stéphane ducasse wrote:

> On 4 avr. 04, at 10:50, Lothar Schenk wrote:
> > stéphane ducasse wrote:
> >> ...the idea was you have multiple Smalltalk
> >> environments (difficult to express again)....
> >
> > Mmm, that's something I've been idly speculating about. I don't know
> > if this
> > holds up to the scrutiny of experts, but has anybody ever considered
> > the
> > possibility of splitting up a Smalltalk system at the image level? It
> > is, a
> > Smalltalk system should be comprised of multiple images, not just one.
>
> No just in term of symbol. from within a classbox I only see the symbol
> (class and variables)
> I defines, or I imported from other. So in each classbox I have the
> impression of a flat world
> but you can have different landscape.
>
> Stef

As far as I understand this, you are essentially building what I think of as 
virtual images by using classboxes. Since you have to accomodate them within 
a single "real" image, it follows that your development tools must be 
specially adapted to handle interoperability and noninterference issues 
within the context of one image. That's a possible benefit I see by making 
such virtual images into "real" images. In principle, you should be able to 
navigate such an image using the traditional tools. Noninterference would 
result automatically by virtue of having separate images. Version control and 
interoperability issues would have to be handled by additional tools whose 
conceptual scope would be outside of a particular image (implying a versioned 
subsytem/package/class load and unload mechanism - a la Squat, maybe), thus 
making them an inter-image issue instead of an intra-image one, which in my 
thinking might possibly lead to a better seperation of concerns.

But, again, I must state that I have no in-depth grasp of the intricacies and 
inherent difficulties in implementing such a scheme. It's just a hazy idea 
I'm throwing out in the hopes that it might give someone somewhere a new idea 
to chew on.

Regards, Lothar