I did some more NFA work. Character tests now match against a character class datastructure like the one I intend on using for Unicode's 24 bit character space, and I slightly altered the semantics of the min / max values of loop sentinels. I also discovered a few bits of extraneous code/data in nfa2.py, which I took out of nfa3.py.

The code: nfa3.py and out3.txt.

Update: Oops, failed to upload the code.