When I was living with David Levitt, who worked with Alan Kay at Atari Cambridge Research, I was snooping around his bookshelves and read through some of the fascinating notebooks, in which Alan Kay wrote about the idea of a two way mapping/projection between the model and the view.

It's always seemed to me that the "controller" was just extra baggage, because the view knew the most about drawing the model, and naturally would also know what it needed to reverse that projection back from user input through to manipulating the model. But right now we usually meticulously write the view drawing and mouse tracking code by hand, without a lot of help from the system. (That's something I consider a high art form, which I enjoy diong, so I'd probably be horrified and out of a job it it were too easy! ;)

I think Ben Shneiderman's term "direct manipulation" describes the style of user interface that Alan Kay was getting at -- or at least DM has to deal with a lot of the same problems: "continuous representation of objects of interest, and rapid, reversible, incremental actions and feedback" according to http://en.wikipedia.org/wiki/Direct_manipulation_interface ...

By that I mean DM user interface programming could benefit from some kind of automatic inverter, but usually it's just implemented in an ad-hoc manner, usually by input tracking code in the view.

I think the way Smalltalk and Self/Morphic use reflection to present editable user interfaces to objects is another example of a two-way mapping between projecting the view on the screen, and editing the model via user input. Booleans whose values you can toggle, numbers whose values you can drag up and down, color pickers and things like that are all fine and good, but for many properties it has to fall back to standard widgets that don't know much about the semantics of the data they're viewing or editing.

I had the opportunity to play around with Sk8 when I was at Kaleida. That certainly let you drill down, see and modify everything there was in the system, and it was totally prototype based! Plus it had cool looking windows with neat shapes, which I like! ;)

I'm curious about what Kay meant by "projection", and what successful examples there are of automatic bi-directional mapping, and powerful user composition. I take it that projection is the problem that you lose dimensions and have to hide information by flattening the structure out onto the screen (via a 3d projection or scaling or filtering or whatever infovis technique you're using), and there's no visible place to click or affordance to edit some every property of an object (or you have to build scaffolding in the form of clumsy indirect manipulation techniques like browsers, outliners and pop-up menus).