Ralph Johnson and I descibed some work we did to allow more dynamic,
per-instance
changes to objects so that they could act more like prototypes
many moons ago in Smalltalk itself in
http://laputa.isdn.uiuc.edu/ref89/ref89.html
at OOPSLA '89.

In those days, there was no changeClassToThatOf:,
so we had to <add> a class change primitive ourselves.
We also added a bit to the object headers to force a doesNotUnderstand: like
exception
(dispatchMessage:) at the start of the lookup process instead of the end.

Once you can clone yourself quickly, and change your parent/class pointer,
the one thing left is adding a method or variable slot.  You can do that by
adding a LightWeight
class (a.k.a. HeavyweightObject or MetaObject), or with instance "clone and
becomes:" tricks.
Adding a lightweight Behavior gives you a place to put methods or state.

Extensible dynamic instance variables can also be simulated using
dictionaries (http://laputa.isdn.uiuc.edu/dfc/discussion#Accessible)
and doesNotUnderstand:.

(The change class and dispatch hacks
were patched in in binary, with MPW C , the assembler, and the debugger.
We got a lot of bits under our fingernails...)

Obligatory Squeak Content:  these sorts of VM alterations are a <lot> easier
under Squeak.

The thing that was interesting about how Self did dynamic slots was the
implicit
meta-architecture under the hood (the maps and all) that made it possible to
do these dynamic
things, and yet run efficiently when you knew you weren't doing them.

Since Self's VM was a big C++ program,
making this part of the system first-class objects wasn't something they
were too enthusiastic about.  I was always frustrated that you
couldn't get your hands on these too.

I sense considerable enthusiasm for fleshing out this part of the
architectural
tale in this community: the objects that connect the object models and
implementation.  These can make experimentation with a variety of models
possible.  In any case, it's good to see the MOP hackers getting
flushed out into the open...

As for Mike's point about dangerous facilities, I'm for refactoring
things in such a way as to give us the dynamic power to make
such facilities easier to hide.

--BF