Some other points worth mentioning

• The generated lexer supports Unicode.
• The first passes of the lexer preserves comments and white space. This turned out to be useful when reusing the lexer in an Haskell-to-HTML converter.
• How is the "maximal munch" rule implemented?