my @data;
foreach ( 1 .. 100 ) {
    push @data, [ map { rand(200) - 100 } 1 .. 100 ];
}

my $pivot = 0;
my @sorted = keysort {
    my $v = $_;
    my @v = keysort { $_ > $pivot ? 1 : -1 } @{$v};
    $v[50]
} @data;
[download]

well, what are you trying to do?

The inner keysort on your code is equivalent to:

@v=((grep { $_ <= $pivot } @$v), (grep { $_ > $pivot } @$v));
[download]

is that what you intended?

I suppose that what you really want is to mimic the sort on the original post but numerically: sort first the elements > $pivot and then the rest.

Then, for your specific data, integers between -100 and 100, this works:

my @v=nkeysort { $_<$pivot ? $_ + 200 : $_ } @$v;
[download]

but if you know nothing about the numbers in $@v, then figuring a convenient sorting key can be quite difficult, it's easier to just use a grep/sort combination:

my @v=((sort {$a<=>$b} (grep { $_ >= $pivot } @$v)),
       (sort {$a<=>$b} (grep { $_ < $pivot } @$v)));
[download]

that inserted on the outer sort becomes:

my @sorted = nkeysort {
  ( ( sort {$a<=>$b} (grep { $_ >= $pivot } @$_) ),
    ( sort {$a<=>$b} (grep { $_ <  $pivot } @$_) ) )[50]
} @data;
[download]

BTW, note that for numeric keys you have to use nkeysort.

I'm trying to see if you know what should happen if you have a keysort within a keysort. You built your module based on the &@ prototype. You don't have a test for a keysort in a keysort, to make sure you don't trip anything, especially on earlier implementations of prototypes (like 5.6.0). You should have a test for that. :-)

---

My wife's blog