Code review comment for lp:~vanvugt/mir/predictive-bypass-2

Also remember "6ms latency" means from the scene snapshot to display. For swap-interval-1 clients that are double buffered you need to add another 16ms to that (or 33ms for triple). But you can avoid the 16-32ms addition by using swap interval zero, which is why I've recommended trying "mir_demo_client_target -n" on desktop.

This branch only shortens the top layer (system compositor) in my original diagram. While using a swap interval zero shortens the middle/bottom layer:
https://docs.google.com/document/d/116i4TC0rls4wKFmbaRrHL_UT_Jg22XI8YqpXGLLTVCc/edit

The eye and the brain can easily perceive latency lower than 16ms and you quickly get used to it (and _want_ it lower than 16ms). Low enough latency and you start to imagine a fixed line or bar between your mouse and the screen. And then you never want it any other way...