=pod

=head1 NAME

Moose::Cookbook::Recipe4 - Subtypes, and modeling a simple B<Company> class hierarchy

=head1 SYNOPSIS
  
  package Address;
  use Moose;
  use Moose::Util::TypeConstraints;
  
  use Locale::US;
  use Regexp::Common 'zip';
  
  my $STATES = Locale::US->new;
  
  subtype USState 
      => as Str
      => where {
          (exists $STATES->{code2state}{uc($_)} || 
           exists $STATES->{state2code}{uc($_)})
      };
      
  subtype USZipCode 
      => as Value
      => where {
          /^$RE{zip}{US}{-extended => 'allow'}$/            
      };
  
  has 'street'   => (is => 'rw', isa => 'Str');
  has 'city'     => (is => 'rw', isa => 'Str');
  has 'state'    => (is => 'rw', isa => 'USState');
  has 'zip_code' => (is => 'rw', isa => 'USZipCode');   
  
  package Company;
  use Moose;
  use Moose::Util::TypeConstraints;
  
  has 'name'      => (is => 'rw', isa => 'Str', required => 1);
  has 'address'   => (is => 'rw', isa => 'Address'); 
  has 'employees' => (is => 'rw', isa => 'ArrayRef[Employee]');    
  
  sub BUILD {
      my ($self, $params) = @_;
      if ($params->{employees}) {
          foreach my $employee (@{$params->{employees}}) {
              $employee->company($self);
          }
      }
  }
  
  after 'employees' => sub {
      my ($self, $employees) = @_;
      if (defined $employees) {
          foreach my $employee (@{$employees}) {
              $employee->company($self);
          }            
      }
  };  
  
  package Person;
  use Moose;
  
  has 'first_name'     => (is => 'rw', isa => 'Str', required => 1);
  has 'last_name'      => (is => 'rw', isa => 'Str', required => 1);       
  has 'middle_initial' => (is => 'rw', isa => 'Str', 
                           predicate => 'has_middle_initial');  
  has 'address'        => (is => 'rw', isa => 'Address');
  
  sub full_name {
      my $self = shift;
      return $self->first_name . 
            ($self->has_middle_initial ? 
                ' ' . $self->middle_initial . '. ' 
                : 
                ' ') .
             $self->last_name;
  }
    
  package Employee;
  use Moose;  
  
  extends 'Person';
  
  has 'title'   => (is => 'rw', isa => 'Str', required => 1);
  has 'company' => (is => 'rw', isa => 'Company', weak_ref => 1);  
  
  override 'full_name' => sub {
      my $self = shift;
      super() . ', ' . $self->title
  };

=head1 DESCRIPTION

In this recipe we introduce the C<subtype> keyword, and show 
how it can be useful for specifying type constraints 
without building an entire class to represent them. We 
will also show how this feature can be used to leverage the 
usefulness of CPAN modules. In addition to this, we will 
introduce another attribute option.

Let's first look at the C<subtype> feature. In the B<Address> class we have
defined two subtypes. The first C<subtype> uses the L<Locale::US> module, which
provides two hashes which can be used to perform existential checks for state
names and their two letter state codes. It is a very simple and very useful
module, and perfect for use in a C<subtype> constraint.
  
  my $STATES = Locale::US->new;  
  subtype USState 
      => as Str
      => where {
          (exists $STATES->{code2state}{uc($_)} || 
           exists $STATES->{state2code}{uc($_)})
      };

Because we know that states will be passed to us as strings, we 
can make C<USState> a subtype of the built-in type constraint 
C<Str>. This will ensure that anything which is a C<USState> will 
also pass as a C<Str>. Next, we create a constraint specializer 
using the C<where> keyword. The value being checked against in 
the C<where> clause can be found in the C<$_> variable (1). Our 
constraint specializer will then check whether the given string 
is either a state name or a state code. If the string meets this 
criteria, then the constraint will pass, otherwise it will fail.
We can now use this as we would any built-in constraint, like so:

  has 'state' => (is => 'rw', isa => 'USState');

The C<state> accessor will now check all values against the 
C<USState> constraint, thereby only allowing valid state names or 
state codes to be stored in the C<state> slot. 

The next C<subtype> does pretty much the same thing using the L<Regexp::Common>
module, and is used as the constraint for the C<zip_code> slot.

  subtype USZipCode 
      => as Value
      => where {
          /^$RE{zip}{US}{-extended => 'allow'}$/            
      };

Using subtypes can save a lot of unnecessary abstraction by not requiring you to
create many small classes for these relatively simple values. They also allow
you to reuse the same constraints in a number of classes (thereby avoiding
duplication), since all type constraints are stored in a global registry and
always accessible to C<has>.

With these two subtypes and some attributes, we have defined
as much as we need for a basic B<Address> class. Next, we define 
a basic B<Company> class, which itself has an address. As we saw in 
earlier recipes, we can use the C<Address> type constraint that 
Moose automatically created for us:

  has 'address' => (is => 'rw', isa => 'Address');

A company also needs a name, so we define that as well:

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

Here we introduce another attribute option, the C<required> option. 
This option tells Moose that C<name> is a required parameter in 
the B<Company> constructor, and that the C<name> accessor cannot 
accept an undefined value for the slot. The result is that C<name> 
will always have a value. 

The next attribute option is not actually new, but a new variant 
of options we have already introduced:
  
  has 'employees' => (is => 'rw', isa => 'ArrayRef[Employee]');

Here, we are passing a more complex string to the C<isa> option, we 
are passing a container type constraint. Container type constraints 
can either be C<ArrayRef> or C<HashRef> with a contained type given 
inside the square brackets. This basically checks that all the values 
in the ARRAY ref are instances of the B<Employee> class. 

This will ensure that our employees will all be of the correct type. However,
the B<Employee> object (which we will see in a moment) also maintains a
reference to its associated B<Company>. In order to maintain this relationship
(and preserve the referential integrity of our objects), we need to perform some
processing of the employees over and above that of the type constraint check.
This is accomplished in two places. First we need to be sure that any employees
array passed to the constructor is properly initialized. For this we can use the
C<BUILD> method (2):
  
  sub BUILD {
      my ($self, $params) = @_;
      if ($params->{employees}) {
          foreach my $employee (@{$params->{employees}}) {
              $employee->company($self);
          }
      }
  }

The C<BUILD> method will be executed after the initial type constraint 
check, so we can simply perform a basic existential check on the C<employees>
param here, and assume that if it does exist, it is both an ARRAY ref 
and contains I<only> instances of B<Employee>.

The next aspect we need to address is the C<employees> read/write 
accessor (see the C<employees> attribute declaration above). This 
accessor will correctly check the type constraint, but we need to extend it
with some additional processing. For this we use an C<after> method modifier,
like so:

  after 'employees' => sub {
      my ($self, $employees) = @_;
      if (defined $employees) {
          foreach my $employee (@{$employees}) {
              $employee->company($self);
          }            
      }
  };

Again, as with the C<BUILD> method, we know that the type constraint 
check has already happened, so we can just check for defined-ness on the 
C<$employees> argument.

At this point, our B<Company> class is complete. Next comes our B<Person> 
class and its subclass, the previously mentioned B<Employee> class. 

The B<Person> class should be obvious to you at this point. It has a few 
C<required> attributes, and the C<middle_initial> slot has an additional 
C<predicate> method (which we saw in the previous recipe with the 
B<BinaryTree> class). 

Next, the B<Employee> class, which should also be pretty obvious at this 
point. It requires a C<title>, and maintains a weakened reference to a 
B<Company> instance. The only new item, which we have seen before in 
examples, but never in the recipe itself, is the C<override> method 
modifier:

  override 'full_name' => sub {
      my $self = shift;
      super() . ', ' . $self->title
  };

This just tells Moose that I am intentionally overriding the superclass 
C<full_name> method here, and adding the value of the C<title> slot at 
the end of the employee's full name.

And that's about it.

Once again, as with all the other recipes, you can go about using 
these classes like any other Perl 5 class. A more detailed example of 
usage can be found in F<t/004_recipe.t>.

=head1 CONCLUSION

This recipe was intentionally longer and more complex to illustrate both 
how easily Moose classes can interact (using class type constraints, etc.)
and the sheer density of information and behaviors which Moose can pack 
into a relatively small amount of typing. Ponder for a moment how much 
more code a non-Moose plain old Perl 5 version of this recipe would have 
been (including all the type constraint checks, weak references, and so on).

And of course, this recipe also introduced the C<subtype> keyword, and 
its usefulness within the Moose toolkit. In the next recipe we will 
focus more on subtypes, and introduce the idea of type coercion as well.

=head1 FOOTNOTES

=over 4

=item (1)

The value being checked is also passed as the first argument to 
the C<where> block as well, so it can also be accessed as C<$_[0]> 
as well.

=item (2)

The C<BUILD> method is called by C<Moose::Object::BUILDALL>, which is 
called by C<Moose::Object::new>. C<BUILDALL> will climb the object 
inheritance graph and call the appropriate C<BUILD> methods in the 
correct order.

=back

=head1 AUTHOR

Stevan Little E<lt>stevan@iinteractive.comE<gt>

=head1 COPYRIGHT AND LICENSE

Copyright 2006-2008 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