On Mon, 24 Jan 2005 15:24:47 -0500, pete-cl-ppcre@kazmier.com wrote:
It is better now. However, I am no longer using DO-REGISTER-GROUPS as I realized that I intended to use REGISTER-GROUPS-BIND all along. In any case, I am thrilled with the performance of cl-ppcre now that I have set *regex-char-code-limit* to an appropriate value.
Actually, if your Lisp implementation has a decent hash table implementation the value of *REGEX-CHAR-CODE-LIMIT* shouldn't matter much during scanning - only when a regular expression is created.
The problem that I fixed in 1.1.0 was that your function created a new scanner each time it was called although this wasn't necessary. This resulted in much too much consing (because of the huge hash tables that were generated).
I think your library is great, my only problem now is the performance of READ-LINE in the free CL implementations (which is certainly not your problem). Its absolutely terrible in SBCL. I've inquired about it, but have not really received any good responses yet:
http://sourceforge.net/mailarchive/forum.php?thread_id=6400934&forum_id=...
Yeah, I've read that and I'm concerned, too. I'd love to see an encouraging response. Have you tried to download the trial versions of AllegroCL and/or LispWorks just to check how they cope with this?
Cheers, Edi.