We want to have an efficient way of integrating content like WebGL, WebGPU, HTML5 Canvas and Media between each such subsystem as well as deeper in the graphics/display pipeline. Undergoing architectural changes also impose the necessity of cross-process migration for these buffers.
On Linux, DMABuf seems an ideal reference solution. This would allow ANGLE to be used to render into DMABuf objects. GStreamer pipelines with hardware decoders outputting DMABufs would be most efficient to handle, while raw decoded data from fallback software decoders would be manually uploaded into dmabufs.
DMABufs aren't universally available. For other platforms, we would require equivalent mechanisms (like e.g. AHardwareBuffer on Android).
This is a tracking bug for all the relevant changes and regressions coming from this work.