From: Xavier Noria <fxn@...>
Date: 2006-03-10T20:35:18+09:00
Subject: Re: Small regexp question

On Mar 10, 2006, at 11:58, francisrammeloo@hotmail.com wrote:

> The pattern I used to find a class definition line is:
>
>     line =~ /^\s*class\s+(\w+)/
>
> But I want to exclude forward class declarations ( class MyClass; )
>
> So I changed my pattern to:
>
>     line =~ /^\s*class\s+(\w+)\s*[^;]/    --> don't match if line ends
> with ";"
>
> But it doesn't work... Why?

I don't know exactly in what sense it does not work, but negations in  
regexps are tricky.

A regexp engine *always* tries to match. If in a first attempt \w+  
matches the whole class name and then the rest does not match, then  
the regexp engine backtracks and happens to find a "shorter class  
name" whose remaining characters are not semicolons, so it still  
matches.

    class Foo; (\w+ -> "Foo", fails, backtrack)
             ^
    class Foo; (\w+ -> "Fo", no whitespace, "o" is not a semicolon,  
matched)
            ^

A solution is to add an anchor for end of string. Another one is to  
prevent \w+ from backtracking, that is known as "atomic grouping":

    (?>\w+) # grab word characters and do not backtrack

In addition, the idiomatic way to say "and at this point I don't what  
this to happen" is to use a negative look-ahead assertion. All in all  
we get this:

    /^\s*class\s+(?>\w+)(?!\s*;)/

-- fxn