Daniel Vainsencher <danielv at netvision.net.il> wrote:
> Just as a spectator - sounds to me like you're mixing several projects
> that could be orthogonal.
> - In-image VM (which I'm assuming could just as easily have current
> Squeak semantics)
> - Distributed Squeak
> - Yet Another Go At namespaces (YAGA ;-)
> All are fascinating, and each of them somewhat risky. Mixing them
> multiplies the risks...
> Does doing them at once make it all so much simpler that it's worth it?

I'm starting fresh, which I think will be easier than trying to modify
the existing Squeak VM.  Of course, I will copy code and algorithms
where I can.

Smalltalk semantics (message lookup and namespaces) is independent of
the bytecode level, and can be easily changed by changing the compiler. 
All compilers will generate Squid bytecode, which is low-level enough
(think assembly language) to support most high-level language semantics.
 I'm going to implement the kernel using the new Squid semantics
(modules with method delegation), because I think it will yield cleaner,
more modular, code.  A Squeak compiler that supports current Squeak
semantics can be added later if desired.

Segmentation, which will facilitate distribution, is at the low level
and will be implemented in conjuction with other VM level behavior.

So, I'm implementing the VM and segmentation together using Squid
modules and semantics.  That is why I'm doing it all together.
Dynamic optimization is one thing that can be done later.

Cheers,
Anthony