perlboot - Beginner's Object-Oriented Tutorial

If you're not familiar with objects from other languages, some of the other Perl object documentation may be a little daunting, such as perlobj, a basic reference in using objects, and perltoot, which introduces readers to the peculiarities of Perl's object system in a tutorial way.

So, let's take a different approach, presuming no prior object experience. It helps if you know about subroutines (perlsub), references (perlref et. seq.), and packages (perlmod), so become familiar with those first if you haven't already.

Let's let the animals talk for a moment:

This results in:

Nothing spectacular here. Simple subroutines, albeit from separate packages, and called using the full package name. So let's create an entire pasture:

This results in:

Wow. That symbolic coderef de-referencing there is pretty nasty. We're counting on no strict refs mode, certainly not recommended for larger programs. And why was that necessary? Because the name of the package seems to be inseparable from the name of the subroutine we want to invoke within that package.

Or is it?

For now, let's say that Class->method invokes subroutine method in package Class. (Here, "Class" is used in its "category" meaning, not its "scholastic" meaning.) That's not completely accurate, but we'll do this one step at a time. Now let's use it like so:

And once again, this results in:

That's not fun yet. Same number of characters, all constant, no variables. But yet, the parts are separable now. Watch:

Ahh! Now that the package name has been parted from the subroutine name, we can use a variable package name. And this time, we've got something that works even when use strict refs is enabled.

Let's take that new arrow invocation and put it back in the barnyard example:

There! Now we have the animals all talking, and safely at that, without the use of symbolic coderefs.

But look at all that common code. Each of the speak routines has a similar structure: a print operator and a string that contains common text, except for two of the words. It'd be nice if we could factor out the commonality, in case we decide later to change it all to says instead of goes.

And we actually have a way of doing that without much fuss, but we have to hear a bit more about what the method invocation arrow is actually doing for us.

The invocation of:

attempts to invoke subroutine Class::method as:

(If the subroutine can't be found, "inheritance" kicks in, but we'll get to that later.) This means that we get the class name as the first parameter (the only parameter, if no arguments are given). So we can rewrite the Sheep speaking subroutine as:

And the other two animals come out similarly:

In each case, $class will get the value appropriate for that subroutine. But once again, we have a lot of similar structure. Can we factor that out even further? Yes, by calling another method in the same class.

Let's call out from speak to a helper method called sound. This method provides the constant text for the sound itself.

Now, when we call Cow->speak, we get a $class of Cow in speak. This in turn selects the Cow->sound method, which returns moooo. But how different would this be for the Horse?

Only the name of the package and the specific sound change. So can we somehow share the definition for speak between the Cow and the Horse? Yes, with inheritance!

We'll define a common subroutine package called Animal, with the definition for speak:

Then, for each animal, we say it "inherits" from Animal, along with the animal-specific sound:

Note the added @ISA array (pronounced "is a"). We'll get to that in a minute.

But what happens when we invoke Cow->speak now?

First, Perl constructs the argument list. In this case, it's just Cow. Then Perl looks for Cow::speak. But that's not there, so Perl checks for the inheritance array @Cow::ISA. It's there, and contains the single name Animal.

Perl next checks for speak inside Animal instead, as in Animal::speak. And that's found, so Perl invokes that subroutine with the already frozen argument list.

Inside the Animal::speak subroutine, $class becomes Cow (the first argument). So when we get to the step of invoking $class->sound, it'll be looking for Cow->sound, which gets it on the first try without looking at @ISA. Success!

This magical @ISA variable has declared that Cow "is a" Animal. Note that it's an array, not a simple single value, because on rare occasions, it makes sense to have more than one parent class searched for the missing methods.

If Animal also had an @ISA, then we'd check there too. The search is recursive, depth-first, left-to-right in each @ISA by default (see mro for alternatives). Typically, each @ISA has only one element (multiple elements means multiple inheritance and multiple headaches), so we get a nice tree of inheritance.

When we turn on use strict, we'll get complaints on @ISA, since it's not a variable containing an explicit package name, nor is it a lexical ("my") variable. We can't make it a lexical variable though (it has to belong to the package to be found by the inheritance mechanism), so there's a couple of straightforward ways to handle that.

The easiest is to just spell the package name out:

Or declare it as package global variable:

Or allow it as an implicitly named package variable:

If the Animal class comes from another (object-oriented) module, then just employ use base to specify that Animal should serve as the basis for the Cow class:

Now that's pretty darn simple!

Let's add a mouse, which can barely be heard:

which results in:

Here, Mouse has its own speaking routine, so Mouse->speak doesn't immediately invoke Animal->speak. This is known as "overriding". In fact, we don't even need to say that a Mouse is an Animal at all, because all of the methods needed for speak are completely defined for Mouse; this is known as "duck typing": "If it walks like a duck and quacks like a duck, I would call it a duck" (James Whitcomb). However, it would probably be beneficial to allow a closer examination to conclude that a Mouse is indeed an Animal, so it is actually better to define Mouse with Animal as its base (that is, it is better to "derive Mouse from Animal").

Moreover, this duplication of code could become a maintenance headache (though code-reuse is not actually a good reason for inheritance; good design practices dictate that a derived class should be usable wherever its base class is usable, which might not be the outcome if code-reuse is the sole criterion for inheritance. Just remember that a Mouse should always act like an Animal).

So, let's make Mouse an Animal!

The obvious solution is to invoke Animal::speak directly:

Note that we're using Animal::speak. If we were to invoke Animal->speak instead, the first parameter to Animal::speak would automatically be "Animal" rather than "Mouse", so that the call to $class->sound in Animal::speak would become Animal->sound rather than Mouse->sound.

Also, without the method arrow ->, it becomes necessary to specify the first parameter to Animal::speak ourselves, which is why $class is explicitly passed: Animal::speak($class).

However, invoking Animal::speak directly is a mess: Firstly, it assumes that the speak method is a member of the Animal class; what if Animal actually inherits speak from its own base? Because we are no longer using -> to access speak, the special method look up mechanism wouldn't be used, so speak wouldn't even be found!

The second problem is more subtle: Animal is now hardwired into the subroutine selection. Let's assume that Animal::speak does exist. What happens when, at a later time, someone expands the class hierarchy by having Mouse inherit from Mus instead of Animal. Unless the invocation of Animal::speak is also changed to an invocation of Mus::speak, centuries worth of taxonomical classification could be obliterated!

What we have here is a fragile or leaky abstraction; it is the beginning of a maintenance nightmare. What we need is the ability to search for the right method wih as few assumptions as possible.

A better solution is to tell Perl where in the inheritance chain to begin searching for speak. This can be achieved with a modified version of the method arrow ->:

So, the improved Mouse class is:

Using this syntax, we start with Animal to find speak, and then use all of Animal's inheritance chain if it is not found immediately. As usual, the first parameter to speak would be $class, so we no longer need to pass $class explicitly to speak.

But what about the second problem? We're still hardwiring Animal into the method lookup.

If Animal is replaced with the special placeholder SUPER in that invocation, then the contents of Mouse's @ISA are used for the search, beginning with $ISA[0]. So, all of the problems can be fixed as follows:

In general, SUPER::speak means look in the current package's @ISA for a class that implements speak, and invoke the first one found. The placeholder is called SUPER, because many other languages refer to base classes as "superclasses", and Perl likes to be eclectic.

Note that a call such as

does not look in the @ISA of $class unless $class happens to be the current package.

So far, we've seen the method arrow syntax:

or the equivalent:

which constructs an argument list of:

and attempts to invoke:

However, if Class::method is not found, then @Class::ISA is examined (recursively) to locate a class (a package) that does indeed contain method, and that subroutine is invoked instead.

Using this simple syntax, we have class methods, (multiple) inheritance, overriding, and extending. Using just what we've seen so far, we've been able to factor out common code (though that's never a good reason for inheritance!), and provide a nice way to reuse implementations with variations.

Now, what about data?

Let's start with the code for the Animal class and the Horse class:

This lets us invoke Horse->speak to ripple upward to Animal::speak, calling back to Horse::sound to get the specific sound, and the output of:

But all of our Horse objects would have to be absolutely identical. If we add a subroutine, all horses automatically share it. That's great for making horses the same, but how do we capture the distinctions of an individual horse? For example, suppose we want to give our first horse a name. There's got to be a way to keep its name separate from the other horses.

That is to say, we want particular instances of Horse to have different names.

In Perl, any reference can be an "instance", so let's start with the simplest reference that can hold a horse's name: a scalar reference.

So, now $horse is a reference to what will be the instance-specific data (the name). The final step is to turn this reference into a real instance of a Horse by using the special operator bless:

This operator stores information about the package named Horse into the thing pointed at by the reference. At this point, we say $horse is an instance of Horse. That is, it's a specific horse. The reference is otherwise unchanged, and can still be used with traditional dereferencing operators.

The method arrow can be used on instances, as well as classes (the names of packages). So, let's get the sound that $horse makes:

To invoke sound, Perl first notes that $horse is a blessed reference (and thus an instance). It then constructs an argument list, as per usual.

Now for the fun part: Perl takes the class in which the instance was blessed, in this case Horse, and uses that class to locate the subroutine. In this case, Horse::sound is found directly (without using inheritance). In the end, it is as though our initial line were written as follows:

Note that the first parameter here is still the instance, not the name of the class as before. We'll get neigh as the return value, and that'll end up as the $noise variable above.

If Horse::sound had not been found, we'd be wandering up the @Horse::ISA array, trying to find the method in one of the superclasses. The only difference between a class method and an instance method is whether the first parameter is an instance (a blessed reference) or a class name (a string).

Because we get the instance as the first parameter, we can now access the instance-specific data. In this case, let's add a way to get at the name:

Inside Horse::name, the @_ array contains:

so the shift stores $horse into $self. Then, $self gets de-referenced with $$self as normal, yielding "Mr. Ed".

It's traditional to shift the first parameter into a variable named $self for instance methods and into a variable named $class for class methods.

Then, the following line:

outputs:

Of course, if we constructed all of our horses by hand, we'd most likely make mistakes from time to time. We're also violating one of the properties of object-oriented programming, in that the "inside guts" of a Horse are visible. That's good if you're a veterinarian, but not if you just like to own horses. So, let's have the Horse class handle the details inside a class method:

Now with the new named method, we can build a horse as follows:

Notice we're back to a class method, so the two arguments to Horse::named are Horse and Mr. Ed. The bless operator not only blesses \$name, it also returns that reference.

This Horse::named method is called a "constructor".

We've called the constructor named here, so that it quickly denotes the constructor's argument as the name for this particular Horse. You can use different constructors with different names for different ways of "giving birth" to the object (like maybe recording its pedigree or date of birth). However, you'll find that most people coming to Perl from more limited languages use a single constructor named new, with various ways of interpreting the arguments to new. Either style is fine, as long as you document your particular way of giving birth to an object. (And you were going to do that, right?)

But was there anything specific to Horse in that method? No. Therefore, it's also the same recipe for building anything else that inherited from Animal, so let's put name and named there:

Ahh, but what happens if we invoke speak on an instance?

We get a debugging value:

Why? Because the Animal::speak routine is expecting a classname as its first parameter, not an instance. When the instance is passed in, we'll end up using a blessed scalar reference as a string, and that shows up as we saw it just now.

All we need is for a method to detect if it is being called on a class or called on an instance. The most straightforward way is with the ref operator. This returns a string (the classname) when used on a blessed reference, and an empty string when used on a string (like a classname). Let's modify the name method first to notice the change:

Here, the ?: operator comes in handy to select either the dereference or a derived string. Now we can use this with either an instance or a class. Note that I've changed the first parameter holder to $either to show that this is intended:

and now we'll fix speak to use this:

And since sound already worked with either a class or an instance, we're done!

Let's train our animals to eat:

And now try it out:

which prints:

An instance method with parameters gets invoked with the instance, and then the list of parameters. So that first invocation is like:

What if an instance needs more data? Most interesting instances are made of many items, each of which can in turn be a reference or even another object. The easiest way to store these is often in a hash. The keys of the hash serve as the names of parts of the object (often called "instance variables" or "member variables"), and the corresponding values are, well, the values.

But how do we turn the horse into a hash? Recall that an object was any blessed reference. We can just as easily make it a blessed hash reference as a blessed scalar reference, as long as everything that looks at the reference is changed accordingly.

Let's make a sheep that has a name and a color:

so $bad->{Name} has Evil, and $bad->{Color} has black. But we want to make $bad->name access the name, and that's now messed up because it's expecting a scalar reference. Not to worry, because that's pretty easy to fix up.

One solution is to override Animal::name and Animal::named by defining them anew in Sheep, but then any methods added later to Animal might still mess up, and we'd have to override all of those too. Therefore, it's never a good idea to define the data layout in a way that's different from the data layout of the base classes. In fact, it's a good idea to use blessed hash references in all cases. Also, this is why it's important to have constructors do the low-level work. So, let's redefine Animal:

Of course, we still need to override named in order to handle constructing a Sheep with a certain color:

(Note that @_ contains the parameters to named.)

What's this default_color? Well, if named has only the name, we still need to set a color, so we'll have a class-specific default color. For a sheep, we might define it as white:

Now:

outputs:

Now, there's nothing particularly specific to Sheep when it comes to color, so let's remove Sheep::named and implement Animal::named to handle color instead:

And then to keep from having to define default_color for each additional class, we'll define a method that serves as the "default default" directly in Animal:

Of course, because name and named were the only methods that referenced the "structure" of the object, the rest of the methods can remain the same, so speak still works as before.

But having all our horses be brown would be boring. So let's add a method or two to get and set the color.

Note the alternate way of accessing the arguments: $_[0] is used in-place, rather than with a shift. (This saves us a bit of time for something that may be invoked frequently.) And now we can fix that color for Mr. Ed:

which results in:

So, now we have class methods, constructors, instance methods, instance data, and even accessors. But that's still just the beginning of what Perl has to offer. We haven't even begun to talk about accessors that double as getters and setters, destructors, indirect object notation, overloading, "isa" and "can" tests, the UNIVERSAL class, and so on. That's for the rest of the Perl documentation to cover. Hopefully, this gets you started, though.

For more information, see perlobj (for all the gritty details about Perl objects, now that you've seen the basics), perltoot (the tutorial for those who already know objects), perltooc (dealing with class data), perlbot (for some more tricks), and books such as Damian Conway's excellent Object Oriented Perl.

Some modules which might prove interesting are Class::Accessor, Class::Class, Class::Contract, Class::Data::Inheritable, Class::MethodMaker and Tie::SecureHash