I wrote ($nl) = $data =~ m{(\15\12?|\12)} because your usage of \n is still problematic - in this case the newline value for mac, *nix and windows is handled. Anyway, the whole point to this code makes my head hurt - I'm wondering why gmpassos didn't just use one of the existing template engines.

A /better/ idea would be to use this more like a state machine - here's a sample implementation:

my $data = qq`\nHTML1\n<% CODE1 %>\nHTML2\n<% CODE2 %>\nHTML3\n`;

my $reader = get_reader( $data );
while (my $blob = $reader->()) {
    print "$blob->{'type'}: $blob->{'data'}\n";;
}

sub get_reader {
    my $input = shift;
    my $state = 'plain';
    return sub {
        my $temp;
        return unless defined $input;

        if ($state eq 'plain') {
            if ($input =~ s/(.*?)<%//s) {
                $state = 'code';
                return { type => 'plain',
                         data => $1 };
            } else {
                $temp = $input;
                undef $input;
                return { type => 'plain',
                         data => $temp };
            }
        }

        else { # state eq 'code'
            if ($input =~ s/(.*?)\%>//s) {
                $state = 'plain';
                return { type => 'code',
                         data => $1 };
            } else {
                $temp = $input;
                undef $input;
                return { type => 'code',
                         data => $temp };
            }
        }
    }
}
__RETURNS__
plain:
HTML1

code:  CODE1
plain:
HTML2

code:  CODE2
plain:
HTML3
[download]

Seeking Green geeks in Minnesota