merge trunk to pluggable errors

[gitmo/Moose.git] / lib / Moose / Cookbook / Basics / Recipe5.pod

=pod

=head1 NAME

Moose::Cookbook::Basics::Recipe5 - More subtypes, coercion in a B<Request> class

=head1 SYNOPSIS

  package Request;
  use Moose;
  use Moose::Util::TypeConstraints;
  
  use HTTP::Headers  ();
  use Params::Coerce ();
  use URI            ();
  
  subtype 'Header'
      => as 'Object'
      => where { $_->isa('HTTP::Headers') };
  
  coerce 'Header'
      => from 'ArrayRef'
          => via { HTTP::Headers->new( @{ $_ } ) }
      => from 'HashRef'
          => via { HTTP::Headers->new( %{ $_ } ) };
  
  subtype 'Uri'
      => as 'Object'
      => where { $_->isa('URI') };
  
  coerce 'Uri'
      => from 'Object'
          => via { $_->isa('URI') 
                    ? $_ 
                    : Params::Coerce::coerce( 'URI', $_ ) }
      => from 'Str'
          => via { URI->new( $_, 'http' ) };
  
  subtype 'Protocol'
      => as Str
      => where { /^HTTP\/[0-9]\.[0-9]$/ };
  
  has 'base'     => (is => 'rw', isa => 'Uri', coerce  => 1);
  has 'uri'      => (is => 'rw', isa => 'Uri', coerce  => 1);	
  has 'method'   => (is => 'rw', isa => 'Str');	
  has 'protocol' => (is => 'rw', isa => 'Protocol');		
  has 'headers'  => (
      is      => 'rw',
      isa     => 'Header',
      coerce  => 1,
      default => sub { HTTP::Headers->new } 
  );

=head1 DESCRIPTION

This recipe introduces the idea of type coercions, and the C<coerce> 
keyword. Coercions can be attached to existing type constraints, 
and can be used to transform input of one type into input of another 
type. This can be an extremely powerful tool if used correctly, which 
is why it is off by default. If you want your accessor to attempt 
a coercion, you must specifically ask for it with the B<coerce> option.

Now, onto the coercions. 

First we need to create a subtype to attach our coercion to. Here we 
create a basic I<Header> subtype, which matches any instance of the 
class B<HTTP::Headers>:

  subtype 'Header'
      => as 'Object'
      => where { $_->isa('HTTP::Headers') };

The simplest thing from here would be create an accessor declaration
like this:

  has 'headers'  => (
      is      => 'rw',
      isa     => 'Header',
      default => sub { HTTP::Headers->new } 
  );

We would then have a self-validating accessor whose default value is 
an empty instance of B<HTTP::Headers>. This is nice, but it is not 
ideal.

The constructor for B<HTTP::Headers> accepts a list of key-value pairs
representing the HTTP header fields. In Perl, such a list could 
easily be stored in an ARRAY or HASH reference. We would like our 
class's interface to be able to accept this list of key-value pairs 
in place of the B<HTTP::Headers> instance, and just DWIM. This is where
coercion can help. First, let's declare our coercion:

  coerce 'Header'
      => from 'ArrayRef'
          => via { HTTP::Headers->new( @{ $_ } ) }
      => from 'HashRef'
          => via { HTTP::Headers->new( %{ $_ } ) };

We first tell it that we are attaching the coercion to the 'Header'
subtype. We then give it a set of C<from> clauses which map other 
subtypes to coercion routines (through the C<via> keyword). Fairly 
simple really; however, this alone does nothing. We have to tell 
our attribute declaration to actually use the coercion, like so:

  has 'headers'  => (
      is      => 'rw',
      isa     => 'Header',
      coerce  => 1,
      default => sub { HTTP::Headers->new } 
  );

This will coerce any B<ArrayRef> or B<HashRef> which is passed into 
the C<headers> accessor into an instance of B<HTTP::Headers>. So the
the following lines of code are all equivalent:

  $foo->headers(HTTP::Headers->new(bar => 1, baz => 2));
  $foo->headers([ 'bar', 1, 'baz', 2 ]);  
  $foo->headers({ bar => 1, baz => 2 });  

As you can see, careful use of coercions can produce a very open 
interface for your class, while still retaining the "safety" of 
your type constraint checks.

Our next coercion takes advantage of the power of CPAN to handle 
the details of our coercion. In this particular case it uses the 
L<Params::Coerce> module, which fits in rather nicely with L<Moose>.

Again, we create a simple subtype to represent instances of the 
B<URI> class:

  subtype 'Uri'
      => as 'Object'
      => where { $_->isa('URI') };

Then we add the coercion:

  coerce 'Uri'
      => from 'Object'
          => via { $_->isa('URI') 
                    ? $_ 
                    : Params::Coerce::coerce( 'URI', $_ ) }
      => from 'Str'
          => via { URI->new( $_, 'http' ) };

The first C<from> clause we introduce is for the 'Object' subtype. An 'Object'
is simply any C<bless>ed value. This means that if the coercion encounters
another object, it should use this clause. Now we look at the C<via> block.
First it checks to see if the object is a B<URI> instance. Since the coercion
process occurs prior to any type constraint checking, it is entirely possible
for this to happen, and if it does happen, we simply want to pass the instance
on through. However, if it is not an instance of B<URI>, then we need to coerce
it. This is where L<Params::Coerce> can do its magic, and we can just use its
return value. Simple really, and much less work since we used a module from CPAN
:)

The second C<from> clause is attached to the 'Str' subtype, and 
illustrates how coercions can also be used to handle certain 
'default' behaviors. In this coercion, we simple take any string 
and pass it to the B<URI> constructor along with the default 
'http' scheme type. 

And of course, our coercions do nothing unless they are told to, 
like so:

  has 'base' => (is => 'rw', isa => 'Uri', coerce => 1);
  has 'uri'  => (is => 'rw', isa => 'Uri', coerce => 1);

As you can see, re-using the coercion allows us to enforce a 
consistent and very flexible API across multiple accessors.

=head1 CONCLUSION

This recipe illustrated the power of coercions to build a more 
flexible and open API for your accessors, while still retaining 
all the safety that comes from using Moose's type constraints. 
Using coercions it becomes simple to manage (from a single 
location) a consistent API not only across multiple accessors, 
but across multiple classes as well. 

In the next recipe, we will introduce roles, a concept originally 
borrowed from Smalltalk, which made it's way into Perl 6, and 
now into Moose.

=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