Why is this necessary? Well, if you think about it, sorting is putting a bunch of things in order by comparing them all. Since you can't compare them all at once, you need to compare two at a time, eventually using what you find out about each pair's order to put the whole kit'n'caboodle in line. Perl already understands all of those steps except for the part about how you'd like to compare the items, so that's all you have to write.

This means that the sort subroutine doesn't need to sort many items after all. It merely has to be able to compare two items. If it can put two items in the proper order, Perl will be able to tell (by repeatedly consulting the sort subroutine) what order you want for your data.

The sort subroutine is defined like an ordinary subroutine (well, almost). This routine will be called repeatedly, each time checking on a pair of elements from the list to be sorted.

Now, if you were writing a subroutine that's expecting to get two parameters that need sorting, you might write something like this to start:

sub any_sort_sub {  # It doesn't really work this way
  my($a, $b) = @_;  # Get and name the two parameters
  # start comparing $a and $b here
  ...
}

But the sort subroutine will be called again and again, often hundreds or thousands of times. Declaring the variables $a and $b and assigning them values at the top of the subroutine will take just a little time, but multiply that by the thousands of times that the routine will be called, and you can see that it contributes significantly to the overall execution speed.

The subroutine returns a coded value describing how the elements compare (like C's qsort(3) does, but it's Perl's own internal sort implementation). If $a should appear before $b in the final list, the sort subroutine returns -1 to say so. If $b should appear before $a, it returns 1.

If the order of $a and $b doesn't matter, the subroutine returns 0. Why would it not matter? Perhaps you're doing a case-insensitive sort and the two strings are fred and Fred. Or perhaps you're doing a numeric sort, and the two numbers are equal.

We could now write a numeric sort subroutine like this:

sub by_number {
  # a sort subroutine, expect $a and $b
  if ($a < $b) { -1 } elsif ($a > $b) { 1 } else { 0 }
}

To use the sort subroutine, just put its name (without an ampersand) between the keyword sort and the list to be sorted. This example puts a numerically sorted list of numbers into @result:

my @result = sort by_number @some_numbers;

We called the subroutine by_number because that describes how it's sorting. But more importantly, you can read the line of code that uses it with sort as saying "sort by number," as you would in English. Many sort-subroutine names begin with by_ to describe how they sort. Or we could have called this one numerically, for a similar reason, but that's more typing and more chance to mess something up.

Notice that we don't have to do anything in the sort subroutine to declare $a and $b, and to set their values -- and if we did, the subroutine wouldn't work right. We just let Perl set up $a and $b for us, and so all we need to write is the comparison.

sub by_number { $a <=> $b }

sub ASCIIbetically { $a cmp $b }
my @strings = sort ASCIIbetically @any_strings;

But you can use cmp to build a more complex sort order, like a case-insensitive sort:

sub case_insensitive { "\L$a" cmp "\L$b" }

In this case, we're comparing the string from $a (forced to lowercase) against the string from $b (forced to lowercase), giving a case-insensitive sort order.

Note that we're not modifying the elements themselves; we're merely using their values. That's actually important: for efficiency reasons, $a and $b aren't copies of the data items. They're actually new, temporary aliases for elements of the original list, so if we changed them we'd be mangling the original data. Don't do that -- it's neither supported nor recommended.

When your sort subroutine is as simple as the ones we show here (and most of the time, it is), you can make the code even simpler yet, by replacing the name of the sort routine with the entire sort routine "in line," like so:

my @numbers = sort { $a <=> $b } @some_numbers;

In fact, in modern Perl, you'll hardly ever see a separate sort subroutine; you'll frequently find sort routines written inline as we've done here.

my @descending = reverse sort { $a <=> $b } @some_numbers;

But here's a neat trick. The comparison operators (<=> and cmp) are very nearsighted; that is, they can't see which operand is $a and which is $b, but only which value is on the left and which is on the right. So if $a and $b were to swap places, the comparison operator would get the results backwards every time. That means that this is another way to get a reversed numeric sort:

my @descending = sort { $b <=> $a } @some_numbers;

You can (with a little practice) read this at a glance. It's a descending-order comparison (because $b comes before $a, which is descending order), and it's a numeric comparison (because it uses the spaceship instead of cmp). So, it's sorting numbers in reverse order.