> The code of the RE stuff seems [too] procedural for my tastes.
> I propose a more OO approach.
>
> So instead of having a program that is made of an array of character
> nodes use java objects > instead.
> This allows for one to instantiate re compilers/matchers from a
> factory with varying functionality.
> I'm not sure what the overhead would be like but I am thinking that most of it would
>be in the
> compiling.
I once wrote an RE package like this. Preliminary results are that it
was about 10 x slower (yes, a full order of magnitude) slower than
Apache Regexp for simple grepping, and almost two orders of magnitude slower
than Unix (BSD) grep, all on the same hardware. I may have had some horrible
bug that made it slower than it needed to be, but I advise you to
move cautiously, and not inflict this on the main CVS branch, unless
you can prove that the overhead is really negligible. Mine was not.
