Object Construction (Programming Perl)

12.4. Object Construction

All objects are references, but not all references are objects. A reference won't work as an object unless its referent is specially marked to tell Perl what package it belongs to. The act of marking a referent with a package name--and therefore, its class, since a class is just a package--is known as blessing. You can think of the blessing as turning a reference into an object, although it's more accurate to say that it turns the reference into an object reference.

The bless function takes either one or two arguments. The first argument is a reference and the second is the package to bless the referent into. If the second argument is omitted, the current package is used.

$obj = { };                 # Get reference to anonymous hash.
bless($obj);                # Bless hash into current package.
bless($obj, "Critter");     # Bless hash into class Critter.

Here we've used a reference to an anonymous hash, which is what people usually use as the data structure for their objects. Hashes are extremely flexible, after all. But allow us to emphasize that you can bless a reference to anything you can make a reference to in Perl, including scalars, arrays, subroutines, and typeglobs. You can even bless a reference to a package's symbol table hash if you can think of a good reason to. (Or even if you can't.) Object orientation in Perl is completely orthogonal to data structure.

Once the referent has been blessed, calling the built-in ref function on its reference returns the name of the blessed class instead of the built-in type, such as HASH. If you want the built-in type, use the reftype function from the attributes module. See use attributes in Chapter 31, "Pragmatic Modules".

And that's how to make an object. Just take a reference to something, give it a class by blessing it into a package, and you're done. That's all there is to it if you're designing a minimal class. If you're using a class, there's even less to it, because the author of a class will have hidden the bless inside a method called a constructor, which creates and returns instances of the class. Because bless returns its first argument, a typical constructor can be as simple as this:

package Critter;
sub spawn { bless {}; }

Or, spelled out slightly more explicitly:

package Critter;
sub spawn {
    my     $self = {};       # Reference to an empty anonymous hash
    bless  $self, "Critter"; # Make that hash a Critter object
    return $self;            # Return the freshly generated Critter
}

With that definition in hand, here's how one might create a Critter object:

$pet = Critter->spawn;

12.4.1. Inheritable Constructors

Like all methods, a constructor is just a subroutine, but we don't call it as a subroutine. We always invoke it as a method--a class method, in this particular case, because the invocant is a package name. Method invocations differ from regular subroutine calls in two ways. First, they get the extra argument we discussed earlier. Second, they obey inheritance, allowing one class to use another's methods.

We'll describe the underlying mechanics of inheritance more rigorously in the next section, but for now, some simple examples of its effects should help you design your constructors. For instance, suppose we have a Spider class that inherits methods from the Critter class. In particular, suppose the Spider class doesn't have its own spawn method. The following correspondences apply:

Method Call	Resulting Subroutine Call
`Critter->spawn()`	`Critter::spawn("Critter")`
`Spider->spawn()`	`Critter::spawn("Spider")`

The subroutine called is the same in both cases, but the argument differs. Note that our spawn constructor above completely ignored its argument, which means our Spider object was incorrectly blessed into class Critter. A better constructor would provide the package name (passed in as the first argument) to bless:

sub spawn {
    my $class =  shift;       # Store the package name
    my $self  =  { };
    bless($self, $class);     # Bless the reference into that package
    return $self;
}

Now you could use the same subroutine for both these cases:

$vermin = Critter->spawn;
$shelob = Spider->spawn;

And each object would be of the proper class. This even works indirectly, as in:

$type  = "Spider";
$shelob = $type->spawn;         # same as "Spider"->spawn

That's still a class method, not an instance method, because its invocant holds a string and not a reference.

If $type were an object instead of a class name, the previous constructor definition wouldn't have worked, because bless needs a class name. But for many classes, it makes sense to use an existing object as the template from which to create another. In these cases, you can design your constructors so that they work with either objects or class names:

sub spawn {
    my $invocant = shift;
    my $class    = ref($invocant) || $invocant;  # Object or class name
    my $self     = { };
    bless($self, $class);
    return $self;
}

12.4.2. Initializers

Most objects maintain internal information that is indirectly manipulated by the object's methods. All our constructors so far have created empty hashes, but there's no reason to leave them empty. For instance, we could have the constructor accept extra arguments to store into the hash as key/value pairs. The OO literature often refers to such data as properties, attributes, accessors, member data, instance data, or instance variables. The section "Instance Variables" later in this chapter discusses attributes in more detail.

Imagine a Horse class with instance attributes like "name" and "color":

$steed = Horse->new(name => "Shadowfax", color => "white");

If the object is implemented as a hash reference, the key/value pairs can be interpolated directly into the hash once the invocant is removed from the argument list:

sub new {
    my $invocant = shift;
    my $class = ref($invocant) || $invocant;
    my $self = { @_ };          # Remaining args become attributes
    bless($self, $class);       # Bestow objecthood
    return $self;
}

This time we used a method named new for the class's constructor, which just might lull C++ programmers into thinking they know what's going on. But Perl doesn't consider "new" to be anything special; you may name your constructors whatever you like. Any method that happens to create and return an object is a de facto constructor. In general, we recommend that you name your constructors whatever makes sense in the context of the problem you're solving. For example, constructors in the Tk module are named after the widgets they create. In the DBI module, a constructor named connect returns a database handle object, and another constructor named prepare is invoked as an instance method and returns a statement handle object. But if there is no suitable context-specific constructor name, new is perhaps not a terrible choice. Then again, maybe it's not such a bad thing to pick a random name to force people to read the interface contract (meaning the class documentation) before they use its constructors.

Elaborating further, you can set up your constructor with default key/value pairs, which the user could later override by supplying them as arguments:

sub new {
    my $invocant = shift;
    my $class   = ref($invocant) || $invocant;
    my $self = {
        color  => "bay",
        legs   => 4,
        owner  => undef,
        @_,                 # Override previous attributes
    };
    return bless $self, $class;
}

$ed       = Horse->new;                    # A 4-legged bay horse
$stallion = Horse->new(color => "black");  # A 4-legged black horse

This Horse constructor ignores its invocant's existing attributes when used as an instance method. You could create a second constructor designed to be called as an instance method, and if designed properly, you could use the values from the invoking object as defaults for the new one:

$steed  = Horse->new(color => "dun");
$foal   = $steed->clone(owner => "EquuGen Guild, Ltd.");

sub clone {
    my $model = shift;
    my $self  = $model->new(%$model, @_);
    return $self;     # Previously blessed by ->new
}

(You could also have rolled this functionality directly into new, but then the name wouldn't quite fit the function.)

Notice how even in the clone constructor, we don't hardcode the name of the Horse class. We have the original object invoke its own new method, whatever that may be. If we had written that as Horse->new instead of $model->new, then the class wouldn't have facilitated inheritance by a Zebra or Unicorn class. You wouldn't want to clone Pegasus and suddenly find yourself with a horse of a different color.

Sometimes, however, you have the opposite problem: rather than trying to share one constructor among several classes, you're trying to have several constructors share one class's object. This happens whenever a constructor wants to call a base class's constructor to do part of the construction work. Perl doesn't do hierarchical construction for you. That is, Perl does not automatically call the constructors (or the destructors) for any base classes of the class requested, so your constructor will have to do that itself and then add any additional attributes the derived class needs. So the situation is not unlike the clone routine, except that instead of copying an existing object into the new object, you want to call your base class's constructor and then transmogrify the new base object into your new derived object.


12.3. Method Invocation		12.5. Class Inheritance