I'm currently designing and implementing a native code compiling forth "interpreter" and building an exokernel on top of it. The kernel will be something like a secure forth virtual machine.
Shared libraries, etc. will not be used. Instead I will modularize the system by creating namespaces with forth dictionaries implemented as linked lists. I can link the dictionary lists in a hierarchial way so that local dictionary entries can be used to provide data hiding. I'm also considering making this "forth" a typed language so that I could have both "objects" and polymorphism.
I'm trying to move as much code as possible outside the kernel. To do that I'm considering the possibility that kernel might implement only the most basic primitives needed to access the hardware, control access to those primitives and cache security data. Security (and the stuff that must be known as it can be executed only in the kernel) will be the only thing the kernel knows about. Things such as device drivers may possibly be replaced by code passed to the kernel in order to be executed. The kernel will no longer know what to do or how to do it however it will know and control what a thread or a process is allowed to do. All of the security related data remains strictly within the kernel and is not exposed to anyone.
The keywords are simplicity and safety. I'm calling the project "fxk" at the moment.
Any comments???
- Esa