In other words, I don't want this:

package A::Class;

sub some_method
{
    my ($self, $object) = @_;
    die "Object is not what I expect"
        unless UNIVERSAL::isa( $object, 'Some::Class' );
    $object->some_other_method();
    $object->some_method_or_other();
}
[download]

That unnecessarily dictates that $object has to inherit from Some::Class, or at least fake up enough so that it can fool UNIVERSAL::isa(). Consider a test case:

use Test::More 'no_plan';

use Test::MockObject;
my $mock = Test::MockObject->new();

use_ok( 'A::Class' );
my $a = A::Class->new();
isa_ok( $a, 'A::Class' );

$mock->set_true( 'some_other_method' )
     ->set_true( 'some_method_or_other' );

$a->some_method( $mock );
[download]

That dies, unnecessarily, because Test::MockObject does not subclass Some::Class. That's not what I want to express, either.

What I really want to express is that $object, whatever it is, ought to act like a Some::Class object. If it inherits from Some::Class, that's fine. If it delegates to a Some::Class object, that's fine. If it contains a Some::Class object, that's fine. If it simply conforms to the Some::Class interface, that's fine.

I don't think "reworking the object model" to make the other three possibilities fit into the inheritance scheme is the right approach. I think the interface approach is more general and, if implemented correctly, can handle all four possibilities easily.

I do think my examples were unclear though, and apologize for the confusion they caused.