That (flyweight strategy) doesn't really solve the problem since you still end up destroying objects in a random order. It will help with multiple objects from the same module having a special order that is required (so long as you design your own way to keep track of this special order which may boil down to you reimplementing your own form of reference counting), but if you have two modules that reference each other in some general way, then using a END-based destructor is just as bad as Perl's global destruction problem (except that Perl's problem will be fixed soon).

You are right that when I think it through carefully there are plenty of cases where it doesn't work. But if you have objects in a class which are backed by available data that you want to get a chance to save to disk, well then protecting that class in a Flyweight pattern gives you an opportunity to allow the data to be saved correctly.

I whipped up a proof of concept at Class::FlyweightWrapper. Take a look at how that works then look at the root node of this thread. I believe that if GMS::MemberFile was made into a wrapper around a private implementation through my new module, then DESTROY would work as desired. Sure, you can find situations where it would break, but it is at least a solution to the problem at hand.