In essence, all you need is a custom string comparison routine. And the easiest, most efficient way to do that in Perl is to use the tr/// operator to 'encode' the strings and then use the standard cmp operator.

Say your custom sort rules call for 0-4 to be sorted before alphas, and 5-9 after. And within the alphas, you want upper and lower case of any given character to be sorted together. Then you set up a mapping that maps the original strings to characters that will sort in the required order.

For the given example:

tr[0-4AaBbCcDdEeFfGgHhIiJjKkLlMmNnOoPpQqRrSsTtUuVvWwXxYyZz5-9][\x00-\x
+ff];
[download]

#! perl -slw
use strict;

sub trans {
    my $in = shift;
    $in =~ tr[0-4AaBbCcDdEeFfGgHhIiJjKkLlMmNnOoPpQqRrSsTtUuVvWwXxYyZz5
+-9]
             [\x00-\xff];
    return $in;
}

chomp( my @data = <DATA> );

my @sorted = sort{
    trans( $a ) cmp trans( $b )
} @data;

print for @sorted;

__DATA__
cdef
0123456
abcd
50011
ABCD
4999
Zxyw
CDEF
zxyw
9999
[download]

Produces:

c:\test>junk78
0123456
4999
ABCD
abcd
CDEF
cdef
Zxyw
zxyw
50011
9999
[download]

Of course, you can now apply all the usual forms of sort optimisations--ST, GRT etc.--to that, but the mechanism remains the same.

The ugly head of UTF will probably complicate things a little, but, at least in the eyes of my non-UTF aware brain, it should be possible to apply the same mechanism.