Re: extended rx.el

[Stefan Monnier]

> But you said that's false!  \(?:foo\) normally isn't slower.

It's in any case marginally slower for the simple reason that it takes more
time to compile the regexp to its bytecode form.

> The issue is whether when generating regexps a la Rx, for instance,
> you worry about extra groups and expend effort eliminating them (as Rx
> already does, actually).

And I think the answer is that it should try to avoid using them
when possible.

> It would be useful to have the info _somewhere_ anyway -- regex.c is far
> from transparent, I guess especially if you don't know/remember your
> theory.  (I don't mean that's surprising or a problem, of course.)

Yes, the should be a blurb at the beginning of regex.c discussing the
performance of the code.
I don't have time to write that now,


        Stefan