 * LALR ? Not sure it's worth it.
 * ProgressiveLexer: subset construction in byte mode
 * Serialization, maybe
 * Include position in input as attribute of token
