This seems to come up pretty regularly. The best answer seems to be Text::Balanced, which I recently bookmarked on my home node due to it's awesomeness. Sadly, I can't remember who pointed it out to me.

With difficulty.

our $re;
local $re = qr{
   \[
   (?:
      (?> [^\[\]]+ )
   |
      (??{ $re })
   )*
   \]
}x;

my $s = <<'__EOI__';
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah
__EOI__

$s =~ s/$re/moo/g;

print($s);
[download]

Partial credits to perlre.

that's beautiful... so, if i have this right... match the first bracket, IF, non-backtracking, we find an additional or , then recurse, match the last bracket? what is the purpose of the ^?

sorry, i meant to say...

that's beautiful...

so, if i have this right... match the first bracket; IF, non-backtracking, we find an additional [ or ], then recurse; match the last bracket?

what is the purpose of the ^?

Surely this will do it:

$str = '
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah';

$str =~ m/\[(.*)\]/s;
print $1;
[download]

Because the regex finds the leftmost pattern, /s allows dot to match linebreaks and dot-star is greedy, that's all you need. Am I missing something?

good idea, but this won't work because there may be multiple pairs of outermost (i.e., same level) brackets. so, a greedy regular expression will grab beyond. i should have clarified this. for example:

$str = '
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah
blah blah blah blah
blah
blah blah blah [blah
[blah blah] [blah blah blah
blah] blah]
blah blah';
[download]

in other words, there are multiple top-level bracket pairs, which, may or may not, contain additional pairs (etc).

btw, here is the code snippet that i finally used (which has some nuances that i didn't include in the original question), based on the previous suggestion:

        my $re = qr{\[(?:(?>[^\[\]]+)|(??{$re}))*\]}s;
        for (;;) {
                last unless $tempstr =~ s/(\[\w+?\s*=\s*($re|\n|[^\[\]
+])+\])/&assign($1)/gies;
        }
[download]

That code is broken.

• You can't declare $re on the same line as you're using it. It won't even compile (under strict vars).

• There's a I reason I used a package variable instead of a lexical. It'll bite you if you interpolate into $re. Best to always use a package variable to avoid the problem entirely.

• While not a bug per say, the s modifiers on both regeps are useless because you don't use ..

        our $re
        local $re = qr{\[(?:(?>[^\[\]]+)|(??{$re}))*\]};
        for (;;) {
                last unless $tempstr =~ s/(\[\w+?\s*=\s*($re|\n|[^\[\]
+])+\])/assign($1)/gie;
        }
[download]

It's unfortunate that you modify things without knowing why they were done in the first place.

I should point out that ikegami's suggestion by itself would surely have worked, but I couldn't get it to for the nuances that I needed, and so I added some possibly unnecessary cruft to account for my needs. I.e., I don't really know why |\n|[^\[\]+])+\] was necessary, but after many hours of trial and error, it was that which made everything (appear to) work.