When you change the grammar, does ALL parser/tokenizer state get wiped, or is some of it retained? Just wondering if you could change only the rules or only the tokens and leave the other cached.