Re: rx-1.5 ...

[Tom Lord]

   On Mon, Oct 14, 2002 at 10:24:06PM -0700, Tom Lord wrote:
   > approximations, the rule in the free software world is "ignore Tom
   > Lord as much as possible". 

   OK, I'll ignore the rule, as it was said by Tom Lord.
   Thus I won't ignore you.

Thanks.

   So you recommend rx from libhackerlab from ftp.regexps.com .

   Another new implementation of regexps is part of GNU libc CVS;
   I understand it was donated by IBM Japan.
   It also has dfa integrated.

Last I checked (quite a while ago) that implementation used an array
of pointers the same length as the string being searched.  So, it
isn't suitable for all applications.  This may no longer be true -- I
don't know.


   And Peter Tillier told me that Henry Spencer's new dfa implementation
   is capable to handle backreferences.  (If I remember that
   correctly.)

"new?"   The most recent I'm aware of is the one in Tcl, that does
indeed handle backreferences.


   Are there other regex implementations?  Which one should one choose
   and why?


There's a datasheet in the rx source tree that might help.

Though I don't know of any semantic bugs in Rx, most or all other
matchers had bugs, last time I checked.

There's some controversy about the right semantics of some slightly
obscure operations.  Last I heard, the glibc implementation had 
decided to take what I think is the wrong side in this controversy.

Rx affords a wide range of space/time trade-offs.  I use it by
default, everywhere.  The compiler (`regcomp') is too slow
(esp. compared to other implementations).  It can do unicode matching
-- but only for XML regular expressions, not POSIX regexps.  You do
have to tune it (by setting some run-time parameters) for most
applications.

Good ol' GNU regex 0.12 has a fast compiler and a tiny implementation.
I'm not sure how it's doing on bugs -- I think there was some
bug-fixing activity on it not long ago.  It is probably good for
applications without outrageous performance requirements.


-t