Code review comment for lp:~mir-team/mir/cursor-spike-phase-2-resubmit

(6) The API implication that a surface has a single cursor setting is misleading. Obviously the cursor will vary over the surface depending on the widget. So although the one-cursor-set-per-surface sounds wrong, I now realize it does work so long as you make the app/toolkit do the hard work of changing the cursor as it moves.

On the other hand, Mir does already have the concept internally of input regions (still needs work). But since we already have multiple input rectangle support per-surface in libmirserver (BasicSurface::input_rectangles), wouldn't it be better to expose that to the client API and simply attach a MirCursor to each input rectangle? Then the server could do the cursor changes for you. I think that's what X and Win32 does... And it certainly eliminates the redundancy of asking every toolkit to re-implement cursor tracking.