The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
package Moose::Spec::Role;

# ABSTRACT: Formal spec for Role behavior



=pod

=head1 NAME

Moose::Spec::Role - Formal spec for Role behavior

=head1 VERSION

version 2.0604

=head1 DESCRIPTION

B<NOTE:> This document is currently incomplete.

=head2 Components of a Role

=over 4

=item Excluded Roles

A role can have a list of excluded roles, these are basically
roles that they shouldn't be composed with. This is not just
direct composition either, but also "inherited" composition.

This feature was taken from the Fortress language and is really
of most use when building a large set of role "building blocks"
some of which should never be used together.

=item Attributes

A roles attributes are similar to those of a class, except that
they are not actually applied. This means that methods that are
generated by an attributes accessor will not be generated in the
role, but only created once the role is applied to a class.

=item Methods

These are the methods defined within the role. Simple as that.

=item Required Methods

A role can require a consuming class (or role) to provide a
given method. Failure to do so for classes is a fatal error,
while for roles it simply passes on the method requirement to
the consuming role.

=item Required Attributes

Just as a role can require methods, it can also require attributes.
The requirement fulfilling attribute must implement at least as much
as is required. That means, for instance, that if the role requires
that the attribute be read-only, then it must at least have a reader
and can also have a writer. It means that if the role requires that
the attribute be an ArrayRef, then it must either be an ArrayRef or
a subtype of an ArrayRef.

=item Overridden Methods

The C<override> and C<super> keywords are allowed in roles, but
their behavior is different from that of its class counterparts.
The C<super> in a class refers directly to that class's superclass,
while the C<super> in a role is deferred and only has meaning once
the role is composed into a class. Once that composition occurs,
C<super> then refers to that class's superclass.

It is key to remember that roles do not have hierarchy, so they
can never have a I<super> role.

=item Method Modifiers

These are the C<before>, C<around> and C<after> modifiers provided
in Moose classes. The difference here is that the modifiers are not
actually applied until the role is composed into a class (this is
just like attributes and the C<override> keyword).

=back

=head2 Role Composition

=head3 Composing into a Class

=over 4

=item Excluded Roles

=item Required Methods

=item Required Attributes

=item Attributes

=item Methods

=item Overridden methods

=item Method Modifiers (before, around, after)

=back

=head3 Composing into a Instance

=head3 Composing into a Role

=over 4

=item Excluded Roles

=item Required Methods

=item Required Attributes

=item Attributes

=item Methods

=item Overridden methods

=item Method Modifiers (before, around, after)

=back

=head3 Role Summation

When multiple roles are added to another role (using the
C<with @roles> keyword) the roles are composed symmetrically.
The product of the composition is a composite role
(L<Moose::Meta::Role::Composite>).

=over 4

=item Excluded Roles

=item Required Methods

=item Required Attributes

=item Attributes

Attributes with the same name will conflict and are considered
a unrecoverable error. No other aspect of the attribute is
examined, it is enough that just the attribute names conflict.

The reason for such early and harsh conflicts with attributes
is because there is so much room for variance between two
attributes that the problem quickly explodes and rules get
very complex. It is my opinion that this complexity is not
worth the trouble.

=item Methods

Methods with the same name will conflict, but no error is
thrown, instead the method name is added to the list of
I<required> methods for the new composite role.

To look at this in terms of set theory, each role can be
said to have a set of methods. The symmetric difference of
these two sets is the new set of methods for the composite
role, while the intersection of these two sets are the
conflicts. This can be illustrated like so:

   Role A has method set { a, b, c }
   Role B has method set { c, d, e }

   The composite role (A,B) has
       method   set { a, b, d, e }
       conflict set { c }

=item Overridden methods

An overridden method can conflict in one of two ways.

The first way is with another overridden method of the same
name, and this is considered an unrecoverable error. This
is an obvious error since you cannot override a method twice
in the same class.

The second way for conflict is for an overridden method and a
regular method to have the same name. This is also an unrecoverable
error since there is no way to combine these two, nor is it
okay for both items to be composed into a single class at some
point.

The use of override in roles can be tricky, but if used
carefully they can be a very powerful tool.

=item Method Modifiers (before, around, after)

Method modifiers are the only place where the ordering of
role composition matters. This is due to the nature of
method modifiers themselves.

Since a method can have multiple method modifiers, these
are just collected in order to be later applied to the
class in that same order.

In general, great care should be taken in using method
modifiers in roles. The order sensitivity can possibly
lead to subtle and difficult to find bugs if they are
overused. As with all good things in life, moderation
is the key.

=back

=head3 Composition Edge Cases

This is a just a set of complex edge cases which can easily get
confused. This attempts to clarify those cases and provide an
explanation of what is going on in them.

=over 4

=item Role Method Overriding

Many people want to "override" methods in roles they are consuming.
This works fine for classes, since the local class method is favored
over the role method. However in roles it is trickier, this is because
conflicts result in neither method being chosen and the method being
"required" instead.

Here is an example of this (incorrect) type of overriding.

    package Role::Foo;
    use Moose::Role;

    sub foo { ... }

    package Role::FooBar;
    use Moose::Role;

    with 'Role::Foo';

    sub foo { ... }
    sub bar { ... }

Here the C<foo> methods conflict and the Role::FooBar now requires a
class or role consuming it to implement C<foo>. This is very often not
what the user wants.

Now here is an example of the (correct) type of overriding, only it is
not overriding at all, as is explained in the text below.

    package Role::Foo;
    use Moose::Role;

    sub foo { ... }

    package Role::Bar;
    use Moose::Role;

    sub foo { ... }
    sub bar { ... }

    package Role::FooBar;
    use Moose::Role;

    with 'Role::Foo', 'Role::Bar';

    sub foo { ... }

This works because the combination of Role::Foo and Role::Bar produce
a conflict with the C<foo> method. This conflict results in the
composite role (that was created by the combination of Role::Foo
and Role::Bar using the I<with> keyword) having a method requirement
of C<foo>. The Role::FooBar then fulfills this requirement.

It is important to note that Role::FooBar is simply fulfilling the
required C<foo> method, and **NOT** overriding C<foo>. This is an
important distinction to make.

Now here is another example of a (correct) type of overriding, this
time using the I<excludes> option.

    package Role::Foo;
    use Moose::Role;

    sub foo { ... }

    package Role::FooBar;
    use Moose::Role;

    with 'Role::Foo' => { -excludes => 'foo' };

    sub foo { ... }
    sub bar { ... }

By specifically excluding the C<foo> method during composition,
we allow B<Role::FooBar> to define its own version of C<foo>.

=back

=head1 SEE ALSO

=over 4

=item Traits

Roles are based on Traits, which originated in the Smalltalk
community.

=over 4

=item L<http://www.iam.unibe.ch/~scg/Research/Traits/>

This is the main site for the original Traits papers.

=item L<Class::Trait>

I created this implementation of traits several years ago,
after reading the papers linked above. (This module is now
maintained by Ovid and I am no longer involved with it).

=back

=item Roles

Since they are relatively new, and the Moose implementation
is probably the most mature out there, roles don't have much
to link to. However, here is some bits worth looking at (mostly
related to Perl 6)

=over 4

=item L<http://www.oreillynet.com/onlamp/blog/2006/08/roles_composable_units_of_obje.html>

This is chromatic's take on roles, which is worth reading since
he was/is one of the big proponents of them.

=item L<http://svn.perl.org/perl6/doc/trunk/design/syn/S12.pod>

This is Synopsis 12, which is all about the Perl 6 Object System.
Which, of course, includes roles.

=back

=back

=head1 AUTHOR

Moose is maintained by the Moose Cabal, along with the help of many contributors. See L<Moose/CABAL> and L<Moose/CONTRIBUTORS> for details.

=head1 COPYRIGHT AND LICENSE

This software is copyright (c) 2012 by Infinity Interactive, Inc..

This is free software; you can redistribute it and/or modify it under
the same terms as the Perl 5 programming language system itself.

=cut


__END__