Hi Doug:

thank you for the update. So Demeter/Java already allows API
definitions a la Persephone with the restrictions:

1. API methods may not have arguments.
2. no object transportation.
3. no shortest path semantics.

We all agree that object transportation and shortest path
semantics (as an option) are good ideas.

Regarding 1, I was referring to the Persephone methods that have
local variables which are used as arguments to the derived edges
that have arguments.

I would like to know how you judge that part of Persephone.
Should that be added to Demeter/Java or do you have a better idea.

-- Karl
=================

From dougo@ccs.neu.edu Mon Oct 26 12:52:43 1998
To: Karl Lieberherr <lieber@ccs.neu.edu>
Cc: jayantha@ccs.neu.edu, johan@ccs.neu.edu, lorenz@ccs.neu.edu,
        mira@informatik.uni-siegen.de, dougo@ccs.neu.edu
Subject: Re: derived edges/APIs

Karl Lieberherr writes:
 > Can Demeter/Java handle traversals through derived edges?

Yep.

 > Do you have an example somewhere?

Well, the new version of the tg package (err, the AP Library) (which
isn't checked in yet) has a derived edge to keep track of intercopy
edges incident with a traversal graph vertex:

TGVertex {
  (@
    private int_int_Pair_List real_intercopyEdges = new int_int_Pair_List();;
    private int trav = 0;
  @)
  
  get intercopyEdges (@
    if (trav != TraversalGraph.curTrav)
      set_intercopyEdges(new int_int_Pair_List());
    return real_intercopyEdges;
  @)

  set intercopyEdges (@
    real_intercopyEdges = dest;
    trav = TraversalGraph.curTrav;
  @)
}

Since the TraversalGraph object (which is a collection of TGVertex
objects) is re-used for all the traversals, the intercopy edges get
recalculated for each traversal.  But the above trick lets me only
calculate the intercopy edges for vertices that are in the strategy
graph, rather than the entire class graph; if the intercopyEdges part
hasn't been set for the current traversal (checked by looking at
curTrav, which is a global counter that's incremented for each
traversal computation) then just use an empty list.  While I don't
really do a traversal through this derived edge as part of the main
functionality, for debugging I have a toAll(UniversalVisitor)
traversal which does go through it, so i can (e.g.) print the
intercopy edges for all the vertices.

Hm, I thought there was another example in the suite-derived directory 
in the regression tests, but none of those tests actually traverse
through a derived edge.  I should probably add one or two.

 > What happens if a derived edge has arguments?

Currently you can't define a derived edge with arguments.

 > Do you think the Persephone solution is a good one that should be
 > added to Demeter/Java?

You mean object transportation?  Sure.  Also, the Persephone model of
generating a single method with a chain of getters is similar to what
we've been discussing as a good optimization (when possible, i.e. no
cycles or alternation edges in the traversal graph).  Actually, Johan
might have already done this in his one-class-per-traversal hacking?

--dougo@ccs.neu.edu