Hi Johan:

You say:
 > a more restrictive
 > compatibility def DOES allow just a name map to be used.

What kind of graph relationship would that be?
Even under a refinement relationship as defined in:

http://www.ccs.neu.edu/research/demeter/biblio/graph-refine.html

you may still get spurios paths. The traversal must get
modified to reflect the original intent in the context of the
concrete class graph.

I agree with you that refinement eliminates some spurious paths
we don't want but not all of them.

I-Chen's project is about eliminating spurious paths:

Input:
Class graph G with strategies S1 and S2 where S2 is a
refinement of S1.
Input object O for class graph G.
Output:
Traversal of O according to both S1 and S2. (According to S2 would be enough
because S2 is a refinement of S1).

-- Karl

>From johan@ccs.neu.edu Wed Nov 11 16:06:34 1998
>To: generic@ccs.neu.edu
>From: Johan Ovlinger <johan@ccs.neu.edu>
>
>I use traversals as the motivating example. Imagine a traversal
>defined over the ICG. By "traversal" I mean the textual representation
>of the traversal- not the expanded paths. 
>
>I favor a very lax compatibility defintion.  This is better for views,
>as it increases flexibility, and the traversal will be expaned over
>the ICG anyway.  Ie, we traverse the ICG, but each edge of the ICG
>corresponds to several edges of the CCG.  The important distinction is
>that the traversal only sees the ICG.
>
>I could never see the point of more restrictive defs. Until now. 
>
>APPCs want to have their traversals expanded over the CCG. This is a
>good idea for efficiency, if nothing else.  But we have to be careful.
>Before expansion, the traversal must be translated from the ICG domain
>to the CCG domain in such a way that only paths that are compatible
>with the ICG traversal will be expanded.  If we have a lax CG
>compatiblity def, then just mapping the names will not work.  It is
>easy to come up with a ICG, CCG, and traversal that when just name
>mapped will introduce spurious paths (those who were at the seminar
>will recall the "no C w/o seeing a B first" rant I was on about).
>Instead, the traversal must be modified to reflect the original
>intent.  What I finally understood was that a more restrictive
>compatibility def DOES allow just a name map to be used.  Which is
>kinda useful. 
>
>Johan
>