I am thinking that
be common code which is then branched off to architecture specific stuff. That
way operations can be checked and shared accordingly. Otherwise I will have to
implement checks for every single instance. I can also hide the internal stuff
through the package system and such.
This will help reduce the duplicated code when the wrapping public interface changes. Easier to maintain this way since it can just be a symlink.
This is a good thing. Because the kernel's internal class will have to be different from sub-processes internal classes. So when a process is linked in and initialized, the Internal will point to a completely different class implementation. Since some kind of message passing or system calls could be used in these cases. For example the creating of instances would need the kernel to initialize a new object for that processs but not in its own kernel.
One thing that can be done for table reads is I can commonize it. I would also need a temporary register. So two new methods. Then I can also commonize the invoke by having it where execution is jumped off the register.