Hi Binoy and Doug:

>From dougo@ccs.neu.edu Fri Aug  1 16:49:15 1997
>From: Doug Orleans <dougo@ccs.neu.edu>
>To: Binoy Samuel <binoy@ccs.neu.edu>
>Cc: Karl Lieberherr <lieber@ccs.neu.edu>, dem@ccs.neu.edu
>Subject: Re: visual developmentof strategies
>
>Binoy Samuel writes:
> > > we need an elegant way to specify strategy graphs interactively in the context
> > > of a class graph. We define the edges of strategy graphs by clicking on
> > > from and to nodes in the class graph. Identical nodes in the
> > > class graph correspond to identical nodes in the strategy graph.
> > > 
> > 
> > Shouldnt it be the other way round? Strategy graphs come first, and then
> > any class graph satisfying it? And maybe a name-mapper if there is no
> > one-to-one correspondence.
>
>Both use cases should be supported: (A) build a strategy graph from
>scratch, and (B) build a strategy graph from an existing class graph.
>Case (A) is a lot like building a class graph without alternation
>classes or labeled edges, plus constraints.
>
>--Doug
>

I agree with Doug that both use cases are useful:
We have two pure use case for Demeter/Java:

BOTTOM-UP:
object graphs -> class graphs -> strategy graphs

TOP-DOWN:
strategy graphs -> class graphs -> object graphs

In reality we will use a mixture of the two; I tend towards
bottum-up. Note that the "object graphs -> class graphs" step
can be automated (see Paul Bergstein's thesis: class graph
learning).

-- Karl