This and other RFCs are available on the web at http://dev.perl.org/rfc/ =head1 TITLE Class Methods Introspection: what methods does this object support? =head1 VERSION Maintainer: Mark Summerfield <[EMAIL PROTECTED]> Date: 28 Sep 2000 Last Modified: 29 Sep 2000 Mailing List: [EMAIL PROTECTED] Number: 335 Version: 2 Status: Frozen =head1 NOTES ON FREEZE In view of the lateness of my submission I've only had one response, from Piers Cawley; we've exchanged several emails and Piers' suggestions and code has been incorporated. There are no issues to resolve. This RFC does not depend on any other RFC. =head1 ABSTRACT This RFC proposes a new UNIVERSAL method which would be used thus: my @method_names = $object->methods(); # OR MyClass->methods(); The C<methods> call would return the public methods of the C<$object> or class; calling with the argument 'all' would return all the public methods that the C<$object> or class could respond to. =head1 DESCRIPTION How can we tell which methods an object supports? If we know the names of the methods we're interested in we can use C<$object-E<gt>can('methodname')>. However if we don't know the possible names then we're stuck. This RFC proposes a new UNIVERSAL method which would be used thus: @method_names = $object->methods(); # OR MyClass->methods(); @method_names = $object->methods('all'); # OR MyClass->methods('all'); The first example returns the public methods of the C<$object> or class itself; the second example returns any public methods that the C<$object> or class could respond to, i.e. all the methods of the class plus those of any classes it inherits from. The C<methods()> method may I<not> be able to determine definitively all the methods that are available: firstly it is perfectly possible to define new methods at runtime using C<eval>; secondly if the class has an C<AUTOLOAD> method then the methods it supports may not be determinable. In the case of classes that use C<AUTOLOAD> the string 'AUTOLOAD' would naturally be one of the method names returned so at least the programmer would be aware that any number of other methods may exist. Thus C<methods> would return the list of methods known to be available to the object at the time that C<methods> is called. Piers Cawley suggested this: "I would say that in an OO context there's a case for walking the @ISA tree to find out all the methods that the class will respond to, say C<$object->all_methods>, after all, in a complex class hierarchy there may be classes that implement only one or two methods that are distinct from the methods of their parent class." I think that Piers extension is a valuable one, but propose that it should be achieved by having an optional argument, 'all' to C<methods> rather than a separate new method; this ensures the minimum addition to UNIVERSAL and also leaves open the prospect of additional or different parameters being added in future. C<methods> could of course be over-ridden on a class-by-class basis; for example, if your class supported fifty methods, but half of them were so rarely needed that they were created on-the-fly in AUTOLOAD you could return all their names if you override C<methods> and return the list of names yourself. Furthermore as a method of UNIVERSAL C<methods> would not collide with any existing or subsequent function of that name. In terms of RFC 188 I would expect C<methods> to return public methods only. Why not simply use C<isa> to determine the class? Often knowing the class is all that is necessary; however sometimes it is useful to know what the possible methods are. Imagine we have a simulation system with a core application that sends and receives signals to the objects it is simulating (by calling their methods). Users may implement the objects they want simulated by creating their own classes for each type of object. One approach to this is to supply the user with a base class from which they inherit. This has a disadvantage -- a given method is always called in response to a given event because each event in the simulator maps to a method in the base class. Sometimes we want to "wire up" the simulated object so that on a particular run event A calls method A, but on another run event A calls method B. Hardwiring this isn't a problem, but what if we want to provide the user with this choice interactively -- well, if we don't know what methods the class supports in the first place we cannot give the user a list to choose from. The solution is to be able to introspect on the class they've chosen and show them the list of methods available; they can then choose which to wire up to which events in the simulator. I am sure that there are other situations where object introspection would be useful, for example a class browser. =head1 IMPLEMENTATION I think this needs to be implemented in Perl itself so that it can avoid private methods and can be implemented efficiently, but here's some code which provides a combined implementation and example: #!/usr/bin/perl -w # filename: test.pl # # Unlike a built-in this method cannot distinguish between `real' # methods and `constants' defined with use constant (but hopefully # that pragma will go away). # We skip private methods by ignoring names that begin with an # underscore; we skip overloaded methods by ignoring names that # begin with an open parenthesis. use strict; package BaseClass; sub baseclass {} package DerivedClass1; use base 'BaseClass'; sub derived {} sub one {} package DerivedClass2; use base 'BaseClass'; sub derived {} sub two {} package BabyClass; use base qw(DerivedClass1 DerivedClass2); sub three {} package main; print "BaseClass: ", join(", ", sort BaseClass->methods()), "\n"; print "BaseClass (all): ", join(", ", sort BaseClass->methods('all')), "\n"; print "DerivedClass1: ", join(", ", sort DerivedClass1->methods()), "\n"; print "DerivedClass1 (all): ", join(", ", sort DerivedClass1->methods('all')), "\n"; print "DerivedClass2: ", join(", ", sort DerivedClass2->methods()), "\n"; print "DerivedClass2 (all): ", join(", ", sort DerivedClass2->methods('all')), "\n"; print "BabyClass: ", join(", ", sort BabyClass->methods()), "\n"; print "BabyClass (all): ", join(", ", sort BabyClass->methods('all')), "\n"; package UNIVERSAL; use strict; sub methods { my ($class, $types) = @_; $class = ref $class || $class; $types ||= ''; my %classes_seen; my %methods; my @class = ($class); no strict 'refs'; while ($class = shift @class) { next if $classes_seen{$class}++; unshift @class, @{"${class}::ISA"} if $types eq 'all'; # Based on methods_via() in perl5db.pl for my $method (grep {not /^[(_]/ and defined &{${"${class}::"}{$_}}} keys %{"${class}::"}) { $methods{$method} = wantarray ? undef : $class->can($method); } } wantarray ? keys %methods : \%methods; } 1; Here's the output: # ./test.pl BaseClass: baseclass BaseClass (all): baseclass DerivedClass1: derived, one DerivedClass1 (all): baseclass, derived, one DerivedClass2: derived, two DerivedClass2 (all): baseclass, derived, two BabyClass: three BabyClass (all): baseclass, derived, one, three, two You can test it on any class you like: # perl -de0 main::(-e:1): 0 DB<1> require 'test.pl' [snipped the tests which get executed] DB<2> use CPAN; DB<3> p join ", ", sort CPAN->methods() AUTOLOAD, DESTROY, all, all_objects, checklock, cleanup, cwd, delete, exists, fill, find, finddepth, getcwd, has_inst, instance, new, shell, wrap DB<4> p join ", ", sort CPAN->methods('all') AUTOLOAD, DESTROY, all, all_objects, checklock, cleanup, cwd, debug, delete, exists, export, export_fail, export_ok_tags, export_tags, export_to_level, fill, find, finddepth, getcwd, has_inst, import, instance, new, require_version, shell, wrap Piers suggested that the C<methods> method should return an array of method names in a list context or a reference to a hash of C<method_name =E<gt> method_coderef> pairs in a scalar context; this has been incorporated. Piers also rewrote the traversal so that it is done in the correct order, i.e. the same way that Perl itself searches for methods. =head1 REFERENCES RFC 188: Objects : Private keys and methods RFC 193: Objects : Core support for method delegation RFC 265: Interface polymorphism considered lovely perl5db.pl =head1 ACKNOWLEDGEMENTS Piers Cawley for reworking the example implementation and explaining why it needed reworking! I<The Librarians> for doing great work!
