In message <9B3E950CB293D411ADF4009027B0A4D203091090@PARSV011>, Stephane Bailli ez writes: >For the records, when you say "can regain the performance lead", do you mean >that you have some figures with other packages ? If yes could you give use >more info ?
Jeffrey Friedl (author of Mastering Regular Expressions) claims that the J2SE 1.4 java.util.regex package is faster, but he doesn't make his benchmarks publicly available because he doesn't have the time to deal with questions or other support. I trust him, but I think a contributing factor to the performance discrepancy is simply this char[] conversion issue. Eliminate that and with HotSpot 1.4 JVM my intuition is that jakarta-oro will be comparable or faster for most cases. Traditionally, the Perl-style byte-code based NFA approach used by Perl5Matcher has beaten out object-based NFAs like the 1.4 implementation. But it's hard to say how the new HotSpot optimizations will affect that. All I know for sure is that the toCharArray() approach in Perl5Matcher is now a loser. I haven't gotten around to seeing what the impact of moving to a generic CharSequence instead of direct array indexing will do. If HotSpot does everything its performance docs claim, there should be little difference, but I don't understand how HotSpot inlines virtual method calls when different interface implementations may be called from the same generic code. daniel -- To unsubscribe, e-mail: <mailto:[EMAIL PROTECTED]> For additional commands, e-mail: <mailto:[EMAIL PROTECTED]>
