I don't think re2c does backreferences, so it might not be usable for Lynn's purposes. It also means any comparison would be apples to oranges, since regex matching with backreferences is a much harder problem.
Some of this is simply curiousity on my part. (... and too much time on my hands?) Also, I've been quite impressed by the performance of xpressive and regex (haven't looked much at spirit) ... my initial expectation was that something general purpose and powerful would be worse than an order of magnitude slower than "rolling your own" hand-tuned recognizer. I think it might be valuable for a person in the position of evaluating whether to put in the time to learn boost::regex and/or boost::xpressive could find information on the performance trade-offs involved: "So, how slow is it?" (and then, "How bloated is it?" "How hard is it to learn?" "How hard is it to build and integrate?") Epecially that it probably won't be nearly as negative as a first guess might be. (but of course, "it depends" applies to everything ). BTW, my first impression is that re2c might be useful for some of the things I work on .... it could get the code to the "area of interest", where more elaborate replacements could be done. I don't think I understand "backreferences" ... my mind hasn't absorbed past DFA's and "greedy" always seems to be a liability. Another BTW: my impression is that pcre doesn't do substitutions .... based on an email from its author, Dr. Philip Hazel. Did I misunderstand? I didn't see anything in its api to indicate substitutions were possible.