You may remember my
recent node about extending regular
expressions to allow recursion.
I've completely rewritten the implementation, so that it backtracks
properly. All the known bugs have been fixed, and both the regexes
that
Juerd posted last time now seem to work.
The web page
has been updated to reflect the changes, and the new patch is available
from there.
The most significant functional change is that calls must be
recursive. With the previous version you could write
/(a*)b(?1)/
and it would mean the same as
/(a*)ba*/. Now you can only
call out to a group that's still open. I made that change for technical
reasons related to the implementation (so that I could use stack rather than
heap storage, basically), but I don't think it's a big problem. Let me
know if you disagree.
Proper detection of left recursion is implemented as well,
meaning that a regex like /((?1))/ won't even compile now.
On a slight tangent, is anyone else interested in the idea of
(as an experiment) hacking perl to use the PCRE library in place
of its built-in engine? I don't think it would be all that hard.