PERLOOTUT(phpman.html) - phpMan

PERLOOTUT(1)           Perl Programmers Reference Guide           PERLOOTUT(1)

NAME
       perlootut - Object-Oriented Programming in Perl Tutorial
DATE
       This document was created in February, 2011.
DESCRIPTION
       This document provides an introduction to object-oriented programming
       in Perl. It begins with a brief overview of the concepts behind object
       oriented design. Then it introduces several different OO systems from
       CPAN <http://search.cpan.org>; which build on top of what Perl provides.
       By default, Perl's built-in OO system is very minimal, leaving you to
       do most of the work. This minimalism made a lot of sense in 1994, but
       in the years since Perl 5.0 we've seen a number of common patterns
       emerge in Perl OO. Fortunately, Perl's flexibility has allowed a rich
       ecosystem of Perl OO systems to flourish.
       If you want to know how Perl OO works under the hood, the perlobj
       document explains the nitty gritty details.
       This document assumes that you already understand the basics of Perl
       syntax, variable types, operators, and subroutine calls. If you don't
       understand these concepts yet, please read perlintro first. You should
       also read the perlsyn, perlop, and perlsub documents.
OBJECT-ORIENTED FUNDAMENTALS
       Most object systems share a number of common concepts. You've probably
       heard terms like "class", "object, "method", and "attribute" before.
       Understanding the concepts will make it much easier to read and write
       object-oriented code. If you're already familiar with these terms, you
       should still skim this section, since it explains each concept in terms
       of Perl's OO implementation.
       Perl's OO system is class-based. Class-based OO is fairly common. It's
       used by Java, C++, C#, Python, Ruby, and many other languages. There
       are other object orientation paradigms as well. JavaScript is the most
       popular language to use another paradigm. JavaScript's OO system is
       prototype-based.
   Object
       An object is a data structure that bundles together data and
       subroutines which operate on that data. An object's data is called
       attributes, and its subroutines are called methods. An object can be
       thought of as a noun (a person, a web service, a computer).
       An object represents a single discrete thing. For example, an object
       might represent a file. The attributes for a file object might include
       its path, content, and last modification time. If we created an object
       to represent /etc/hostname on a machine named "foo.example.com", that
       object's path would be "/etc/hostname", its content would be "foo\n",
       and it's last modification time would be 1304974868 seconds since the
       beginning of the epoch.
       The methods associated with a file might include "rename()" and
       "write()".
       In Perl most objects are hashes, but the OO systems we recommend keep
       you from having to worry about this. In practice, it's best to consider
       an object's internal data structure opaque.
   Class
       A class defines the behavior of a category of objects. A class is a
       name for a category (like "File"), and a class also defines the
       behavior of objects in that category.
       All objects belong to a specific class. For example, our /etc/hostname
       object belongs to the "File" class. When we want to create a specific
       object, we start with its class, and construct or instantiate an
       object. A specific object is often referred to as an instance of a
       class.
       In Perl, any package can be a class. The difference between a package
       which is a class and one which isn't is based on how the package is
       used. Here's our "class declaration" for the "File" class:
         package File;
       In Perl, there is no special keyword for constructing an object.
       However, most OO modules on CPAN use a method named "new()" to
       construct a new object:
         my $hostname = File->new(
             path          => '/etc/hostname',
             content       => "foo\n",
             last_mod_time => 1304974868,
         );
       (Don't worry about that "->" operator, it will be explained later.)
       Blessing
       As we said earlier, most Perl objects are hashes, but an object can be
       an instance of any Perl data type (scalar, array, etc.). Turning a
       plain data structure into an object is done by blessing that data
       structure using Perl's "bless" function.
       While we strongly suggest you don't build your objects from scratch,
       you should know the term bless. A blessed data structure (aka "a
       referent") is an object. We sometimes say that an object has been
       "blessed into a class".
       Once a referent has been blessed, the "blessed" function from the
       Scalar::Util core module can tell us its class name. This subroutine
       returns an object's class when passed an object, and false otherwise.
         use Scalar::Util 'blessed';
         print blessed($hash);      # undef
         print blessed($hostname);  # File
       Constructor
       A constructor creates a new object. In Perl, a class's constructor is
       just another method, unlike some other languages, which provide syntax
       for constructors. Most Perl classes use "new" as the name for their
       constructor:
         my $file = File->new(...);
   Methods
       You already learned that a method is a subroutine that operates on an
       object. You can think of a method as the things that an object can do.
       If an object is a noun, then methods are its verbs (save, print, open).
       In Perl, methods are simply subroutines that live in a class's package.
       Methods are always written to receive the object as their first
       argument:
         sub print_info {
             my $self = shift;
             print "This file is at ", $self->path, "\n";
         }
         $file->print_info;
         # The file is at /etc/hostname
       What makes a method special is how it's called. The arrow operator
       ("->") tells Perl that we are calling a method.
       When we make a method call, Perl arranges for the method's invocant to
       be passed as the first argument. Invocant is a fancy name for the thing
       on the left side of the arrow. The invocant can either be a class name
       or an object. We can also pass additional arguments to the method:
         sub print_info {
             my $self   = shift;
             my $prefix = shift // "This file is at ";
             print $prefix, ", ", $self->path, "\n";
         }
         $file->print_info("The file is located at ");
         # The file is located at /etc/hostname
   Attributes
       Each class can define its attributes. When we instantiate an object, we
       assign values to those attributes. For example, every "File" object has
       a path. Attributes are sometimes called properties.
       Perl has no special syntax for attributes. Under the hood, attributes
       are often stored as keys in the object's underlying hash, but don't
       worry about this.
       We recommend that you only access attributes via accessor methods.
       These are methods that can get or set the value of each attribute. We
       saw this earlier in the "print_info()" example, which calls
       "$self->path".
       You might also see the terms getter and setter. These are two types of
       accessors. A getter gets the attribute's value, while a setter sets it.
       Another term for a setter is mutator
       Attributes are typically defined as read-only or read-write. Read-only
       attributes can only be set when the object is first created, while
       read-write attributes can be altered at any time.
       The value of an attribute may itself be another object. For example,
       instead of returning its last mod time as a number, the "File" class
       could return a DateTime object representing that value.
       It's possible to have a class that does not expose any publicly
       settable attributes. Not every class has attributes and methods.
   Polymorphism
       Polymorphism is a fancy way of saying that objects from two different
       classes share an API. For example, we could have "File" and "WebPage"
       classes which both have a "print_content()" method. This method might
       produce different output for each class, but they share a common
       interface.
       While the two classes may differ in many ways, when it comes to the
       "print_content()" method, they are the same. This means that we can try
       to call the "print_content()" method on an object of either class, and
       we don't have to know what class the object belongs to!
       Polymorphism is one of the key concepts of object-oriented design.
   Inheritance
       Inheritance lets you create a specialized version of an existing class.
       Inheritance lets the new class to reuse the methods and attributes of
       another class.
       For example, we could create an "File::MP3" class which inherits from
       "File". An "File::MP3" is-a more specific type of "File".  All mp3
       files are files, but not all files are mp3 files.
       We often refer to inheritance relationships as parent-child or
       "superclass/subclass" relationships. Sometimes we say that the child
       has an is-a relationship with its parent class.
       "File" is a superclass of "File::MP3", and "File::MP3" is a subclass of
       "File".
         package File::MP3;
         use parent 'File';
       The parent module is one of several ways that Perl lets you define
       inheritance relationships.
       Perl allows multiple inheritance, which means that a class can inherit
       from multiple parents. While this is possible, we strongly recommend
       against it. Generally, you can use roles to do everything you can do
       with multiple inheritance, but in a cleaner way.
       Note that there's nothing wrong with defining multiple subclasses of a
       given class. This is both common and safe. For example, we might define
       "File::MP3::FixedBitrate" and "File::MP3::VariableBitrate" classes to
       distinguish between different types of mp3 file.
       Overriding methods and method resolution
       Inheritance allows two classes to share code. By default, every method
       in the parent class is also available in the child. The child can
       explicitly override a parent's method to provide its own
       implementation. For example, if we have an "File::MP3" object, it has
       the "print_info()" method from "File":
         my $cage = File::MP3->new(
             path          => 'mp3s/My-Body-Is-a-Cage.mp3',
             content       => $mp3_data,
             last_mod_time => 1304974868,
             title         => 'My Body Is a Cage',
         );
         $cage->print_info;
         # The file is at mp3s/My-Body-Is-a-Cage.mp3
       If we wanted to include the mp3's title in the greeting, we could
       override the method:
         package File::MP3;
         use parent 'File';
         sub print_info {
             my $self = shift;
             print "This file is at ", $self->path, "\n";
             print "Its title is ", $self->title, "\n";
         }
         $cage->print_info;
         # The file is at mp3s/My-Body-Is-a-Cage.mp3
         # Its title is My Body Is a Cage
       The process of determining what method should be used is called method
       resolution. What Perl does is look at the object's class first
       ("File::MP3" in this case). If that class defines the method, then that
       class's version of the method is called. If not, Perl looks at each
       parent class in turn. For "File::MP3", its only parent is "File". If
       "File::MP3" does not define the method, but "File" does, then Perl
       calls the method in "File".
       If "File" inherited from "DataSource", which inherited from "Thing",
       then Perl would keep looking "up the chain" if necessary.
       It is possible to explicitly call a parent method from a child:
         package File::MP3;
         use parent 'File';
         sub print_info {
             my $self = shift;
             $self->SUPER::print_info();
             print "Its title is ", $self->title, "\n";
         }
       The "SUPER::" bit tells Perl to look for the "print_info()" in the
       "File::MP3" class's inheritance chain. When it finds the parent class
       that implements this method, the method is called.
       We mentioned multiple inheritance earlier. The main problem with
       multiple inheritance is that it greatly complicates method resolution.
       See perlobj for more details.
   Encapsulation
       Encapsulation is the idea that an object is opaque. When another
       developer uses your class, they don't need to know how it is
       implemented, they just need to know what it does.
       Encapsulation is important for several reasons. First, it allows you to
       separate the public API from the private implementation. This means you
       can change that implementation without breaking the API.
       Second, when classes are well encapsulated, they become easier to
       subclass. Ideally, a subclass uses the same APIs to access object data
       that its parent class uses. In reality, subclassing sometimes involves
       violating encapsulation, but a good API can minimize the need to do
       this.
       We mentioned earlier that most Perl objects are implemented as hashes
       under the hood. The principle of encapsulation tells us that we should
       not rely on this. Instead, we should use accessor methods to access the
       data in that hash. The object systems that we recommend below all
       automate the generation of accessor methods. If you use one of them,
       you should never have to access the object as a hash directly.
   Composition
       In object-oriented code, we often find that one object references
       another object. This is called composition, or a has-a relationship.
       Earlier, we mentioned that the "File" class's "last_mod_time" accessor
       could return a DateTime object. This is a perfect example of
       composition. We could go even further, and make the "path" and
       "content" accessors return objects as well. The "File" class would then
       be composed of several other objects.
   Roles
       Roles are something that a class does, rather than something that it
       is. Roles are relatively new to Perl, but have become rather popular.
       Roles are applied to classes. Sometimes we say that classes consume
       roles.
       Roles are an alternative to inheritance for providing polymorphism.
       Let's assume we have two classes, "Radio" and "Computer". Both of these
       things have on/off switches. We want to model that in our class
       definitions.
       We could have both classes inherit from a common parent, like
       "Machine", but not all machines have on/off switches. We could create a
       parent class called "HasOnOffSwitch", but that is very artificial.
       Radios and computers are not specializations of this parent. This
       parent is really a rather ridiculous creation.
       This is where roles come in. It makes a lot of sense to create a
       "HasOnOffSwitch" role and apply it to both classes. This role would
       define a known API like providing "turn_on()" and "turn_off()" methods.
       Perl does not have any built-in way to express roles. In the past,
       people just bit the bullet and used multiple inheritance. Nowadays,
       there are several good choices on CPAN for using roles.
   When to Use OO
       Object Orientation is not the best solution to every problem. In Perl
       Best Practices (copyright 2004, Published by O'Reilly Media, Inc.),
       Damian Conway provides a list of criteria to use when deciding if OO is
       the right fit for your problem:
       o   The system being designed is large, or is likely to become large.
       o   The data can be aggregated into obvious structures, especially if
           there's a large amount of data in each aggregate.
       o   The various types of data aggregate form a natural hierarchy that
           facilitates the use of inheritance and polymorphism.
       o   You have a piece of data on which many different operations are
           applied.
       o   You need to perform the same general operations on related types of
           data, but with slight variations depending on the specific type of
           data the operations are applied to.
       o   It's likely you'll have to add new data types later.
       o   The typical interactions between pieces of data are best
           represented by operators.
       o   The implementation of individual components of the system is likely
           to change over time.
       o   The system design is already object-oriented.
       o   Large numbers of other programmers will be using your code modules.
PERL OO SYSTEMS
       As we mentioned before, Perl's built-in OO system is very minimal, but
       also quite flexible. Over the years, many people have developed systems
       which build on top of Perl's built-in system to provide more features
       and convenience.
       We strongly recommend that you use one of these systems. Even the most
       minimal of them eliminates a lot of repetitive boilerplate. There's
       really no good reason to write your classes from scratch in Perl.
       If you are interested in the guts underlying these systems, check out
       perlobj.
   Moose
       Moose bills itself as a "postmodern object system for Perl 5". Don't be
       scared, the "postmodern" label is a callback to Larry's description of
       Perl as "the first postmodern computer language".
       "Moose" provides a complete, modern OO system. Its biggest influence is
       the Common Lisp Object System, but it also borrows ideas from Smalltalk
       and several other languages. "Moose" was created by Stevan Little, and
       draws heavily from his work on the Perl 6 OO design.
       Here is our "File" class using "Moose":
         package File;
         use Moose;
         has path          => ( is => 'ro' );
         has content       => ( is => 'ro' );
         has last_mod_time => ( is => 'ro' );
         sub print_info {
             my $self = shift;
             print "This file is at ", $self->path, "\n";
         }
       "Moose" provides a number of features:
       o   Declarative sugar
           "Moose" provides a layer of declarative "sugar" for defining
           classes.  That sugar is just a set of exported functions that make
           declaring how your class works simpler and more palatable.  This
           lets you describe what your class is, rather than having to tell
           Perl how to implement your class.
           The "has()" subroutine declares an attribute, and "Moose"
           automatically creates accessors for these attributes. It also takes
           care of creating a "new()" method for you. This constructor knows
           about the attributes you declared, so you can set them when
           creating a new "File".
       o   Roles built-in
           "Moose" lets you define roles the same way you define classes:
             package HasOnOfSwitch;
             use Moose::Role;
             has is_on => (
                 is  => 'rw',
                 isa => 'Bool',
             );
             sub turn_on {
                 my $self = shift;
                 $self->is_on(1);
             }
             sub turn_off {
                 my $self = shift;
                 $self->is_on(0);
             }
       o   A miniature type system
           In the example above, you can see that we passed "isa => 'Bool'" to
           "has()" when creating our "is_on" attribute. This tells "Moose"
           that this attribute must be a boolean value. If we try to set it to
           an invalid value, our code will throw an error.
       o   Full introspection and manipulation
           Perl's built-in introspection features are fairly minimal. "Moose"
           builds on top of them and creates a full introspection layer for
           your classes. This lets you ask questions like "what methods does
           the File class implement?" It also lets you modify your classes
           programmatically.
       o   Self-hosted and extensible
           "Moose" describes itself using its own introspection API. Besides
           being a cool trick, this means that you can extend "Moose" using
           "Moose" itself.
       o   Rich ecosystem
           There is a rich ecosystem of "Moose" extensions on CPAN under the
           MooseX <http://search.cpan.org/search?query=MooseX&mode=dist>;
           namespace. In addition, many modules on CPAN already use "Moose",
           providing you with lots of examples to learn from.
       o   Many more features
           "Moose" is a very powerful tool, and we can't cover all of its
           features here. We encourage you to learn more by reading the
           "Moose" documentation, starting with Moose::Manual
           <http://search.cpan.org/perldoc?Moose::Manual>.
       Of course, "Moose" isn't perfect.
       "Moose" can make your code slower to load. "Moose" itself is not small,
       and it does a lot of code generation when you define your class. This
       code generation means that your runtime code is as fast as it can be,
       but you pay for this when your modules are first loaded.
       This load time hit can be a problem when startup speed is important,
       such as with a command-line script or a "plain vanilla" CGI script that
       must be loaded each time it is executed.
       Before you panic, know that many people do use "Moose" for command-line
       tools and other startup-sensitive code. We encourage you to try "Moose"
       out first before worrying about startup speed.
       "Moose" also has several dependencies on other modules. Most of these
       are small stand-alone modules, a number of which have been spun off
       from "Moose". "Moose" itself, and some of its dependencies, require a
       compiler. If you need to install your software on a system without a
       compiler, or if having any dependencies is a problem, then "Moose" may
       not be right for you.
       Mouse
       If you try "Moose" and find that one of these issues is preventing you
       from using "Moose", we encourage you to consider Mouse next.  "Mouse"
       implements a subset of "Moose"'s functionality in a simpler package.
       For all features that it does implement, the end-user API is identical
       to "Moose", meaning you can switch from "Mouse" to "Moose" quite
       easily.
       "Mouse" does not implement most of "Moose"'s introspection API, so it's
       often faster when loading your modules. Additionally, all of its
       required dependencies ship with the Perl core, and it can run without a
       compiler. If you do have a compiler, "Mouse" will use it to compile
       some of its code for a speed boost.
       Finally, it ships with a "Mouse::Tiny" module that takes most of
       "Mouse"'s features and bundles them up in a single module file. You can
       copy this module file into your application's library directory for
       easy bundling.
       The "Moose" authors hope that one day "Mouse" can be made obsolete by
       improving "Moose" enough, but for now it provides a worthwhile
       alternative to "Moose".
   Class::Accessor
       Class::Accessor is the polar opposite of "Moose". It provides very few
       features, nor is it self-hosting.
       It is, however, very simple, pure Perl, and it has no non-core
       dependencies. It also provides a "Moose-like" API on demand for the
       features it supports.
       Even though it doesn't do much, it is still preferable to writing your
       own classes from scratch.
       Here's our "File" class with "Class::Accessor":
         package File;
         use Class::Accessor 'antlers';
         has path          => ( is => 'ro' );
         has content       => ( is => 'ro' );
         has last_mod_time => ( is => 'ro' );
         sub print_info {
             my $self = shift;
             print "This file is at ", $self->path, "\n";
         }
       The "antlers" import flag tells "Class::Accessor" that you want to
       define your attributes using "Moose"-like syntax. The only parameter
       that you can pass to "has" is "is". We recommend that you use this
       Moose-like syntax if you choose "Class::Accessor" since it means you
       will have a smoother upgrade path if you later decide to move to
       "Moose".
       Like "Moose", "Class::Accessor" generates accessor methods and a
       constructor for your class.
   Object::Tiny
       Finally, we have Object::Tiny. This module truly lives up to its name.
       It has an incredibly minimal API and absolutely no dependencies (core
       or not). Still, we think it's a lot easier to use than writing your own
       OO code from scratch.
       Here's our "File" class once more:
         package File;
         use Object::Tiny qw( path content last_mod_time );
         sub print_info {
             my $self = shift;
             print "This file is at ", $self->path, "\n";
         }
       That's it!
       With "Object::Tiny", all accessors are read-only. It generates a
       constructor for you, as well as the accessors you define.
   Role::Tiny
       As we mentioned before, roles provide an alternative to inheritance,
       but Perl does not have any built-in role support. If you choose to use
       Moose, it comes with a full-fledged role implementation. However, if
       you use one of our other recommended OO modules, you can still use
       roles with Role::Tiny
       "Role::Tiny" provides some of the same features as Moose's role system,
       but in a much smaller package. Most notably, it doesn't support any
       sort of attribute declaration, so you have to do that by hand.  Still,
       it's useful, and works well with "Class::Accessor" and "Object::Tiny"
   OO System Summary
       Here's a brief recap of the options we covered:
       o   Moose
           "Moose" is the maximal option. It has a lot of features, a big
           ecosystem, and a thriving user base. We also covered Mouse briefly.
           "Mouse" is "Moose" lite, and a reasonable alternative when Moose
           doesn't work for your application.
       o   Class::Accessor
           "Class::Accessor" does a lot less than "Moose", and is a nice
           alternative if you find "Moose" overwhelming. It's been around a
           long time and is well battle-tested. It also has a minimal "Moose"
           compatibility mode which makes moving from "Class::Accessor" to
           "Moose" easy.
       o   Object::Tiny
           "Object::Tiny" is the absolute minimal option. It has no
           dependencies, and almost no syntax to learn. It's a good option for
           a super minimal environment and for throwing something together
           quickly without having to worry about details.
       o   Role::Tiny
           Use "Role::Tiny" with "Class::Accessor" or "Object::Tiny" if you
           find yourself considering multiple inheritance. If you go with
           "Moose", it comes with its own role implementation.
   Other OO Systems
       There are literally dozens of other OO-related modules on CPAN besides
       those covered here, and you're likely to run across one or more of them
       if you work with other people's code.
       In addition, plenty of code in the wild does all of its OO "by hand",
       using just the Perl built-in OO features. If you need to maintain such
       code, you should read perlobj to understand exactly how Perl's built-in
       OO works.
CONCLUSION
       As we said before, Perl's minimal OO system has led to a profusion of
       OO systems on CPAN. While you can still drop down to the bare metal and
       write your classes by hand, there's really no reason to do that with
       modern Perl.
       For small systems, Object::Tiny and Class::Accessor both provide
       minimal object systems that take care of basic boilerplate for you.
       For bigger projects, Moose provides a rich set of features that will
       let you focus on implementing your business logic.
       We encourage you to play with and evaluate Moose, Class::Accessor, and
       Object::Tiny to see which OO system is right for you.

perl v5.16.3                      2013-03-04                      PERLOOTUT(1)