I realised that there are even more problems with the original glob2pat in the Cookbook. The POSIX standard says that the only character inside [] that changes between glob and regular expressions is ! to ^. The Cookbook code does not do that conversion, but does others that are wrong and escapes other special characters, like the ':' in a POSIX characers class.

I gave up trying to use an RE to do this and ended-up with a brute force approach. Elegant it is not, but it does work:

sub glob2pat {
    my $globstr = shift;
    my $inside_br = 0;
    my @chars = (split '', $globstr);
    
    # C style used because I need to skip-ahead and look-behind
    for (my $i; $i < @chars; $i++) {
        if ($chars[$i] eq '\\') {
            $i++;          # ignore next char
        }
        elsif ($chars[$i] eq '[') {
            $inside_br++;  # Allow for nested []
        }
        elsif ($chars[$i] eq ']' && $inside_br) {
            $inside_br--;
        }
        elsif ($chars[$i] eq '!' && $inside_br && $chars[$i-1] eq '[')
+ { 
            # ! only means 'not' at the front of the [] list
            $chars[$i] = '^'
        }
        elsif (! $inside_br) {
            if ($chars[$i] eq '*') {
                $chars[$i] = '.*'
            }
            elsif ($chars[$i] eq '?') {
                $chars[$i] = '.'
            }
        }
    }
    
    local $" = '';
    return "^@chars\$";
}
[download]

Of course improvements are welcome, here is one of my test patterns: '?\?*  [!-0-9?[:upper:]!*?]*'
which gives: '^.\?.*  [^-0-9?[:upper:]!*?].*$'