This bugged me for a while so I thought I'd mention the solution I found, as well as ask if there is a better way to do it:

Basically I wanted to do a rather complex substitution globally but /g did not work for me. The reason /g did not work is that I need part of the string that has already be used in order to do each match.

I guess a simple example will illustrate the problem better: in the string 'a1b1a2b1b2a3b1b2b3'db1b2 I want to move all the b\d+ following a\d+ to the end of the string and wrap them in (). In this case the simple s{(a\d+)(b\d+)(.*)}{$1$3($2)}g will not work: it will only move the first b, as the regexp engine has to go all the way to the end of the string to match the (.*) part.

There might be ways to do this in one substitution, and I'd like to hear about them, but in fact the real regexps are a bit more complex, so easy tricks are not likely to work, I have searched for them for a while ;--( The main problems are that the (a\d+) is of variable length, which prevents using a look behind (?<=a\d+), each (b\d+) has to be processed separately, and the final (.*) will go through the whole string anyway...

In any case here is the solution I found: use a while with an empty body: while( $string=~ s{(a\d+)(b\d+)(.*)}{$1$3($2)}){}. This works fine, is conceptually clean, but I don't like the empty {}, it looks just weird. Which makes me wonder if there is an other way to do this, one that would be blessed by better hackers than me?

#!/bin/perl -w
use strict;

while( my $string=<DATA>)
  { # this works
    while( $string=~ s{(a\d+)(b\d+)(.*)}{$1$3($2)}){}
    # this does not
    #$string=~ s{(?<=a\d+)(b\d+)(.*)}{$3($2)}g;

    print "string: $string\n"; 
  }


__DATA__
a1b1b2b3
a1b1a12b1b2a3b1b2b3db1b2
[download]