CFF automatic subroutinization about fontkit

[Pomax]

making this work efficiently is a good source for a paper or two - naively, it's the byte equivalent of finding the largest common substring in a set of strings, marking it a subroutine, reducing all strings, and then running the algorithm again, repeating the process until there are "common substrings" left, so having a good guide to what is worth turning into substrings, which charstrings are worth not splitting up at all even if they do have something that could conceivably be a substring, which boundaries a substring should not cross, etc. is far from trivial, and a good thing to write a paper around (even if it turns into a white paper. Tracking hypotheses, testing, and (in)validation makes for good guided code, with reference material to fall back to; thing Adobe Tech Notes)

It's easy to run offline as a production step, even if it takes an hour to find the optimal subroutine mapping - just turn on the subroutiniser, have it run in the background, eventually it'll be done. But it's much harder to write a real-time subroutine abstractor that is fast enough to feel like an acceptable delay while also finding creating a meaningful set of subroutines: pulling that off would certainly be worth either a long post on a blog, or a paper-like document that people can read to understand the work that will be going into this.

[devongovett]

Also, currently fontkit doesn't produce fonts from scratch, it only subsets them using the existing subroutines from the original font. Unless we add support for creating fonts from scratch, I think this issue may be better sent to opentype.js, which is more of a font creation library.

[be5invis]

@Pomax No, it is the “optimal grammar problem” in the area of data compression. Some other algorithms works in linear time and performs quite well, like SEQUITUR or RE-PAIR.