Moose - A postmodern object system for Perl 5


Moose - A postmodern object system for Perl 5


  package Point;
  use Moose; # automatically turns on strict and warnings
  has 'x' => (is => 'rw', isa => 'Int');
  has 'y' => (is => 'rw', isa => 'Int');
  sub clear {
      my $self = shift;
  package Point3D;
  use Moose;
  extends 'Point';
  has 'z' => (is => 'rw', isa => 'Int');
  after 'clear' => sub {
      my $self = shift;


Moose is an extension of the Perl 5 object system.

Another object system!?!?

Yes, I know there has been an explosion recently of new ways to build objects in Perl 5, most of them based on inside-out objects and other such things. Moose is different because it is not a new object system for Perl 5, but instead an extension of the existing object system.

Moose is built on top of the Class::MOP manpage, which is a metaclass system for Perl 5. This means that Moose not only makes building normal Perl 5 objects better, but it also provides the power of metaclass programming.

Is this for real? Or is this just an experiment?

Moose is based on the prototypes and experiments I did for the Perl 6 meta-model. However, Moose is NOT an experiment/prototype; it is for real.

Is this ready for use in production?

Yes, I believe that it is.

Moose has been used successfully in production environemnts by several people and companies (including the one I work for). There are Moose applications which have been in production with little or no issue now for well over a year. I consider it highly stable and we are commited to keeping it stable.

Of course, in the end, you need to make this call yourself. If you have any questions or concerns, please feel free to email me, or even the list or just stop by #moose and ask away.

Is Moose just Perl 6 in Perl 5?

No. While Moose is very much inspired by Perl 6, it is not itself Perl 6. Instead, it is an OO system for Perl 5. I built Moose because I was tired of writing the same old boring Perl 5 OO code, and drooling over Perl 6 OO. So instead of switching to Ruby, I wrote Moose :)

Wait, post modern, I thought it was just modern?

So I was reading Larry Wall's talk from the 1999 Linux World entitled ``Perl, the first postmodern computer language'' in which he talks about how he picked the features for Perl because he thought they were cool and he threw out the ones that he thought sucked. This got me thinking about how we have done the same thing in Moose. For Moose, we have ``borrowed'' features from Perl 6, CLOS (LISP), Smalltalk, Java, BETA, OCaml, Ruby and more, and the bits we didn't like (cause they sucked) we tossed aside. So for this reason (and a few others) I have re-dubbed Moose a postmodern object system.

Nuff Said.

Moose Extensions

The the MooseX:: manpage namespace is the official place to find Moose extensions. There are a number of these modules out on CPAN right now the best way to find them is to search for MooseX:: on


Moose makes every attempt to provide as much convenience as possible during class construction/definition, but still stay out of your way if you want it to. Here are a few items to note when building classes with Moose.

Unless specified with extends, any class which uses Moose will inherit from the Moose::Object manpage.

Moose will also manage all attributes (including inherited ones) that are defined with has. And (assuming you call new, which is inherited from the Moose::Object manpage) this includes properly initializing all instance slots, setting defaults where appropriate, and performing any type constraint checking or coercion.


Moose provides a number of methods to all your classes, mostly through the inheritance of the Moose::Object manpage. There is however, one exception.

This is a method which provides access to the current class's metaclass.


Moose will export a number of functions into the class's namespace which may then be used to set up the class. These functions all work directly on the current class.

extends (@superclasses)
This function will set the superclass(es) for the current class.

This approach is recommended instead of use base, because use base actually pushes onto the class's @ISA, whereas extends will replace it. This is important to ensure that classes which do not have superclasses still properly inherit from the Moose::Object manpage.

with (@roles)
This will apply a given set of @roles to the local class.

has $name => %options
This will install an attribute of a given $name into the current class. The %options are the same as those provided by the Class::MOP::Attribute manpage, in addition to the list below which are provided by Moose (the Moose::Meta::Attribute manpage to be more specific):
is => 'rw'|'ro'
The is option accepts either rw (for read/write) or ro (for read only). These will create either a read/write accessor or a read-only accessor respectively, using the same name as the $name of the attribute.

If you need more control over how your accessors are named, you can use the reader, writer and accessor options inherited from the Class::MOP::Attribute manpage, however if you use those, you won't need the is option.

isa => $type_name
The isa option uses Moose's type constraint facilities to set up runtime type checking for this attribute. Moose will perform the checks during class construction, and within any accessors. The $type_name argument must be a string. The string may be either a class name or a type defined using Moose's type definition features. (Refer to the Moose::Util::TypeConstraints manpage for information on how to define a new type, and how to retrieve type meta-data).

coerce => (1|0)
This will attempt to use coercion with the supplied type constraint to change the value passed into any accessors or constructors. You must have supplied a type constraint in order for this to work. See the Moose::Cookbook::Recipe5 manpage for an example.

does => $role_name
This will accept the name of a role which the value stored in this attribute is expected to have consumed.

required => (1|0)
This marks the attribute as being required. This means a defined value must be supplied during class construction, and the attribute may never be set to undef with an accessor.

weak_ref => (1|0)
This will tell the class to store the value of this attribute as a weakened reference. If an attribute is a weakened reference, it cannot also be coerced.

lazy => (1|0)
This will tell the class to not create this slot until absolutely necessary. If an attribute is marked as lazy it must have a default supplied.

auto_deref => (1|0)
This tells the accessor whether to automatically dereference the value returned. This is only legal if your isa option is either ArrayRef or HashRef.

trigger => $code
The trigger option is a CODE reference which will be called after the value of the attribute is set. The CODE ref will be passed the instance itself, the updated value and the attribute meta-object (this is for more advanced fiddling and can typically be ignored). You cannot have a trigger on a read-only attribute.

handles => ARRAY | HASH | REGEXP | ROLE | CODE
The handles option provides Moose classes with automated delegation features. This is a pretty complex and powerful option. It accepts many different option formats, each with its own benefits and drawbacks.

NOTE: The class being delegated to does not need to be a Moose based class, which is why this feature is especially useful when wrapping non-Moose classes.

All handles option formats share the following traits:

You cannot override a locally defined method with a delegated method; an exception will be thrown if you try. That is to say, if you define foo in your class, you cannot override it with a delegated foo. This is almost never something you would want to do, and if it is, you should do it by hand and not use Moose.

You cannot override any of the methods found in Moose::Object, or the BUILD and DEMOLISH methods. These will not throw an exception, but will silently move on to the next method in the list. My reasoning for this is that you would almost never want to do this, since it usually breaks your class. As with overriding locally defined methods, if you do want to do this, you should do it manually, not with Moose.

You do not need to have a reader (or accessor) for the attribute in order to delegate to it. Moose will create a means of accessing the value for you, however this will be several times less efficient then if you had given the attribute a reader (or accessor) to use.

Below is the documentation for each option format:

This is the most common usage for handles. You basically pass a list of method names to be delegated, and Moose will install a delegation method for each one.

This is the second most common usage for handles. Instead of a list of method names, you pass a HASH ref where each key is the method name you want installed locally, and its value is the name of the original method in the class being delegated to.

This can be very useful for recursive classes like trees. Here is a quick example (soon to be expanded into a Moose::Cookbook::Recipe):

  package Tree;
  use Moose;
  has 'node' => (is => 'rw', isa => 'Any');
  has 'children' => (
      is      => 'ro',
      isa     => 'ArrayRef',
      default => sub { [] }
  has 'parent' => (
      is          => 'rw',
      isa         => 'Tree',
      weak_ref => 1,
      handles     => {
          parent_node => 'node',
          siblings    => 'children',

In this example, the Tree package gets parent_node and siblings methods, which delegate to the node and children methods (respectively) of the Tree instance stored in the parent slot.

The regexp option works very similar to the ARRAY option, except that it builds the list of methods for you. It starts by collecting all possible methods of the class being delegated to, then filters that list using the regexp supplied here.

NOTE: An isa option is required when using the regexp option format. This is so that we can determine (at compile time) the method list from the class. Without an isa this is just not possible.

With the role option, you specify the name of a role whose ``interface'' then becomes the list of methods to handle. The ``interface'' can be defined as; the methods of the role and any required methods of the role. It should be noted that this does not include any method modifiers or generated attribute methods (which is consistent with role composition).

This is the option to use when you really want to do something funky. You should only use it if you really know what you are doing, as it involves manual metaclass twiddling.

This takes a code reference, which should expect two arguments. The first is the attribute meta-object this handles is attached to. The second is the metaclass of the class being delegated to. It expects you to return a hash (not a HASH ref) of the methods you want mapped.

metaclass => $metaclass_name
This tells the class to use a custom attribute metaclass for this particular attribute. Custom attribute metaclasses are useful for extending the capabilities of the has keyword: they are the simplest way to extend the MOP, but they are still a fairly advanced topic and too much to cover here, see the Moose::Cookbook::Recipe11 manpage for more information.

The default behavior here is to just load $metaclass_name; however, we also have a way to alias to a shorter name. This will first look to see if Moose::Meta::Attribute::Custom::$metaclass_name exists. If it does, Moose will then check to see if that has the method register_implementation, which should return the actual name of the custom attribute metaclass. If there is no register_implementation method, it will fall back to using Moose::Meta::Attribute::Custom::$metaclass_name as the metaclass name.

traits => [ @role_names ]
This tells Moose to take the list of @role_names and apply them to the attribute meta-object. This is very similar to the metaclass option, but allows you to use more than one extension at a time. This too is an advanced topic, we don't yet have a cookbook for it though.

As with metaclass, the default behavior is to just load $role_name; however, we also have a way to alias to a shorter name. This will first look to see if Moose::Meta::Attribute::Custom::Trait::$role_name exists. If it does, Moose will then check to see if that has the method register_implementation, which should return the actual name of the custom attribute trait. If there is no register_implementation method, it will fall back to using Moose::Meta::Attribute::Custom::Trait::$metaclass_name as the trait name.

has +$name => %options
This is variation on the normal attibute creator has which allows you to clone and extend an attribute from a superclass or from a role. Here is an example of the superclass usage:
  package Foo;
  use Moose;
  has 'message' => (
      is      => 'rw',
      isa     => 'Str',
      default => 'Hello, I am a Foo'
  package My::Foo;
  use Moose;
  extends 'Foo';
  has '+message' => (default => 'Hello I am My::Foo');

What is happening here is that My::Foo is cloning the message attribute from its parent class Foo, retaining the is => 'rw' and isa => 'Str' characteristics, but changing the value in default.

Here is another example, but within the context of a role:

  package Foo::Role;
  use Moose::Role;
  has 'message' => (
      is      => 'rw',
      isa     => 'Str',
      default => 'Hello, I am a Foo'
  package My::Foo;
  use Moose;
  with 'Foo::Role';
  has '+message' => (default => 'Hello I am My::Foo');

In this case, we are basically taking the attribute which the role supplied and altering it within the bounds of this feature.

Aside from where the attributes come from (one from superclass, the other from a role), this feature works exactly the same. This feature is restricted somewhat, so as to try and force at least some sanity into it. You are only allowed to change the following attributes:

Change the default value of an attribute.

Change whether the attribute attempts to coerce a value passed to it.

Change if the attribute is required to have a value.

Change the documentation string associated with the attribute.

Change if the attribute lazily initializes the slot.

You are allowed to change the type without restriction.

It is recommended that you use this freedom with caution. We used to only allow for extension only if the type was a subtype of the parent's type, but we felt that was too restrictive and is better left as a policy descision.

You are allowed to add a new handles definition, but you are not allowed to change one.

You are allowed to add a new builder definition, but you are not allowed to change one.

You are allowed to add a new metaclass definition, but you are not allowed to change one.

You are allowed to add additional traits to the traits definition. These traits will be composed into the attribute, but pre-existing traits are not overridden, or removed.

before $name|@names => sub { ... }
after $name|@names => sub { ... }
around $name|@names => sub { ... }
This three items are syntactic sugar for the before, after, and around method modifier features that the Class::MOP manpage provides. More information on these may be found in the Class::MOP::Class documentation for now.

The keyword super is a no-op when called outside of an override method. In the context of an override method, it will call the next most appropriate superclass method with the same arguments as the original method.

override ($name, &sub)
An override method is a way of explicitly saying ``I am overriding this method from my superclass''. You can call super within this method, and it will work as expected. The same thing can be accomplished with a normal method call and the SUPER:: pseudo-package; it is really your choice.

The keyword inner, much like super, is a no-op outside of the context of an augment method. You can think of inner as being the inverse of super; the details of how inner and augment work is best described in the the Moose::Cookbook::Recipe7 manpage.

augment ($name, &sub)
An augment method, is a way of explicitly saying ``I am augmenting this method from my superclass''. Once again, the details of how inner and augment work is best described in the the Moose::Cookbook::Recipe7 manpage.

This is the Carp::confess function, and exported here because I use it all the time.

This is the Scalar::Util::blessed function, it is exported here because I use it all the time. It is highly recommended that this is used instead of ref anywhere you need to test for an object's class name.



Moose offers a way to remove the keywords it exports, through the unimport method. You simply have to say no Moose at the bottom of your code for this to work. Here is an example:

    package Person;
    use Moose;
    has 'first_name' => (is => 'rw', isa => 'Str');
    has 'last_name'  => (is => 'rw', isa => 'Str');
    sub full_name {
        my $self = shift;
        $self->first_name . ' ' . $self->last_name
    no Moose; # keywords are removed from the Person package


Moose also offers some options for extending or embedding it into your own framework. The basic premise is to have something that sets up your class' metaclass and export the moose declarators (has, with, extends,...). Here is an example:

    package MyFramework;
    use Moose;
    sub import {
        my $CALLER = caller();
        # we should never export to main
        return if $CALLER eq 'main';
        Moose::init_meta( $CALLER, 'MyFramework::Base' );
        Moose->import({into => $CALLER});
        # Do my custom framework stuff
        return 1;


Moose's import method supports the the Sub::Exporter manpage form of {into => $pkg} and {into_level => 1}

init_meta ($class, $baseclass, $metaclass)

Moose does some boot strapping: it creates a metaclass object for your class, and then injects a meta accessor into your class to retrieve it. Then it sets your baseclass to Moose::Object or the value you pass in unless you already have one. This is all done via init_meta which takes the name of your class and optionally a baseclass and a metaclass as arguments.



I blame Sam Vilain for introducing me to the insanity that is meta-models.
I blame Audrey Tang for then encouraging my meta-model habit in #perl6.
Without Yuval ``nothingmuch'' Kogman this module would not be possible, and it certainly wouldn't have this name ;P
The basis of the TypeContraints module was Rob Kinyon's idea originally, I just ran with it.
Thanks to mst & chansen and the whole #moose posse for all the early ideas/feature-requests/encouragement/bug-finding.
Thanks to David ``Theory'' Wheeler for meta-discussions and spelling fixes.

This is the official web home of Moose, it contains links to our public SVN repo as well as links to a number of talks and articles on Moose and Moose related technologies.

the Class::MOP manpage documentation
The #moose channel on
The Moose mailing list -
Moose stats on -
Several Moose extension modules in the the MooseX:: manpage namespace.


The Art of the MetaObject Protocol
I mention this in the the Class::MOP manpage docs too, this book was critical in the development of both modules and is highly recommended.

This paper (suggested by lbr on #moose) was what lead to the implementation of the super/override and inner/augment features. If you really want to understand them, I suggest you read this.


All complex software has bugs lurking in it, and this module is no exception. If you find a bug please either email me, or add the bug to cpan-RT.


Stevan Little <>

with contributions from:


Adam (Alias) Kennedy

Anders (Debolaz) Nor Berle

Nathan (kolibre) Gray

Christian (chansen) Hansen

Hans Dieter (confound) Pearcey

Eric (ewilhelm) Wilhelm

Guillermo (groditi) Roditi

Jess (castaway) Robinson

Matt (mst) Trout

Robert (phaylon) Sedlacek

Robert (rlb3) Boone

Scott (konobi) McWhirter

Shlomi (rindolf) Fish

Yuval (nothingmuch) Kogman

Chris (perigrin) Prather

Jonathan (jrockway) Rockway

Piotr (dexter) Roszatycki

Sam (mugwump) Vilain

Shawn (sartak) Moore

... and many other #moose folks


Copyright 2006-2008 by Infinity Interactive, Inc.

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

 Moose - A postmodern object system for Perl 5