Henry wrote: > my recollection is that Roger > implements the Boyer-Moore algorithm
That's my recollection too (or some kind of variant, like Horspool), though the only real evidence I could find quickly was [1] . That said, I imagine the other languages k.os tested also use Boyer-Moore or some other commercial-grade search algorithm, and he found their built-in string searches too slow (even when explicitly parallelized). He had to roll his own. I expect he leveraged global information about the search strings, something J doesn't currently do. If not, I'd be interested in how he achieved his performance improvements over standard K/Q (maybe leveraging application-specific constraints? I haven't read his Usenet thread yet). I'd be very interested in how a naive implementation in a recent Perl would compare to k.os' other benchmarks. -Dan [1] Roger mentions Boyer-Moore and E. : http://www.jsoftware.com/pipermail/programming/2007-January/004673.html ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
