Yeap I agree with point c)

> d) One of the reasons why I think type inferencers work so poorly is 
> because
> of the lack of strong enough models in the underlying system. "Just
> guessing" about messages being sent here and there isn't going to work 
> in my
> understanding - any type inferencer which is worth its money probably 
> will
> need some "hints" by framework designers so that users of the 
> framework can
> have the type inferencer do its work. So some form of "type 
> annotations"
> will be required to make it work.

Yes this was the point in anymorphic, You could also give some 
information and based on that
the inferencer can work. Gilad mentioned that at the time they did 
animorphic they used a wrong
algorithm for inferencing but now he knows a better one (I forgot). But 
I keep that in mind.

In fact In VW (so may be in squeak too) the parser accepted name 
<String> for method
definition. We could simply think about the same for instance variables 
and local...
as soon as we are not forced to say it.

stef