Hi,

I read your book because once we dreamed about extracting roles from 
dynamic trace of program. We succeeded partly but I felt always sorry 
that I pushed this topic because this was really difficult. As most of 
the time
people talk about role in abstract manner and never described the key 
information : the protocol in terms of interaction the roles bring to 
the object. Anyway....

> I'm new to Squeak but old to Smalltalk (since 1978). My retirement 
> project is definitely "blue plane", and its scope is expanding at a 
> frightening rate.
This is good.

> The short form: "What should the information systems of the 21st 
> century look like?"

Good question
Especially when we see google now...

> The ideas are mainly from
>  - Smalltalk "everything is represented as an object"
>  - UML, with its meta-levels. An application model is an instance of 
> UML,
>       and UML is an instance of MOF.
>  - The UML specification defines a complex class hierarchy. This 
> hierarchy
>       is simply a comment to the real substance; which is the objects 
> that
>       are instances of the defined classes.

The problem with UML as with any meta-meta model in general is that it
is fairly easy to have a interpreter for a model in no time (one week) 
for the structural aspect
ie this class has attributes (and providing all the query/navigation 
method). I did one (take ER
to describe UMl like model and generate smalltalk code, easy but not 
satisfying).

Now the way to describe the behavior is the key point. Some people use 
OCL like stuff, other Condition
Event Action or semantics action. But in most of the case you are 
limited and end up having a limited language
with your interpreter.

I found the paper in OOPSLA 98 on the Universal UML virtual machine 
depressing (even if I appreciate the work of dirk in general) this is 
too easy to generate the structural aspects and let to the programmer 
to code the behavior in skeleton.

In such a case would not be more interesting to have a Smalltalk like 
meta-model? Said differently what is the next generation of Smalltalk 
that would be much simpler? Does this thing exist?

>  - Richard Pawson's Naked Objects project where information is 
> presented to
>       the user as tangible objects.
>
> and much more. Why should our meta-meta language restrict us to 
> punched card input and line printer output, i.e., BNF.
>
> But the main point is that we have to leave the old life cycle:
> "Code-Compile-Load-Run-Stop". The Smalltalk idea of a continuously 
> executing world of objects is clearly superior. (I belive the 
> execution thread of my current Squeak image started some time back in 
> 1972).

Indeed.

> I have done some experiments with an executing UML where I modified 
> the Smalltalk classes, meta-classes and meta-meta-classes to conform 
> to the UML architecture. Confusing, but informative.
> See 
> http://heim.ifi.uio.no/~trygver/UML2.0-U2P/UML_VirtualMachine-131.pdf
>
> The morphic architecture looks like an important step forward towards 
> tangible objects. It breaks with the ST-80 MVC, but does not seem to 
> greatly invalidate my original MVC from 1978. See the original 
> proposal here: http://heim.ifi.uio.no/~trygver/mvc/index.html

But what is frustrating in Morph to me is that you mix the 
representation of an object and its visual aspect.
I'm not sure that I know what a unificationTerm, process, a tree 
walking, an enumerator
  looks like. The question is that visualisation is important but should 
it be generalized to any object. I think not.

Stef

http://www.iam.unibe.ch/~ducasse/
  "if you knew today was your last day on earth, what would you
  do different? ...  especially if,  by doing something different,
  today might not be your last day on earth" Calvin&Hobbes