Class::SelfMethods - a Module for supporting instance-defined methods

  use Class::SelfMethods;

  package MyClass;
  @ISA = qw(Class::SelfMethods);
  use strict;

  sub _friendly {
    my $self = shift;
    return $self->name;
  }

  package main;
  no strict;

  my $foo = MyClass->new( name => 'foo' );
  my $bar = MyClass->new( name => 'bar', friendly => 'Bar');
  my $bas = MyClass->new( name => 'bas',
                          friendly => sub {
                            my $self = shift;
                            return ucfirst($self->_friendly);
                          }
                        );

  print $foo->friendly, "\n";
  print $bar->friendly, "\n";
  print $bas->friendly, "\n";

  $bas->friendly_SET('a reset friendly');
  print $bas->friendly, "\n";

  $bas->friendly_SET( sub { my $self = shift; return uc($self->_friendly) });
  print $bas->friendly, "\n";

  $bas->friendly_CLEAR;
  print $bas->friendly, "\n";

Development of this module has largely lapsed due to the superior performance and feature set of Class::Prototyped. If you haven't written code that depends upon Class::SelfMethods, I strongly urge you to look at Class::Prototyped first.

Class::SelfMethods merges some features of other Object Oriented languages to build a system for implementing more flexible objects than is provided by default in Perl.

The core features I was looking for when I wrote Class::SelfMethods were:

Class-based inheritance hierarchy

I wanted to retain Perl's normal class-based inheritance hierarchy rather than to write (or use) a completely prototype based system. If you are looking for a purely prototype based system, see Sean M. Burke's Class::Classless. My reasoning on this is that it is easier in file based languages (as opposed to world based languages like Self) to code class based inheritance hierarchies (which are largely static) than to code object based inheritance hierarchies (since objects in such languages have a dynamicism that is not granted to classes).

Instance-defined method overriding

I wanted instances to be able to override their class-defined methods. In the example above, the $bas object has its own friendly method. Instance-defined methods are passed the exact same parameter list as class-defined methods.

Subroutine/Attribute equivalence

Borrowing from Self, I wanted to be able to treat methods and attributes similarly. For instance, in the above example the $bar object has an attribute friendly, whereas the $bas object has a method friendly, and the $foo object uses the class-defined method. The calling syntax is independent of the implementation. Parameters can even be passed in the method call and they will simply be ignored if the method is implemented by a simple attribute

In addition to those core features, I (and Damian) had a wish list of additional features:

Simple syntax

I wanted the system to be reasonable easy to use for both implementers of classes and users of objects. Simple syntax for users is more important than simple syntax for implementers.

Full support for SUPER type concepts

I wanted instance-defined methods to be able to call the class-defined methods they replace.

Support for calling methods at instantiation time

In some circumstances, rather than deal with multiple inheritance it is easier to have a class-defined object method that sets up the various instance-defined methods for a given object. To support this, the new method allows deferred method calls to be passed in as parameters.

Modifying objects post-instantiation

I originally had no need for modifying objects post-instantiation, but Damian Conway thought it would be a Good Thing (TM) to support. Being so very good at these sorts of thing, he instantly came up with a good general syntax to support such. Method calls that end in a _SET result in the first parameter being assigned to the attribute/method. I noticed one remaining hole and added support for _CLEAR.

Your class should inherit from Class::SelfMethods. The class-defined instance methods should be defined with a leading underscore and should be called without a leading underscore. Don't do anything silly like writing methods whose proper names have a leading underscore and whose definitions have two leading underscores - that's just asking for trouble.

Do not, of course, make use of attributes that have leading underscores - that's also just asking for trouble. Also, do not access attributes directly (i.e. $self->{foo}). That will prevent people who use your class from substituting a method for an attribute. Instead, always read attributes by making the corresponding method call ($self->foo).

If you need to call SUPER::methodname, call SUPER::_methodname.

The default new method uses named parameters. Unless you are certifiable, you will too. To specify attributes, simply use the syntax name => 'value' and to specify a method use name => sub { my $self = shift; . . . }. Note that methods and attributes are interchangeable.

Method calls that end in a _SET will result in their first parameter being assigned to the appropriate attribute/method. For instance, in the SYNOPSIS I use $foo->friendly_SET to specify both a value and a method for friendly. Method calls that end in a _CLEAR will delete that attribute/method from the object. The can method will behave just like UNIVERSAL::can - it returns a code reference that will interoperate with the associated object properly using the $obj->$coderef() syntax. For examples of usage, see test.pl.

Standard module installation procedure.

This implementation of can is the heart of the system. By making can responsible for almost everything relating to accessing the objects, the code for deciding how to respond to the various situtations is kept in one place.

In order to get major speed improvements (a factor of 2 to 3 for attribute retrieval and method calls), extensive symbol table manipulation was used to build methods on the fly that react appropriately.

The three types of methods are _SET methods, _CLEAR methods, and "normal" methods. The first two are fairly straight forward as far as implementation goes. First UNIVERSAL::can is called to determine whether an appropriate entry has been made in the package symbol table. If not, an anonymous subroutine (actually, a closure in this case because $func is a lexically scoped variable defined outside the anonymous subroutine and referenced from within) is created and assigned into the package symbol table. In either case, a reference to the appropriate closure is returned (normal can behavior is to return a reference to the code or undef if the method call is not legal).

The "normal" methods are somewhat trickier. The outer if statement exists to ensure that can returns undef for illegal method calls (remember that there may be situations where $self->can($func) should return false even though UNIVERSAL::can($self, $func) returns true). It then checks whether an appropriate entry has been made in the package symbol table. If not, it builds a closure that will do the trick. Remember that the closure could get called on an object that is in any of the four possible states - attribute, instance method, inherited method, or illegal. The closure includes the logic to test for instance methods and attributes, but if neither are present it will make the call to _method regardless of whether or not there is an inherited method with the proper name. It relies on AUTOLOAD to properly deal with unhandled _method calls.

AUTOLOAD gets called the first time a given method call is made. It first strips off the package name from the function call to extract the actual function name. It then checks to see if the function name starts with an underscore. If it does, it's a failed call from the "normal" method closure, so AUTOLOAD calls croak to die with the appropriate error message. Notice that the underscore has been stripped off, so it will die failing to find method.

AUTOLOAD then calls can, which will return a reference to the appropriate CODE entity if the method call is supported. At the same time, can puts an entry into the symbol table for Class::SelfMethods to support future calls to that method. AUTOLOAD jumps to that CODE entity if a valid entity was return. Otherwise, execution continues on to another croak call.

The new method supplied in Class::SelfMethods provides one interesting twist on an otherwise standard named parameters constructor. It strips out any passed parameters that have leading underscores and stores them away. It then creates the hash ref from the remaining parameters and blesses it appropriately. Finally, it takes the stored parameters that have leading underscores and makes the matching method calls - the key is used for the method name and the value is dereferenced to an array and passed as parameters.

Toby Ovod-Everett, tovod-everett@alascom.att.com

Damian Conway, damian@cs.monash.edu.au

Responsible for accessor methods, module name, constructive criticism and moral support. After I responded to Sean's suggestion of implementing a can method, Damian completely rewrote my first attempt by routing everything through can. He also was the first to point out direct symbol table manipulation by implementing it for the _SET and _CLEAR methods. I rebutted his routing everything through can by doing performance testing. He agreed that the performance issues were a problem, but suggested retaining the direct symbol table for the accessor methods. It was then that the lightbulb went off and I realized that a properly written closure could be used for the normal method calls. Damian's criticisms kept me on track and from making a fool of myself, and the result is some very fast (and I hope safe:) code.

I first started writing to Damian as a result of an excellent book he wrote, Object Oriented Perl. I highly recommend it - get it, read it.

Sean M. Burke, sburke@netadventure.net

Suggested implementing a can method. Sean was/is responsible for Class::Classless. If you need a full-featured purely prototype based object system, check it out.