Uncool

I'ld think implementing strings would help somewhat, but I didn't manage to convince myself, and biggest reason: I just don't feel like, having strings as lists of characters works for Haskell too and gives nice algorithms. The real problem is 1GB of character data, which are probably mostly copies of themselves. Other stuff I could do:

1. Make the data unique by storing strings in a map and only extract unique copies.
2. Don't retain specific character data, it's probably all positional information anyway.
3. Change the runtime such that all shareable data is shared.

I like 1 the most, but that would best be solved by having a constant pool referred to as a variable, and I don't have variables at the moment.

(Did I mention using a symbol table? A simple transform which just visits all AST nodes once would do it too, of course.)