Moo - Minimalist Object Orientation (with Moose compatibility)


 package Cat::Food;

 use Moo;
 use namespace::clean;

 sub feed_lion {
   my $self = shift;
   my $amount = shift || 1;

   $self->pounds( $self->pounds - $amount );

 has taste => (
   is => 'ro',

 has brand => (
   is  => 'ro',
   isa => sub {
     die "Only SWEET-TREATZ supported!" unless $_[0] eq 'SWEET-TREATZ'

 has pounds => (
   is  => 'rw',
   isa => sub { die "$_[0] is too much cat food!" unless $_[0] < 15 },


And elsewhere:

 my $full = Cat::Food->new(
    taste  => 'DELICIOUS.',
    brand  => 'SWEET-TREATZ',
    pounds => 10,


 say $full->pounds;


This module is an extremely light-weight subset of Moose optimised for rapid startup and "pay only for what you use".

It also avoids depending on any XS modules to allow simple deployments. The name Moo is based on the idea that it provides almost -- but not quite -- two thirds of Moose.

Unlike Mouse this module does not aim at full compatibility with Moose's surface syntax, preferring instead of provide full interoperability via the metaclass inflation capabilities described in "MOO AND MOOSE".

For a full list of the minor differences between Moose and Moo's surface syntax, see "INCOMPATIBILITIES WITH MOOSE".


If you want a full object system with a rich Metaprotocol, Moose is already wonderful.

However, sometimes you're writing a command line script or a CGI script where fast startup is essential, or code designed to be deployed as a single file via App::FatPacker, or you're writing a CPAN module and you want it to be usable by people with those constraints.

I've tried several times to use Mouse but it's 3x the size of Moo and takes longer to load than most of my Moo based CGI scripts take to run.

If you don't want Moose, you don't want "less metaprotocol" like Mouse, you want "as little as possible" -- which means "no metaprotocol", which is what Moo provides.

Better still, if you install and load Moose, we set up metaclasses for your Moo classes and Moo::Role roles, so you can use them in Moose code without ever noticing that some of your codebase is using Moo.

Hence, Moo exists as its name -- Minimal Object Orientation -- with a pledge to make it smooth to upgrade to Moose when you need more than minimal features.


If Moo detects Moose being loaded, it will automatically register metaclasses for your Moo and Moo::Role packages, so you should be able to use them in Moose code without anybody ever noticing you aren't using Moose everywhere.

Moo will also create Moose type constraints for classes and roles, so that isa => 'MyClass' and isa => 'MyRole' work the same as for Moose classes and roles.

Extending a Moose class or consuming a Moose::Role will also work.

So will extending a Mouse class or consuming a Mouse::Role - but note that we don't provide Mouse metaclasses or metaroles so the other way around doesn't work. This feature exists for Any::Moose users porting to Moo; enabling Mouse users to use Moo classes is not a priority for us.

This means that there is no need for anything like Any::Moose for Moo code - Moo and Moose code should simply interoperate without problem. To handle Mouse code, you'll likely need an empty Moo role or class consuming or extending the Mouse stuff since it doesn't register true Moose metaclasses like Moo does.

If you want types to be upgraded to the Moose types, use MooX::Types::MooseLike and install the MooseX::Types library to match the MooX::Types::MooseLike library you're using - Moo will load the MooseX::Types library and use that type for the newly created metaclass.

If you need to disable the metaclass creation, add:

  no Moo::sification;

to your code before Moose is loaded, but bear in mind that this switch is currently global and turns the mechanism off entirely so don't put this in library code.


If a new enough version of Class::XSAccessor is available, it will be used to generate simple accessors, readers, and writers for a speed boost. Simple accessors are those without lazy defaults, type checks/coercions, or triggers. Readers and writers generated by Class::XSAccessor will behave slightly differently: they will reject attempts to call them with the incorrect number of parameters.


Any::Moose will load Mouse normally, and Moose in a program using Moose - which theoretically allows you to get the startup time of Mouse without disadvantaging Moose users.

Sadly, this doesn't entirely work, since the selection is load order dependent - Moo's metaclass inflation system explained above in "MOO AND MOOSE" is significantly more reliable.

So if you want to write a CPAN module that loads fast or has only pure perl dependencies but is also fully usable by Moose users, you should be using Moo.

For a full explanation, see the article which explains the differing strategies in more detail and provides a direct example of where Moo succeeds and Any::Moose fails.



 Foo::Bar->new( attr1 => 3 );


 Foo::Bar->new({ attr1 => 3 });


   my ( $class, @args ) = @_;

   unshift @args, "attr1" if @args % 2 == 1;

   return { @args };

 Foo::Bar->new( 3 );

The default implementation of this method accepts a hash or hash reference of named parameters. If it receives a single argument that isn't a hash reference it throws an error.

You can override this method in your class to handle other types of options passed to the constructor.

This method should always return a hash reference of named options.


If you are inheriting from a non-Moo class, the arguments passed to the parent class constructor can be manipulated by defining a FOREIGNBUILDARGS method. It will receive the same arguments as BUILDARGS, and should return a list of arguments to pass to the parent class constructor.


Define a BUILD method on your class and the constructor will automatically call the BUILD method from parent down to child after the object has been instantiated. Typically this is used for object validation or possibly logging.


If you have a DEMOLISH method anywhere in your inheritance hierarchy, a DESTROY method is created on first object construction which will call $instance->DEMOLISH($in_global_destruction) for each DEMOLISH method from child upwards to parents.

Note that the DESTROY method is created on first construction of an object of your class in order to not add overhead to classes without DEMOLISH methods; this may prove slightly surprising if you try and define your own.


 if ($foo->does('Some::Role1')) {

Returns true if the object composes in the passed role.



 extends 'Parent::Class';

Declares base class. Multiple superclasses can be passed for multiple inheritance (but please use roles instead). The class will be loaded, however no errors will be triggered if it can't be found and there are already subs in the class.

Calling extends more than once will REPLACE your superclasses, not add to them like 'use base' would.


 with 'Some::Role1';


 with 'Some::Role1', 'Some::Role2';

Composes one or more Moo::Role (or Role::Tiny) roles into the current class. An error will be raised if these roles have conflicting methods. The roles will be loaded using the same mechansim as extends uses.


 has attr => (
   is => 'ro',

Declares an attribute for the class.

 package Foo;
 use Moo;
 has 'attr' => (
   is => 'ro'

 package Bar;
 use Moo;
 extends 'Foo';
 has '+attr' => (
   default => sub { "blah" },

Using the + notation, it's possible to override an attribute.

The options for has are as follows:

  • is

    required, may be ro, lazy, rwp or rw.

    ro generates an accessor that dies if you attempt to write to it - i.e. a getter only - by defaulting reader to the name of the attribute.

    lazy generates a reader like ro, but also sets lazy to 1 and builder to _build_${attribute_name} to allow on-demand generated attributes. This feature was my attempt to fix my incompetence when originally designing lazy_build, and is also implemented by MooseX::AttributeShortcuts. There is, however, nothing to stop you using lazy and builder yourself with rwp or rw - it's just that this isn't generally a good idea so we don't provide a shortcut for it.

    rwp generates a reader like ro, but also sets writer to _set_${attribute_name} for attributes that are designed to be written from inside of the class, but read-only from outside. This feature comes from MooseX::AttributeShortcuts.

    rw generates a normal getter/setter by defaulting accessor to the name of the attribute.

  • isa

    Takes a coderef which is meant to validate the attribute. Unlike Moose, Moo does not include a basic type system, so instead of doing isa => 'Num', one should do

     isa => sub {
       die "$_[0] is not a number!" unless looks_like_number $_[0]

    Note that the return value is ignored, only whether the sub lives or dies matters.

    Sub::Quote aware

    Since Moo does not run the isa check before coerce if a coercion subroutine has been supplied, isa checks are not structural to your code and can, if desired, be omitted on non-debug builds (although if this results in an uncaught bug causing your program to break, the Moo authors guarantee nothing except that you get to keep both halves).

    If you want MooseX::Types style named types, look at MooX::Types::MooseLike.

    To cause your isa entries to be automatically mapped to named Moose::Meta::TypeConstraint objects (rather than the default behaviour of creating an anonymous type), set:

      $Moo::HandleMoose::TYPE_MAP{$isa_coderef} = sub {
        require MooseX::Types::Something;
        return MooseX::Types::Something::TypeName();

    Note that this example is purely illustrative; anything that returns a Moose::Meta::TypeConstraint object or something similar enough to it to make Moose happy is fine.

  • coerce

    Takes a coderef which is meant to coerce the attribute. The basic idea is to do something like the following:

     coerce => sub {
       $_[0] % 2 ? $_[0] : $_[0] + 1

    Note that Moo will always fire your coercion: this is to permit isa entries to be used purely for bug trapping, whereas coercions are always structural to your code. We do, however, apply any supplied isa check after the coercion has run to ensure that it returned a valid value.

    Sub::Quote aware

  • handles

    Takes a string

      handles => 'RobotRole'

    Where RobotRole is a role (Moo::Role) that defines an interface which becomes the list of methods to handle.

    Takes a list of methods

     handles => [ qw( one two ) ]

    Takes a hashref

     handles => {
       un => 'one',
  • trigger

    Takes a coderef which will get called any time the attribute is set. This includes the constructor, but not default or built values. Coderef will be invoked against the object with the new value as an argument.

    If you set this to just 1, it generates a trigger which calls the _trigger_${attr_name} method on $self. This feature comes from MooseX::AttributeShortcuts.

    Note that Moose also passes the old value, if any; this feature is not yet supported.

    Sub::Quote aware

  • default

    Takes a coderef which will get called with $self as its only argument to populate an attribute if no value is supplied to the constructor - or if the attribute is lazy, when the attribute is first retrieved if no value has yet been provided.

    If a simple scalar is provided, it will be inlined as a string. Any non-code reference (hash, array) will result in an error - for that case instead use a code reference that returns the desired value.

    Note that if your default is fired during new() there is no guarantee that other attributes have been populated yet so you should not rely on their existence.

    Sub::Quote aware

  • predicate

    Takes a method name which will return true if an attribute has a value.

    If you set this to just 1, the predicate is automatically named has_${attr_name} if your attribute's name does not start with an underscore, or _has_${attr_name_without_the_underscore} if it does. This feature comes from MooseX::AttributeShortcuts.

  • builder

    Takes a method name which will be called to create the attribute - functions exactly like default except that instead of calling


    Moo will call


    The following features come from MooseX::AttributeShortcuts:

    If you set this to just 1, the builder is automatically named _build_${attr_name}.

    If you set this to a coderef or code-convertible object, that variable will be installed under $class::_build_${attr_name} and the builder set to the same name.

  • clearer

    Takes a method name which will clear the attribute.

    If you set this to just 1, the clearer is automatically named clear_${attr_name} if your attribute's name does not start with an underscore, or _clear_${attr_name_without_the_underscore} if it does. This feature comes from MooseX::AttributeShortcuts.

  • lazy

    Boolean. Set this if you want values for the attribute to be grabbed lazily. This is usually a good idea if you have a "builder" which requires another attribute to be set.

  • required

    Boolean. Set this if the attribute must be passed on instantiation.

  • reader

    The value of this attribute will be the name of the method to get the value of the attribute. If you like Java style methods, you might set this to get_foo

  • writer

    The value of this attribute will be the name of the method to set the value of the attribute. If you like Java style methods, you might set this to set_foo.

  • weak_ref

    Boolean. Set this if you want the reference that the attribute contains to be weakened; use this when circular references are possible, which will cause leaks.

  • init_arg

    Takes the name of the key to look for at instantiation time of the object. A common use of this is to make an underscored attribute have a non-underscored initialization name. undef means that passing the value in on instantiation is ignored.

  • moosify

    Takes either a coderef or array of coderefs which is meant to transform the given attributes specifications if necessary when upgrading to a Moose role or class. You shouldn't need this by default, but is provided as a means of possible extensibility.


 before foo => sub { ... };

See "before method(s) => sub { ... }" in Class::Method::Modifiers for full documentation.


 around foo => sub { ... };

See "around method(s) => sub { ... }" in Class::Method::Modifiers for full documentation.


 after foo => sub { ... };

See "after method(s) => sub { ... }" in Class::Method::Modifiers for full documentation.


"quote_sub" in Sub::Quote allows us to create coderefs that are "inlineable," giving us a handy, XS-free speed boost. Any option that is Sub::Quote aware can take advantage of this.

To do this, you can write

  use Sub::Quote;

  use Moo;
  use namespace::clean;

  has foo => (
    is => 'ro',
    isa => quote_sub(q{ die "Not <3" unless $_[0] < 3 })

which will be inlined as

  do {
    local @_ = ($_[0]->{foo});
    die "Not <3" unless $_[0] < 3;

or to avoid localizing @_,

  has foo => (
    is => 'ro',
    isa => quote_sub(q{ my ($val) = @_; die "Not <3" unless $val < 3 })

which will be inlined as

  do {
    my ($val) = ($_[0]->{foo});
    die "Not <3" unless $val < 3;

See Sub::Quote for more information, including how to pass lexical captures that will also be compiled into the subroutine.


Moo will not clean up imported subroutines for you; you will have to do that manually. The recommended way to do this is to declare your imports first, then use Moo, then use namespace::clean. Anything imported before namespace::clean will be scrubbed. Anything imported or declared after will be still be available.

 package Record;

 use Digest::MD5 qw(md5_hex);

 use Moo;
 use namespace::clean;

 has name => (is => 'ro', required => 1);
 has id => (is => 'lazy');
 sub _build_id {
   my ($self) = @_;
   return md5_hex($self->name);


If you were to import md5_hex after namespace::clean you would be able to call ->md5_hex() on your Record instances (and it probably wouldn't do what you expect!).

Moo::Roles behave slightly differently. Since their methods are composed into the consuming class, they can do a little more for you automatically. As long as you declare your imports before calling use Moo::Role, those imports and the ones Moo::Role itself provides will not be composed into consuming classes, so there's usually no need to use namespace::clean.

On namespace::autoclean: If you're coming to Moo from the Moose world, you may be accustomed to using namespace::autoclean in all your packages. This is not recommended for Moo packages, because namespace::autoclean will inflate your class to a full Moose class. It'll work, but you will lose the benefits of Moo. Instead you are recommended to just use namespace::clean.


There is no built-in type system. isa is verified with a coderef; if you need complex types, just make a library of coderefs, or better yet, functions that return quoted subs. MooX::Types::MooseLike provides a similar API to MooseX::Types::Moose so that you can write

  has days_to_live => (is => 'ro', isa => Int);

and have it work with both; it is hoped that providing only subrefs as an API will encourage the use of other type systems as well, since it's probably the weakest part of Moose design-wise.

initializer is not supported in core since the author considers it to be a bad idea and Moose best practices recommend avoiding it. Meanwhile trigger or coerce are more likely to be able to fulfill your needs.

There is no meta object. If you need this level of complexity you wanted Moose - Moo succeeds at being small because it explicitly does not provide a metaprotocol. However, if you load Moose, then


will return an appropriate metaclass pre-populated by Moo.

No support for super, override, inner, or augment - the author considers augment to be a bad idea, and override can be translated:

  override foo => sub {

  around foo => sub {
    my ($orig, $self) = (shift, shift);

The dump method is not provided by default. The author suggests loading Devel::Dwarn into main:: (via perl -MDevel::Dwarn ... for example) and using $obj->$::Dwarn() instead.

"default" only supports coderefs and plain scalars, because passing a hash or array reference as a default is almost always incorrect since the value is then shared between all objects using that default.

lazy_build is not supported; you are instead encouraged to use the is => 'lazy' option supported by Moo and MooseX::AttributeShortcuts.

auto_deref is not supported since the author considers it a bad idea and it has been considered best practice to avoid it for some time.

documentation will show up in a Moose metaclass created from your class but is otherwise ignored. Then again, Moose ignores it as well, so this is arguably not an incompatibility.

Since coerce does not require isa to be defined but Moose does require it, the metaclass inflation for coerce alone is a trifle insane and if you attempt to subtype the result will almost certainly break.

BUILDARGS is not triggered if your class does not have any attributes. Without attributes, BUILDARGS return value would be ignored, so we just skip calling the method instead.

Handling of warnings: when you use Moo we enable FATAL warnings, and some several extra pragmas when used in development: indirect, multidimensional, and bareword::filehandles. See the strictures documentation for the details on this.

A similar invocation for Moose would be:

  use Moose;
  use warnings FATAL => "all";

Additionally, Moo supports a set of attribute option shortcuts intended to reduce common boilerplate. The set of shortcuts is the same as in the Moose module MooseX::AttributeShortcuts as of its version 0.009+. So if you:

    package MyClass;
    use Moo;

The nearest Moose invocation would be:

    package MyClass;

    use Moose;
    use warnings FATAL => "all";
    use MooseX::AttributeShortcuts;

or, if you're inheriting from a non-Moose class,

    package MyClass;

    use Moose;
    use MooseX::NonMoose;
    use warnings FATAL => "all";
    use MooseX::AttributeShortcuts;

Finally, Moose requires you to call


at the end of your class to get an inlined (i.e. not horribly slow) constructor. Moo does it automatically the first time ->new is called on your class. (make_immutable is a no-op in Moo to ease migration.)

An extension MooX::late exists to ease translating Moose packages to Moo by providing a more Moose-like interface.


Users' IRC: #moose on

(click for instant chatroom login)

Development and contribution IRC: #web-simple on

(click for instant chatroom login)


Git repository: git://

Git browser:


mst - Matt S. Trout (cpan:MSTROUT) <>


dg - David Leadbeater (cpan:DGL) <>

frew - Arthur Axel "fREW" Schmidt (cpan:FREW) <>

hobbs - Andrew Rodland (cpan:ARODLAND) <>

jnap - John Napiorkowski (cpan:JJNAPIORK) <>

ribasushi - Peter Rabbitson (cpan:RIBASUSHI) <>

chip - Chip Salzenberg (cpan:CHIPS) <>

ajgb - Alex J. G. Burzyński (cpan:AJGB) <>

doy - Jesse Luehrs (cpan:DOY) <doy at tozt dot net>

perigrin - Chris Prather (cpan:PERIGRIN) <>

Mithaldu - Christian Walde (cpan:MITHALDU) <>

ilmari - Dagfinn Ilmari Mannsåker (cpan:ILMARI) <>

tobyink - Toby Inkster (cpan:TOBYINK) <>

haarg - Graham Knop (cpan:HAARG) <>

mattp - Matt Phillips (cpan:MATTP) <>

bluefeet - Aran Deltac (cpan:BLUEFEET) <>


Copyright (c) 2010-2011 the Moo "AUTHOR" and "CONTRIBUTORS" as listed above.


This library is free software and may be distributed under the same terms as perl itself. See