[PRL] may be

[Mitchell Wand]

>>>>> On Wed, 5 Nov 2003 14:32:04 -0500, "Karl Lieberherr" <lieber at ccs.neu.edu> said:

KL> Hi Matthias:

KL> I did not have a chance to respond in person. Here is my fix to
KL> the "may-be" formalization:

KL> A B-object may be contained in a C-object iff the traversal specification
KL> [C,B] defines a non-empty set.

KL> The may-be relation is covered in my undergraduate class ((<=.C.=>)*.<=
KL> where . is relation composition). Maybe I misunderstood your statement that
KL> the may-be relation is NOT easy to formalize.

This can't possibly be right, because the model of objects you mention
doesn't distinguish between association and aggregation (or
composition).

And EA is fundamentally about aggregation:  it is inheritance down
aggregation links.

Will talked to David after the seminar, and concluded that may-be is a
parameter of the semantics.  By this he meant the following:

** The class graph includes must-contain, direct-subclass-of, and
   has-as-part edges (and maybe others).

** The language designer provides an algorithm for computing may-be
   from the relations in the class graph.

** The meaning of ACQUIRE is based on this (derived) may-be relation. 

So the programmer does not have to specify the may-be relation when he
writes down the class graph.

Whether a particular object graph is legal wrt a particular class
graph has nothing to do with may-be.  That's different from what I was
saying after the talk.

At least that's my understanding of what Will told me.

--Mitch