Sun Jan 24 23:38:24 CET 2010

Modifying VM1

Adding some simple optimizations might already improve it a lot:

- snap global variable pointers (-> cell)
- snap lexial variable indices (-> number)
- eliminate the STATE update allocation
- mark continuations as not captured to re-use frames

The basic idea is that the bootstrapping required for a compiler not
written in C is a pain.  If some bootstrapping is going to be
necessary anyway, it's probably best to use it to make implementing
the system easier, i.e.  raise the abstraction level of all the

     VM:         scheme -> C/EX
     COMP, LIB:  scheme -> VM

Conclusion about dybvig's stack-based implementation: it's probably
not worth the trouble: better to re-use another Scheme implementation.