps this is from the hardcoded 8 in
?I appear to have opened this PR a few hours before I made a change to support 16 channel audio 28c4dcd
That would explain the slowness/broken audio you are experiencing too
| { | ||
| std::lock_guard<std::mutex> lock(frames_mutex_); | ||
| if (frames_.empty()) { | ||
| presentation_frame wrapped_frame(last_generated_frame_); |
There was a problem hiding this comment.
Again, I don't understand how this would work. When you construct a presentation_frame from a draw_frame, you move the draw_frame into the presentation_frame. Doesn't that invalidate last_generated_frame_?
There was a problem hiding this comment.
Similar to the other question, the constructor parameter isnt a r-value or l-value, so I think it is fine
Inspired by/based upon a start made by nxtedition at https://github.com/nxtedition/casparcg
This needs a bit more testing to ensure it works smoothly and without glitches.
In particular, when audio starts/stops within the page, and when motion starts/stops
This also needs a sync test, to check that the audio is in sync and remains in sync with both video and audio pausing and resuming