In Dercuano I wrote about a “circle portal GUI” ZUI design consisting entirely of circular windows. Your user interface is a circular viewport onto an infinitely zoomable canvas, which contains other such circular viewports onto other parts of itself, each possessed of only a position and size, a destination position and size and orientation, a z-order, a background color, and a translucency. You can create these portals, duplicate them, move them, move their destination, resize them, resize their destination, change their background color, change their z-order, and follow them — they function as hyperlinks. Also you can “snap” them, destroying them and copying the region of canvas they view onto the place where they are displayed.
By arranging some empty portals with a given background color somewhere, you could create a letterform, and then by making portals that view that letterform in various places you could make copies of the letter. By arranging dozens of such letters together in a font, you could have a way to write text in that font; if you made a “font portal” viewing that font, you could make portals onto the positions of letterforms in that portal, which could also be used to spell text; by changing the destination of your font portal, you could change the font used for that text.
(You’d probably want to have a way to interpose a level of indirection like that after the fact.)
If you wanted to draw something in some changeable colors, you could do that by putting your palette in one place, and making your drawing out of portals onto the palette.
So this single type of object serves as a graphical primitive (since you can set the background color), a limited IFS generator (without shearing or nonuniform scaling, since each transform only supports four degrees of freedom rather than the usual 6), a hypermedia navigation system with live preview, a graphical instancing system capable of some use as a stylesheet, a universal “view source” button, and so on, all through direct manipulation, though a sort of direct manipulation that would probably be super confusing, just due to its hall-of-mirrors nature.
(I implemented a tiny prototype of the system one day, but didn’t get far enough to get a good sense for how to use it. I imagine you’d want to add some other graphical primitives and interaction modes; dragging letterforms one by one is maybe not a great way to write text.)
I was thinking today that it would be interesting to do the same thing in 3-D, using spheres instead of circles. The sphere portals could display what was within their target volume, like you might imagine looking into a crystal ball; they could display what was visible by looking through it (either in a given direction, or in the direction you’re looking in); or they could display what would be reflected from the volume around their target volume, just as a silver ball displays what is reflected from the volume around itself. This last item would have the benefit that the target volume could itself be an object in the world, so the hypertext links in this system could be bidirectional, which might actually help a little bit with the confusion, though it would impede uses like instancing a character from a font. (You could probably get an acceptable approximation by using really small target spheres.)