FWIW this part is true, but sort of mixed up. It's not really "wayland" that's at issue (Wayland surfaces are just shared memory like they are with any other compositor). It's that the Android GPU HAL wasn't designed with the sharing of contexts in mind, and so getting accelerated rendering out of the VM isn't a natural thing.
And this isn't my area, and I have no expertise to offer except to say that I too am hopeful folks work this out in both a HAL- and open-source friendly way.
And also to say that I'm running routine wayland apps on a chromebook every day and they render and composite just great with Mesa/LLVMPipe contexts, even things like e.g. STL slicers you'd expect to want the GPU. The need for acceleration is real, but limited to some special use cases.
And this isn't my area, and I have no expertise to offer except to say that I too am hopeful folks work this out in both a HAL- and open-source friendly way.
And also to say that I'm running routine wayland apps on a chromebook every day and they render and composite just great with Mesa/LLVMPipe contexts, even things like e.g. STL slicers you'd expect to want the GPU. The need for acceleration is real, but limited to some special use cases.