You're right that things have come a long way. I'm not scared of the tokeniser any more, for example. The regex code still gives me the willies, but I guess you're happier with it than you were in 2002. :-) The most unsatisfactory thing about the PCRE recursion code is that the match data is still a plain list, when morally it ought to be a tree. If we could get that sorted, you could write an entire parser as a single regex!