it is more or less equivalent to:

sub keysort (&@) {
  my ($keygen, @values) = @_;
  my @keys = map { &$keygen() } @values;
  my @ix = sort { $keys[$a] cmp $keys[$b] } 0..$#keys;
  @values[@ix];
}
[download]

but implemented in C so there is no performance penalty because of the sort whith a custom comparison body (the heavy part of the alghorithm, with O(NlogN) cost).

It also uses native C types to represent the keys when those are numbers, and tries to move or copy data around as less as possible.