Question

A little regex help please.

Why are these different?

Regex.Replace("[phone] /x ext123", "/x.*|[^0-9]", String.Empty)
"9995550000"


Regex.Replace("[phone] /x ext123", "[^0-9]|/x.*", String.Empty)
"9995550000123"

I thought the pipe operator did not care about order… or maybe there is something else that can explain this?

Answer 1

I think you’ve got the wrong idea about alternation (i.e., the pipe). In a pure DFA regex implementation, it’s true that alternation favors the longest match no matter how the alternatives are ordered. In other words, the whole regex, whether it contains alternation or not, always returns the earliest and longest possible match–the “leftmost-longest” rule.

However, the regex implementations in most of today’s popular programming languages, including .NET, are what Friedl calls Traditional NFA engines. One of the most important differences between them and DFA engines is that alternation is not greedy; it attempts the alternatives in the order they’re listed and stops as soon as one of them matches. The only thing that will cause it to change its mind is if the match fails at a later point in the regex, forcing it to backtrack into the alternation.

Note that if you change the [^0-9] to [^0-9]+ in both regexes you’ll get the same result from both–but not the one you want. (I’m assuming the /x.* alternative is supposed to match–and remove–the rest of the string, including the extension number.) I’d suggest something like this:

"[^0-9/]+|/x.*$"

That way, neither alternative can even start to match what the other one matches. Not only will that will prevent the kind of confusion you’re experiencing, it avoids potential performance bottlenecks. One of the other major differences between DFA’s and NFA’s is that badly-written NFA’s are prone to serious (even catastrophic) performance problems, and sloppy alternations are one of the easiest ways to trigger them.