sZweaver@gmail.com (86 the z)

The code as posted here (and everywhere I’ve seen it) for matching balancing constructs does not work.

The problem lies in the meat of the pattern:

(?>
<div (?<DEPTH>) | </div (?<-DEPTH>) | .? )* (?(DEPTH)(?!))
)

let’s say input is: “Beforedivs<div>.stuff</div>more stuff</div>Afterdivs”

The pattern matches:<div>.stuff</div>more stuff</div>

That’s not a balanced set of divs. Here’s why: the <div> is matched and the DEPTH stack is pushed. The first </div> is matched and the stack is popped. Now the (?(DEPTH)(?!) comes in and tests if there’s anything on the stack and there isn’t so the pattern continues greedily. The second </div> is finally matched, the EMTPY stack is popped and again the test sees nothing on the stack so the pattern does not fail. The pattern is forced to backtrack and allow the last </div> in the pattern to match, and we get the erroneous result. The pattern does not account for unmatched </div>’s.

The good news is that solution is easy. Make it:
(?>
<div (?) | </div (?) | .? )*? (?(DEPTH)(?!)
)
I made the whole parentheses subexpression NON-GREEDY with the ? mark. Now the pattern stops after every <div>,</div>, or single ‘anything’ character (.?) matched and sees if it can match that last </div>, and if so it stops. This new pattern now matches:
<div>stuff</div>

sZweaver2112@gmail.com (86 the z)

Expresso (www.ultrapico.com) does use .NET regular expressions, so you might want to give it a try. Thanks for the reference on balancing groups and Friedl’s book. There is no “DEPTH” construct, per se. That is simply the name that Friedl chose for his capture group, any other name works as well. The key concept is “balancing groups”.