++ed by:

17 PAUSE users
8 non-PAUSE users.

David Golden


Class::Tiny - Minimalist class construction


version 0.014


In Person.pm:

  package Person;

  use Class::Tiny qw( name );


In Employee.pm:

  package Employee;
  use parent 'Person';

  use Class::Tiny qw( ssn ), {
    timestamp => sub { time }   # attribute with default


In example.pl:

  use Employee;

  my $obj = Employee->new( name => "Larry", ssn => "111-22-3333" );

  # unknown attributes are fatal:
  eval { Employee->new( name => "Larry", OS => "Linux" ) };
  die "Error creating Employee: $@" if $@;


This module offers a minimalist class construction kit in around 120 lines of code. Here is a list of features:

  • defines attributes via import arguments

  • generates read-write accessors

  • supports lazy attribute defaults

  • supports custom accessors

  • superclass provides a standard new constructor

  • new takes a hash reference or list of key/value pairs

  • new has heuristics to catch constructor attribute typos

  • new calls BUILD for each class from parent to child

  • superclass provides a DESTROY method

  • DESTROY calls DEMOLISH for each class from child to parent

Multiple-inheritance is possible, with superclass order determined via mro::get_linear_isa.

It uses no non-core modules for any recent Perl. On Perls older than v5.10 it requires MRO::Compat. On Perls older than v5.14, it requires Devel::GlobalDestruction.


Defining attributes

Define attributes as a list of import arguments:

    package Foo::Bar;

    use Class::Tiny qw(

For each attribute, a read-write accessor is created unless a subroutine of that name already exists:

    $obj->name;               # getter
    $obj->name( "John Doe" ); # setter

Attribute names must be valid subroutine identifiers or an exception will be thrown.

You can specify lazy defaults by defining attributes with a hash reference. Keys define attribute names and values are constants or code references that will be evaluated when the attribute is first accessed if no value has been set. The object is passed as an argument to a code reference.

    package Foo::WithDefaults;

    use Class::Tiny qw/name id/, {
        title     => 'Peon',
        skills    => sub { [] },
        hire_date => sub { $_[0]->_build_hire_date },

When subclassing, if multiple accessors of the same name exist in different classes, any default (or lack of default) is determined by standard method resolution order.

To make your own custom accessors, just pre-declare the method name before loading Class::Tiny:

    package Foo::Bar;

    use subs 'id';

    use Class::Tiny qw( name id );

    sub id { ... }

By declaring id also with Class::Tiny, you include it in the list of known attributes for introspection. Default values will not be set for custom accessors unless you handle that yourself.

Class::Tiny::Object is your base class

If your class does not already inherit from some class, then Class::Tiny::Object will be added to your @ISA to provide new and DESTROY.

If your class does inherit from something, then no additional inheritance is set up. If the parent subclasses Class::Tiny::Object, then all is well. If not, then you'll get accessors set up but no constructor or destructor. Don't do that unless you really have a special need for it.

Define subclasses as normal. It's best to define them with base, parent or superclass before defining attributes with Class::Tiny so the @ISA array is already populated at compile-time:

    package Foo::Bar::More;

    use parent 'Foo::Bar';

    use Class::Tiny qw( shoe_size );

Object construction

If your class inherits from Class::Tiny::Object (as it should if you followed the advice above), it provides the new constructor for you.

Objects can be created with attributes given as a hash reference or as a list of key/value pairs:

    $obj = Foo::Bar->new( name => "David" );

    $obj = Foo::Bar->new( { name => "David" } );

If a reference is passed as a single argument, it must be able to be dereferenced as a hash or an exception is thrown. A shallow copy is made of the reference provided.

In order to help catch typos in constructor arguments, any argument that it is not also a valid method (e.g. an accessor or other method) will result in a fatal exception. This is not perfect, but should catch typical transposition typos. Also see "BUILD" for how to explicitly hide non-attribute, non-method arguments if desired.


If your class or any superclass defines a BUILD method, it will be called by the constructor from the furthest parent class down to the child class after the object has been created.

It is passed the constructor arguments as a hash reference. The return value is ignored. Use BUILD for validation or setting default values.

    sub BUILD {
        my ($self, $args) = @_;
        $self->foo(42) unless defined $self->foo;
        croak "Foo must be non-negative" if $self->foo < 0;

If you want to hide a non-attribute constructor argument from validation, delete it from the passed-in argument hash reference.

    sub BUILD {
        my ($self, $args) = @_;

        if ( delete $args->{do_something_special} ) {

The argument reference is a copy, so deleting elements won't affect data in the object. You have to delete it from both if that's what you want.

    sub BUILD {
        my ($self, $args) = @_;

        if ( delete $args->{do_something_special} ) {
            delete $self->{do_something_special};


Class::Tiny provides a DESTROY method. If your class or any superclass defines a DEMOLISH method, they will be called from the child class to the furthest parent class during object destruction. It is provided a single boolean argument indicating whether Perl is in global destruction. Return values and errors are ignored.

    sub DEMOLISH {
        my ($self, $global_destruct) = @_;

Introspection and internals

You can retrieve an unsorted list of valid attributes known to Class::Tiny for a class and its superclasses with the get_all_attributes_for class method.

    my @attrs = Class::Tiny->get_all_attributes_for("Employee");
    # returns qw/name ssn timestamp/

Likewise, a hash reference of all valid attributes and default values (or code references) may be retrieved with the get_all_attribute_defaults_for class method. Any attributes without a default will be undef.

    my $def = Class::Tiny->get_all_attribute_defaults_for("Employee");
    # returns {
    #   name => undef,
    #   ssn => undef
    #   timestamp => $coderef
    # }

The import method uses two class methods, prepare_class and create_attributes to set up the @ISA array and attributes. Anyone attempting to extend Class::Tiny itself should use these instead of mocking up a call to import.

When the first object is created, linearized @ISA and various subroutines references are cached for speed. Ensure that all inheritance and methods are in place before creating objects. (You don't want to be changing that once you create objects anyway, right?)


Why this instead of Object::Tiny or Class::Accessor or something else?

I wanted something so simple that it could potentially be used by core Perl modules I help maintain (or hope to write), most of which either use Class::Struct or roll-their-own OO framework each time.

Object::Tiny and Object::Tiny::RW were close to what I wanted, but lacking some features I deemed necessary, and their maintainers have an even more strict philosophy against feature creep than I have.

I also considered Class::Accessor, which has been around a long time and is heavily used, but it, too, lacked features I wanted and did things in ways I considered poor design.

I looked for something else on CPAN, but after checking a dozen class creators I realized I could implement exactly what I wanted faster than I could search CPAN for something merely sufficient.

In general, compared to most things on CPAN (other than Object::Tiny), Class::Tiny is smaller in implementation and simpler in API.

Specifically, here is how Class::Tiny ("C::T") compares to Object::Tiny ("O::T") and Class::Accessor ("C::A"):

 FEATURE                            C::T    O::T      C::A
 attributes defined via import      yes     yes       no
 read/write accessors               yes     no        yes
 lazy attribute defaults            yes     no        no
 provides new                       yes     yes       yes
 provides DESTROY                   yes     no        no
 new takes either hashref or list   yes     no (list) no (hash)
 new validates arguments            yes     no        no
 Moo(se)-like BUILD/DEMOLISH        yes     no        no
 no extraneous methods via @ISA     yes     yes       no

Why this instead of Moose or Moo?

Moose and Moo are both excellent OO frameworks. Moose offers a powerful meta-object protocol (MOP), but is slow to start up and has about 30 non-core dependencies including XS modules. Moo is faster to start up and has about 10 pure Perl dependencies but provides no true MOP, relying instead on its ability to transparently upgrade Moo to Moose when Moose's full feature set is required.

By contrast, Class::Tiny has no MOP and has zero non-core dependencies for Perls in the support window. It has far less code, less complexity and no learning curve. If you don't need or can't afford what Moo or Moose offer, this is intended to be a reasonable fallback.

That said, Class::Tiny offers Moose-like conventions for things like BUILD and DEMOLISH for some minimal interoperability and an easier upgrade path.


Bugs / Feature Requests

Please report any bugs or feature requests through the issue tracker at https://github.com/dagolden/Class-Tiny/issues. You will be notified automatically of any progress on your issue.

Source Code

This is open source software. The code repository is available for public review and contribution under the terms of the license.


  git clone https://github.com/dagolden/Class-Tiny.git


David Golden <dagolden@cpan.org>


  • Dagfinn Ilmari Mannsåker <ilmari@ilmari.org>

  • Gelu Lupas <gelu@devnull.ro>

  • Karen Etheridge <ether@cpan.org>

  • Matt S Trout <mstrout@cpan.org>

  • Olivier Mengué <dolmen@cpan.org>

  • Toby Inkster <tobyink@cpan.org>


This software is Copyright (c) 2013 by David Golden.

This is free software, licensed under:

  The Apache License, Version 2.0, January 2004