Insight:
The more fine-grained the predicates, the better the concerns
can be separated.

traversal function t determined by:
S={A->B B->D A->C C->D}

if method-name = t and
  if following S1={A->B B->D}
     and before D {f1(S1.A, S1.B, S1.D}
  if following S2={A->C C->D}
     and before D {f1(S2.A, S2.C, S2.D}
  if following S2={A->C C->D}
     and before C {f3(S2.A, S2.C)}
  if following S2={A->C C->D}
     and around C {if (cond) {f3(S2.A, S2.C); subtraversal.apply(); xx}}

==========
Hi Doug:

> what to do, and when to do it

> The degree of flexibility that
the programming language allows in specifying the second part
determines how well behavioral concerns can be separated.

> By allowing
more precise declaration of the "when to do it" part of a behavioral
unit, we remove the tangling that would otherwise occur if the
dispatch had be implemented explicitly.

This is worth elaborating!

In DJ: 

what to do is described by the methods of an adaptive visitor.

when to do it is described by a traversal strategy. We can view
the traversal strategy as a set of predicates for all the
visitor methods.

What kind of tangling would occur if "the
dispatch had be implemented explicitly"?

Can you provide other examples that support your claim?
Maybe an AspectJ example?

> The degree of flexibility that
the programming language allows in specifying the second part
determines how well behavioral concerns can be separated.

Is this an important design goal for AOP programming languages?

Compare 
	what to do		when to do it
COOL: 	method managers		exclusion sets
RIDL:   ?			?
DJ:	visitor methods		traversal strategy
AspectJ methods, advice		oo method dispatch, point cut designators

-- Karl