The Visitor Pattern

A Visitor is an object implementing some behaviour for a fixed set of classes, whose implementation would otherwise need to be scattered across those classes' implementations.

Visitors are usually combined with Iterators for traversing either a list or tree of objects.

A Visitor's methods are called using the so-called double dispatch technique. To allow double dispatching, the Visitor implements one method for every class to be handled, whereas every class implements just one method (commonly named "access"), which does nothing more than calling a method on the reference given, with the self object as parameter.

If all this sounds strange, maybe an example helps. Imagine you had a list of person objects and wanted to print out a list of their names (or address stamps or everything else you like). This can easily be implemented with a Visitor:

    package PersonVisitor;
    use Class::Std;    # handles all basic stuff like constructors etc.

    sub visit_Person {
        my ( $self, $object ) = @_;
        print "Person name is ", $object->get_name(), "\n";
    }

    package Person;
    use Class::Std;
    my %name : ATTR(:name<name> :default<anonymous>);

    sub accept { $_[1]->visit_Person( $_[0] ) }

    package main;
    my @person_from = ();
    for (qw(Gamma Helm Johnson Vlissides)) {
        push @person_from, Person->new( { name => $_ } );
    }

    my $visitor = PersonVisitor->new();
    for (@person_from) {
        $_->accept($visitor);
    }

    # will print
    Person name is Gamma
    Person name is Helm
    Person name is Johnson
    Person name is Vlissides

While using this pattern for just printing a list may look a bit over-sized, it may become handy if you need multiple output formats and different classes to operate on.

The main benefits using visitors are:

• Grouping related behaviour in one class

  Related behaviour for several classes can be grouped together in the Visitor class. The behaviour can easily be changed by changing the code in one class, instead of having to change all the visited classes.

• Cleaning up the data classes' implementations

  If classes holding data also implement several different output formats or other (otherwise unrelated) behaviour, they tend to get bloated.

• Adding behaviour is easy

  Swapping out the visitor class allows easy alterations of behaviour. So on a list of Persons, one Visitor may print address stamps, while another one prints out a phone number list.

Of course, there are also drawbacks in the visitor pattern:

• Changes in the visited classes are expensive

  If one of the visited classes changes (or is added), all visitors must be updated to reflect this change. This may be rather expensive if classes change often.

• The visited classes must expose all data required

  Visitors may need to use the internals of a class. This may result in fidelling with a object's internals, or a bloated interface in the visited class.

Visitors are usually accompanied by a Iterator. The Iterator may be implemented in the visited classes, in the Visitor, or somewhere else (in the example it was somewhere else).

The Iterator decides which object to visit next.

Why SOAP::WSDL uses the Visitor pattern for Code Generation

• Typemaps

  For every WSDL definition, a Typemap is created. The Typemap is used later as an aid in parsing the SOAP XML messages.

• Type Classes

  For every type defined in the WSDL's schema, a Type Class is generated.

  These classes are instantiated later as a result of parsing SOAP XML messages.

• Interface Classes

  For every service, a interface class is generated. This class is later used by programmers accessing the service

• Documentation

  Both Type Classes and Interface Classes include documentation. Additional documentation may be generated as a hint for programmers, or later for mimicking .NET's .asmx example pages.

All these behaviours could well be (and have historically been) implemented in the classes holding the WSDL data. This made these classes rather bloated, and made it hard to change behaviour (like supporting SOAP Headers, supporting atomic types, and other features which were missing from early versions of SOAP::WSDL).

Implementing these behaviours in Visitor classes eases adding new behaviours, and reducing the incompletenesses still inherent in SOAP::WSDL's WSDL and XML schema implementation.

Implementation

SOAP::WSDL::Base defines an accept method which expects a Visitor as only parameter.

The method visit_Foo_Bar is called on the visitor, with the self object as parameter.

The actual method name is constructed this way:

• SOAP::WSDL is stripped from the class name
• All remaining :: s are replaced by _

Example:

When visiting a SOAP::WSDL::XSD::ComplexType object, the method visit_XSD_ComplexType is called on the visitor.

Writing your own visitor

To write your own code generating visitor, you should subclass SOAP::WSDL::Generator::Visitor. It implements (empty) default methods for all SOAP::WSDL data classes:

• visit_Definitions
• visit_Binding
• visit_Message
• visit_Operation
• visit_OpMessage
• visit_Part
• visit_Port
• visit_PortType
• visit_Service
• visit_SoapOperation
• visit_Types
• visit_XSD_Schema
• visit_XSD_ComplexType
• visit_XSD_Element
• visit_XSD_SimpleType

In your Visitor, you must implement visit_Foo methods for all classes you wish to visit.

The SOAP::WSDL::Generator::Visitor implementations include part of their own Iterator (which means they know how to find the next objects to visit). You may or may not choose to implement a separate Iterator.

Letting a visitor implementing its own Iterator visit a WSDL definition is as easy as writing something like this:

 my $visitor = MyVisitor->new();
 my $parser = SOAP::WSDL::Expat::WSDLParser->new();
 my $definitions = $parser->parse_file('my.wsdl'):

 $definitions->_accept( $visitor );

If you need an iterator following the somewhat crude path of dependencies in a WSDL1.1 definition, you might want to look at SOAP::WSDL::Generator::Iterator::WSDL11.