Le 07/06/2010 04:58, Eric Niebler wrote:
I'm always suspect of benchmarks, but I figured I'd post this anyway.
http://astl.sourceforge.net/bench.7.html.
It seems someone wrote a generic automata library and a regex engine on top of it, and compared its performance to boost.regex, xpressive, pcre, and greta. Of course, his library wins handily. (The purpose of benchmarking is to make yourself look good, right? ;-) He doesn't say that (a) his engine is a DFA, and he's comparing to NFAs; and (b) he's only implemented a tiny subset of regex features people would expect; e.g., no captures even, let alone backreferences: things that can be tricky or impossible to implement with DFAs. He also doesn't say what version of Boost he's using (tests run Nov. 2009) or whether he's using static or dynamic xpressive (probably dynamic; static can be ~15% faster).
Just an idea: for static xpressive, couldn't you detect at compile-time that the expression is truly regular, and use a DFA in that case?