I include the following in my blog post template:

If you have any questions or comments, I would be glad to hear it.

I ask for GitHub issues or Twitter responses, because I’ve had long-form off-topic arguments in the comments of my posts before, and I wanted to push those things away from my core points while still having interaction with readers.

But that seemed to be a barrier, because few people used these methods to contact me, until last night, when I was asked to go deeper on postderef.


Consider $n[0][0]. By syntax, we know that there’s an array @n, but what is in $n[0]?

0> my @n;
1> $n[0][0] = 3;
$res[0] = 3

2> ref $n[0][0]
$res[1] = ''

3> ref $n[0]
$res[2] = 'ARRAY'

The “Big Book of Perl References” is perlref and should be available via perldoc perlref. The base syntax,from this documentation, is thus:

    $scalarref = \$foo;
    $arrayref  = \@ARGV;
    $hashref   = \%ENV;
    $coderef   = \&handler;
    $globref   = \*foo;

But you can also work more directly.

    $arrayref = [1,2,3];
    $hashref  = { foo => 'bar' };
    $coderef  = sub { print 'FMEP' };

So, yes, you can create an array of functions just like you can create an array of arrays. This is called a dispatch table and is key to Data-Driven Development, and a whole lot of fun.

Standard Dereferencing

Going from the previously-set references, we’d get

    $$scalarref ;
    @$arrayref ;

This is fine. This works. But it does double-up on the sigils in front, and can make reading a bear.

Aside: Why I doubled-down on references

JSON and related modules.

In a way, I had come to default to passing values as hashrefs because remembering if it the function took ($filehandle, $value) or ($value, $filehandle) was frustrating when you could pass ({ filehandle => $filehandle, value=>$value }) and know there’s no ambiguity. This is cool, wonderful and very kind to the maintenence developer who will look at your code in a few years.

But no, it was JSON that pushed me over the edge, because …

17> my $json = JSON->new->pretty->canonical;
$res[12] = bless( do{\(my $o = '')}, 'JSON' )

18> my @x = 1..5

19> $json->encode(@x)
Usage: JSON::XS::encode(self, scalar) at reply input line 1.

20> $json->encode(\@x)
$res[14] = '[

JSON wants references to data structures, so instead of @n and $n[0][0], it’s easier to have @$n and $$n[0][0], except sometimes it gets a bit hard to read. You can do @{$n}, but I wonder if that’s an improvement.


And the Secret Masters of Perl must’ve thought the same thing, because we have postderef, which has been stable since 5.24. So, let’s round up the usual suspects.

    $arrayref->@* ;

Here we’re dereferencing as above, but post or after the reference. And, of course, these can stack.

    my $value = $data->{functions}[0](20);
    # $data is a hashref
    # $data->{functions} is an arrayref
    # $data->{functions}[0] is a coderef

(I’m reasonably sure it was brian d. foy who taught me this, but it could’ve been David Farrell. Props to both on this.)

Aside: The mind-boggling part

A line from my post on Javascript Arrays, actually.

my $hedgehogs->@* = map { int rand 10 } 0..9;

This does not need to be an arrayref, but for the above reasons, I am more likely to make references than hashes and arrays. If I want to dump as JSON to see what I’m dealing with at a random point, it’s just easier for me.

So, this could have been written this way, without any references at all.

my @hedgehogs = map { int rand 10 } 0..9;

I guess I was using the map-grep style of Perl to show off that style of Javascript, assuming that Perl developers had gone through the Schwartzian Transform and understood it.

Basically, we’re talking about functions that take an array and return an array. .. is a range operator, and 0..9 gives you the array [0,1,2,3,4,5,6,7,8,9].

map transforms the array, as in “maps a value onto”. (The ex-English major wonders about the etymology of “map”.) This is in contrast with sort, which changes the order of the array, and filter or grep, which make shorter arrays. In Perl, rand gives you a “random” floating-point number between 0 and 1. rand 10 multiplies that by 10, so, it’s between 0.0… and 9.9…, and int rand 10 coerces it into an integer, so now we get 10 random integers instead of 0 thru 9.

But, as map takes a function as well as an array, the changes could be anything. map { sub ( $x ) { return $_ * $x } } 0..9 would make an array of functions that, basically, multiply the value by the index number.

I go into the pieces of the Transform in JS in a previous blog post.

Back at the ranch

I can think of one frustration and one trick relating to using references and the postderef feature. The frustration might be obvious when you read this: the syntax highlighters for Perl tend to lag behind Perl, so the cool new features that I love are usually not added yet, which means that they don’t know what to do with my code. The one in Markdown/Jekyll/Github Pages is certainly not there yet.

And here’s the trick: Perl knows the difference between my $x = response() and my @x = response(), and you can have your function switch between by using wantarray.

sub response {
    my @array = 0..9;
    return wantarray ? @array : \@array;

sub alt_response {
    my $array->@* = 0..9;
    return wantarray ? $array->@* : $array;

If you have any questions or comments, I would be glad to hear it. Ask me on Twitter or make an issue on my blog repo.