I want to clarify what we intend in this bug before I write a lot of possibly unnecessary code. Completion here means:

* deletion of GLSurfaceAdapter
* deletion of PassThroughImageTransportSurface (PThImTS)
* and hence removing the indirection layer wrapping a GLSurface with a PThImTS.

To do this while preserving the existing functionality, I will have to add a mechanism by which a GLSurface instance can provide similar functionality as PThImTS: recording latency info and augmenting the SwapCompletionCallback.

If this is the intent: how sensitive are we to adding a small number of pointer-sized fields to GLSurface?


Or, did we just mean the (much simpler) task of folding GLSurfaceAdapter into PThImTS?

The latter seems like an obvious win which wouldn't cause any layering issue.

Moving LatencyInfo into GLSurface may be more objectionable - it would be a bit of a layering wart. I would probably prefer moving that functionality into GpuCommandBufferStub if possible. Maybe using a hook into GLSurface (swap observer or some other mechanism)?

SGTM. I'll do the simple case first.

Thanks for idea of moving latency tracking into GpuCommandBufferStub -- it hadn't occurred to me and I think it makes the follow on change much simpler than the scheme I had sketched out yesterday.

Cleaning up old Proj-Mustash labels.