What I've been thinking about doing is hijacking the pseudo-hash syntax and using it for this. So if you did

my Animal $spot = Dog->new;
$spot->Reproduce; # fine
$spot->DriveToWork # not OK
some_other_function_that_returns_an_object()->method # ignore this
[download]

I was thinking of using the B framework to get access to the parsed code. In a very simple implementation you would wander around the parsed code, looking for methods called directly on a variable, find out that the variable is a lexical declared as my Animal and end up checking Animal->can("Reproduce") and Animal->can("DriveToWork"). In a more sophisticated implementation, instead of using can, the method information would be held in say $Animal::INTERFACE and could also include info about method arguments or other useful stuff.

Objections:

• AUTOLOAD methods. These are OK as long as you can supply the list of possible valid methods. If you can't provide that list then static checking is probably not suitable for that class anyway. Even if the list of methods is large/infinite you could always provide a function that says yes or no to a particular method name.
• Things like some_function()->Reproduce. To which I say, if you want real static checking, don't use Perl (5).

I had a quick look at B last weekend and it looks like it should be possible by navigating through a B tree of opcodes looking for the method calls but it could be a lot of work to write the code because you need to cater for each opcode (athough most of the time you're just ignoring them I think).

If anyone else fancies a challenge...