Currently, the external view embedder (which was authored later than the GPU surfaces) can only render into overlays. However, this prevents us from presenting a unified view of composition into the root layer as well as overlays without adding hacks to the GPU surfaces to bypass rendering into the root. These hacks come in the form of the render_to_surface_ ivar on the GPUSurfaceGL. In the absence of these hacks, extra resources may need to be allocated unnecessarily. Notably, this hack is absent from the software surface.

The GPU surface APIs must be refactored to make the external view embedder a first class citizen of the rasterization subsystem.