I understand this, but it doesn't solve the problem. Imagine if someone said: "the developers of grep have come out with an SDK so that other text-oriented programs can integrate text matching!" The idea of grep-as-library is great, but it doesn't replace having pipes so that you can use grep with arbitrary programs, most of which have no idea they're being used with grep.

To build larger abstractions on top of these things, we need looser coupling. I want to draw and erase with Pencil, but use a manual dial to set the pressure because I do very precise architectural drawings. I want to draw with Pencil but use Mighty to make my lines snap to a grid and some other French Curve thing I hacked together with some dials for snapping to curves. I want to use two Pencils at once on two separate iPads and draw together with someone. I want my LeapMotion to understand my hand gestures to rotate data in an Excel pivot table and project it into a graph.

Yes, I want crazy things. But abstractions work to build crazy, unexpected text-oriented programs in Unix. So what are the abstractions we need to build arbitrary crazy programs with UIs and hardware peripherals?

Yes, this is like the old DOS days where e.g., your games had to be programmed with your specific make of sound card or you would have no sound (or PC speaker only). What we need is an OS-level API that allows apps to be mostly agnostic about what device is giving the input. And device manufacturers won't need to release app integration SDKs -- just code the OS driver.

And then we have windows all over again? I'm not sure that SDKs are that much better, but windows drivers have been the cause of many wasted hours...

Many OS's/PC's can even get laptop trackpads right, and we've been making those for over a decade: and shipping tens of millions. Apple gets it right, because they write their own drivers and OS.

I don't have high hopes for abstractions on pressure-sensitive styli.