Briefly discussed the issues around typesetting unjustified paragraphs. The main problem is that we don't have a good method to define a quality function for such paragraphs so that we can compare solutions.

Another requirement is that any such function should not kill the use of dynamic programming.

One way to address this problem is to define a number of functions that a paragraph can take as a shape; however this is perhaps far too limiting and may visually show.

Another approach would be to define a kind of curve class (like the visual classes lines currently can fall in) only that there would need to be a few more than just four and we need a different optimizing criteria, ie, while the visual classes right now are minimized, that is, TeX attempts not to apply drastic changes, the approach here would be to consider some change (but not too much) better than no change. Also the classes would nee to capture in a limited way the fact that the curve is already moving into a certain direction for a while (which may be a problem for dynamic programming).

food for thoughts ...