Clarify how roles do and do not participate in inheritance.

[gitmo/Moose.git] / lib / Moose / Manual / Attributes.pod

=pod

=head1 NAME

Moose::Manual::Attributes - Object attributes with Moose

=head1 INTRODUCTION

Moose attributes have many properties, and attributes are probably the
single most powerful and flexible part of Moose. You can create a
powerful class simply by declaring attributes. In fact, it's possible
to have classes that consist solely of attribute declarations.

An attribute is a property that every member of a class has. For
example, we might say that "every C<Person> object has a first name and
last name". Attributes can be optional, so that we can say "some C<Person>
objects have a social security number (and some don't)".

At its simplest, an attribute can be thought of as a named value (as
in a hash) that can be read and set. However, attributes can also have
defaults, type constraints, delegation and much more.

In other languages, attributes are also referred to as slots or
properties.

=head1 ATTRIBUTE OPTIONS

Use the C<has> function to declare an attribute:

  package Person;

  use Moose;

  has 'first_name' => ( is => 'rw' );

This says that all C<Person> objects have an optional read-write
"first_name" attribute.

=head2 Read-write Vs read-only

The options passed to C<has> define the properties of the
attribute. There are a many options, but in the simplest form you just
need to set C<is>, which can be either C<rw> (read-write) or C<ro>
(read-only).

(In fact, you could even omit C<is>, but that gives you an attribute
that has no accessors, which is pointless unless you're doing some
deep, dark magic).

=head2 Accessor methods

Each attribute has one or more accessor methods. An accessor lets you
read and write the value of that attribute for an object.

By default, the accessor method has the same name as the attribute. If
you declared your attribute as C<ro> then your accessor will be
read-only. If you declared it read-write, you get a read-write
accessor. Simple.

Given our C<Person> example above, we now have a single C<first_name>
accessor that can read or write a C<Person> object's C<first_name>
attribute's value.

If you want, you can also explicitly specify the method names to be
used for reading and writing an attribute's value. This is
particularly handy when you'd like an attribute to be publicly
readable, but only privately settable. For example:

  has 'weight' => (
      is     => 'rw',
      writer => '_set_weight',
  );

This might be useful if weight is calculated based on other methods,
for example every time the C<eat> method is called, we might adjust
weight. This lets us hide the implementation details of weight
changes, but still provide the weight value to users of the class.

Some people might prefer to have distinct methods for reading and
writing. In I<Perl Best Practices>, Damian Conway recommends that
reader methods start with "get_" and writer methods start with "set_".

We can do exactly that by providing names for both the C<reader> and
C<writer> methods:

  has 'weight' => (
      is     => 'rw',
      reader => 'get_weight',
      writer => 'set_weight',
  );

If you're thinking that doing this over and over would be insanely
tedious, you're right! Fortunately, Moose provides a powerful
extension system that lets override the default naming
conventions. See L<Moose::Manual::MooseX> for more details.

=head2 Predicate and clearer methods

Moose allows you to explicitly distinguish between a false or
undefined attribute value and an attribute which has not been set. If
you want to access this information, you must define clearer and
predicate methods for an attribute.

A predicate method tells you whether or not a given attribute is
currently set. Note an attribute can be explicitly set to C<undef> or
some other false value, but the predicate will return true.

The clearer method unsets the attribute. This is I<not> the
same as setting the value to C<undef>, but you can only distinguish
between them if you define a predicate method!

Here's some code to illustrate the relationship between an accessor,
predicate, and clearer method.

  package Person;

  use Moose;

  has 'ssn' => (
      is        => 'rw',
      clearer   => 'clear_ssn',
      predicate => 'has_ssn',
  );

  ...

  my $person = Person->new();
  $person->has_ssn; # false

  $person->ssn(undef);
  $person->ssn; # returns undef
  $person->has_ssn; # true

  $person->clear_ssn;
  $person->ssn; # returns undef
  $person->has_ssn; # false

  $person->ssn('123-45-6789');
  $person->ssn; # returns '123-45-6789'
  $person->has_ssn; # true

  my $person2 = Person->new( ssn => '111-22-3333');
  $person2->has_ssn; # true

By default, Moose does not make a predicate or clearer for you. You
must explicitly provide names for them.

=head2 Required or not?

By default, all attributes are optional, and do not need to be
provided at object construction time. If you want to make an attribute
required, simply set the C<required> option to true:

  has 'name' => (
      is       => 'rw',
      required => 1,
  );

There are a couple caveats worth mentioning in regards to what
"required" actually means.

Basically, all it says is that this attribute (C<name>) must be provided
to the constructor. It does not say anything about its value, so it
could be C<undef>.

If you define a clearer method on a required attribute, the clearer
I<will> work, so even a required attribute can be unset after object
construction.

This means that if you do make an attribute required, providing a
clearer doesn't make much sense. In some cases, it might be handy to
have a I<private> C<clearer> and C<predicate> for a required
attribute.

=head2 Default and builder methods

Attributes can have default values, and Moose provides two ways to
specify that default.

In the simplest form, you simply provide a non-reference scalar value
for the C<default> option:

  has 'size' => (
      is        => 'rw',
      default   => 'medium',
      predicate => 'has_size',
  );

If the size attribute is not provided to the constructor, then it ends
up being set to C<medium>:

  my $person = Person->new();
  $person->size; # medium
  $person->has_size; # true

You can also provide a subroutine reference for C<default>. This
reference will be called as a method on the object.

  has 'size' => (
      is => 'rw',
      default =>
          sub { ( 'small', 'medium', 'large' )[ int( rand 3 ) ] },
      predicate => 'has_size',
  );

This is dumb example, but it illustrates the point that the subroutine
will be called for every new object created.

When you provide a C<default> subroutine reference, it is called as a
method on the object, with no additional parameters:

  has 'size' => (
      is => 'rw',
      default => sub {
          my $self = shift;

          return $self->height > 200 ? 'big' : 'average';
      },
  );

When the C<default> is called during object construction, it may be
called before other attributes have been set. If your default is
dependent on other parts of the object's state, you can make the
attribute C<lazy>. Laziness is covered in the next section.

If you want to use a reference of any sort as the default value, you
must return it from a subroutine. This is necessary because otherwise
Perl would instantiate the reference exactly once, and it would be
shared by all objects:

  has 'mapping' => (
      is      => 'rw',
      default => {}, # wrong!
  );

Moose will throw an error if you pass a bare non-subroutine reference
as the default.

If Moose allowed this then the default mapping attribute could easily
end up shared across many objects. Instead, wrap it in a subroutine
reference:

  has 'mapping' => (
      is      => 'rw',
      default => sub { {} }, # right!
  );

This is a bit awkward, but it's just the way Perl works.

As an alternative to using a subroutine reference, you can instead
supply a C<builder> method for your attribute:

  has 'size' => (
      is        => 'rw',
      builder   => '_build_size',
      predicate => 'has_size',
  );

  sub _build_size {
      return ( 'small', 'medium', 'large' )[ int( rand 3 ) ];
  }

This has several advantages. First, it moves a chunk of code to its
own named method, which improves readability and code
organization.

We strongly recommend that you use a C<builder> instead of a
C<default> for anything beyond the most trivial default.

A C<builder>, just like a C<default>, is called as a method on the
object with no additional parameters.

=head3 Builders allow subclassing

Because the C<builder> is called I<by name>, it goes through Perl's
method resolution. This means that builder methods are both
inheritable and overridable.

If we subclass our C<Person> class, we can override C<_build_size>:

  package Lilliputian;

  use Moose;
  extends 'Person';

  sub _build_size { return 'small' }

=head3 Builders can be composed from roles

Because builders are called by name, they work well with roles. For
example, a role could provide an attribute but require that the
consuming class provide the C<builder>:

  package HasSize;
  use Moose::Role;

  requires '_build_size';

  has 'size' => (
      is      => 'ro',
      lazy    => 1,
      builder => '_build_animal',
  );

  package Lilliputian;
  use Moose;

  with 'HasSize';

  sub _build_size { return 'small' }

Roles are covered in L<Moose::Manual::Roles>.

=head2 Laziness and C<lazy_build>

Moose lets you defer attribute population by making an attribute
C<lazy>:

  has 'size' => (
      is      => 'rw',
      lazy    => 1,
      builder => '_build_size',
  );

When C<lazy> is true, the default is not generated until the reader
method is called, rather than at object construction time. There are
several reasons you might choose to do this.

First, if the default value for this attribute depends on some other
attributes, then the attribute I<must> be C<lazy>. During object
construction, defaults are not generated in a predictable order, so
you cannot count on some other attribute being populated when
generating a default.

Second, there's often no reason to calculate a default before it's
needed. Making an attribute C<lazy> lets you defer the cost until the
attribute is needed. If the attribute is I<never> needed, you save
some CPU time.

We recommend that you make any attribute with a builder or non-trivial
default C<lazy> as a matter of course.

To facilitate this, you can simply specify the C<lazy_build> attribute
option. This bundles up a number of options together:

  has 'size' => (
      is         => 'rw',
      lazy_build => 1,
  );

This is the same as specifying all of these options:

  has 'size' => (
      is        => 'rw',
      lazy      => 1,
      builder   => '_build_size',
      clearer   => 'clear_size',
      predicate => 'has_size',
  );

If your attribute name starts with an underscore (C<_>), then the clearer
and predicate will as well:

  has '_size' => (
      is         => 'rw',
      lazy_build => 1,
  );

becomes:

  has '_size' => (
      is        => 'rw',
      lazy      => 1,
      builder   => '_build__size',
      clearer   => '_clear_size',
      predicate => '_has_size',
  );

Note the doubled underscore in the builder name. Internally, Moose
simply prepends the attribute name with "_build_" to come up with the
builder name.

If you don't like the names that C<lazy_build> generates, you can
always provide your own:

  has 'size' => (
      is         => 'rw',
      lazy_build => 1,
      clearer    => '_clear_size',
  );

Options that you explicitly provide are always used in favor of
Moose's internal defaults.

=head2 Constructor parameters (C<init_arg>)

By default, each attribute can be passed by name to the class's
constructor. On occasion, you may want to use a different name for
the constructor parameter. You may also want to make an attribute
unsettable via the constructor.

Both of these goals can be accomplished with the C<init_arg> option:

  has 'bigness' => (
      is       => 'rw',
      init_arg => 'size',
  );

Now we have an attribute named "bigness", but we pass C<size> to the
constructor.

Even more useful is the ability to disable setting an attribute via
the constructor. This is particularly handy for private attributes:

  has '_genetic_code' => (
      is         => 'rw',
      lazy_build => 1,
      init_arg   => undef,
  );

By setting the C<init_arg> to C<undef>, we make it impossible to set
this attribute when creating a new object.

=head2 Weak references

Moose has built-in support for weak references. If you set the
C<weak_ref> option to a true value, then it will call
C<Scalar::Util::weaken> whenever the attribute is set:

  has 'parent' => (
      is       => 'rw',
      weak_ref => 1,
  );

  $node->parent($parent_node);

This is very useful when you're building objects that may contain
circular references.

=head2 Triggers

A C<trigger> is a subroutine that is called whenever the attribute is
set:

  has 'size' => (
      is      => 'rw',
      trigger => \&_size_set,
  );

  sub _size_set {
      my ( $self, $size, $meta_attr ) = @_;

      warn $self->name, " size is now $size\n";
  }

The trigger is called as a method, and receives the new value as well
as the L<Moose::Meta::Attribute> object for the attribute. The trigger
is called I<after> the value is set.

This differs from an after method modifier in two ways. First, a
trigger is only called when the attribute is set, as opposed to
whenever the accessor is called. Second, it is also called if the
attribute is set via a lazy default or builder.

=head2 Attribute types

Attributes can be restricted to only accept certain types:

  has 'first_name' => (
      is  => 'rw',
      isa => 'Str',
  );

This says that the C<first_name> attribute must be a string.

Moose also provides a shortcut for specifying that an attribute only
accepts objects that do a certain role:

  has 'weapon' => (
      is   => 'rw',
      does => 'MyApp::Weapon',
  );

See the L<Moose::Manual::Types> documentation for a complete
discussion of Moose's type system.

=head2 Delegation

Attributes can define methods which simply delegate to their values:

  has 'hair_color' => (
      is      => 'rw',
      isa     => 'Graphics::Color::RGB',
      handles => { hair_color_hex => 'as_hex_string' },
  );

This adds a new method, C<hair_color_hex>. When someone calls
C<hair_color_hex>, internally, the object just calls C<<
$self->hair_color->as_hex_string >>.

See L<Moose::Manual::Delegation> for documentation on how to set up
delegation methods.

=head2 Metaclass and traits

One of Moose's best features is that it can be extended in all sorts
of ways through the use of custom metaclasses and metaclass traits.

When declaring an attribute, you can declare a metaclass or a set of
traits for the attribute:

  use MooseX::AttributeHelpers;

  has 'mapping' => (
      metaclass => 'Collection::Hash',
      is        => 'ro',
      default   => sub { {} },
  );

In this case, the metaclass C<Collection::Hash> really refers to
L<MooseX::AttributeHelpers::Collection::Hash>.

You can also apply one or more traits to an attribute:

  use MooseX::MetaDescription;

  has 'size' => (
      is          => 'rw',
      traits      => ['MooseX::MetaDescription::Meta::Trait'],
      description => {
          html_widget  => 'text_input',
          serialize_as => 'element',
      },
  );

The advantage of traits is that you can mix more than one of them
together easily (in fact, a trait is just a role under the hood).

There are a number of MooseX modules on CPAN which provide useful
attribute metaclasses and traits. See L<Moose::Manual::MooseX> for
some examples. You can also write your own metaclasses and traits. See
the "Meta" and "Extending" recipes in L<Moose::Cookbook> for examples.

=head1 ATTRIBUTE INHERITANCE

By default, a child inherits all of its parent class(es)' attributes
as-is. However, you can explicitly change some aspects of the
inherited attribute in the child class.

The options that can be overridden in a subclass are:

=over 4

=item * default

=item * coerce

=item * required

=item * documentation

=item * lazy

=item * isa

=item * handles

=item * builder

=item * metaclass

=item * traits

=back

To override an attribute, you simply prepend its name with a plus sign
(C<+>):

  package LazyPerson;

  use Moose;

  extends 'Person';

  has '+first_name' => (
      lazy    => 1,
      default => 'Bill',
  );

Now the C<first_name> attribute in C<LazyPerson> is lazy, and defaults
to C<'Bill'>.

We recommend that you exercise caution when changing the type (C<isa>)
of an inherited attribute.

=head1 MORE ON ATTRIBUTES

Moose attributes are a big topic, and this document glosses over a few
aspects of their aspects. We recommend that you read the
L<Moose::Manual::Delegation> and L<Moose::Manual::Types> documents to
get a more complete understanding of attribute features.

=head1 A FEW MORE OPTIONS

Moose has lots of attribute options. The ones listed below are
superseded by some more modern features, but are covered for the sake
of completeness.

=head2 The C<documentation> option

You can provide a piece of documentation as a string for an attribute:

  has 'first_name' => (
      is            => 'rw',
      documentation => q{The person's first (personal) name},
  );

Moose does absolutely nothing with this information other than store
it.

=head2 The C<auto_deref> option

If your attribute is an array reference or hash reference, the
C<auto_deref> option will make Moose dereference the value when it is
returned from the reader method:

  my %map = $object->mapping;

This option only works if your attribute is explicitly typed as an
C<ArrayRef> or C<HashRef>.

However, we recommend that you use L<MooseX::AttributeHelpers> for
these types of attributes, which gives you much more control over how
they are accessed and manipulated.

=head2 Initializer

Moose provides an attribute option called C<initializer>. This is
similar to C<builder>, except that it is I<only> called during object
construction.

This option is inherited from L<Class::MOP>, but we recommend that you
use a C<builder> (which is Moose-only) instead.

=head1 AUTHOR

Dave Rolsky E<lt>autarch@urth.orgE<gt>

=head1 COPYRIGHT AND LICENSE

Copyright 2009 by Infinity Interactive, Inc.

L<http://www.iinteractive.com>

This library is free software; you can redistribute it and/or modify
it under the same terms as Perl itself.

=cut