From: Xavier Noria <fxn@...>
Date: 2011-04-13T16:20:23+09:00
Subject: [ruby-core:35733] Re: caching of the ancestor chain

What's the downside of the naive implementation?

Each class has two slots: list of pointers to direct mixins (no flat
list, only the direct ones that were mixed in and in the correct
order) and a pointer to the parent class. Modules have a slot with the
list of direct mixins.

Method dispatch loops/recurses. We ask each mixin in turn whether he
responds to the method. At that point of execution, mixins ask
themselves, and if not loop over their direct mixins. Recurse. If all
of them fail, I go to the parent class and repeat. Of course the
complete algorithm would include method_missing, singleton classes...
but just to give the idea.

I have NO experience writing interpreters, is that a terrible
implementation? The number of pointers to follow seems to be about the
same more or less doesn't it... if it is bad why is it bad? Not
proposing anything of course (I can't propose anything I have no ida
about VMs :), only doing this as an exercise to try to understand why
it works the way it works.