Hi Mira:

I have written down some ideas probably relevant to our paper.
For example, how to handle strategy types.

-- Karl

based on your picture, it looks like we need the following concepts:

strategy types
  defines family of strategies
strategies
path-set equivalence
  two strategies are path-set equivalent if for all class graphs they
  define the same path set
path-set extending
  s1 is path-set extending, if for all class graphs, s1 defines a larger
  path set than s2.

adjuster types
  defines interface to adjuster
adjusters
  a reusable microarchitecture?
  parameterized by strategy type

  compose adjusters: an adjuster may modify at most one other adjuster
  Each adjuster adds more behavior. Incremental behavior composition.

class graph types
  defines interface to class graph
class graphs
object equivalence
object extending



A strategy type defines a family of strategies. A strategy type
is defined by a strategy st. A strategy s conforms to type st,
if st selects the entire strategy graph s viewed as a class graph.

How to deal with constraint maps?

Example:
st = from A via B to C
s  = from A via X via B via Y to C.
class graph corresponding to s:

A=X.
X=B.
B=Y.
Y=C.

A class graph type defines an interface to a class graph.
A class graph type is defined by a class graph gt.
Some of the edges are functional edges expressed in terms 
of a lower level class graph.

Methods are expressed in terms of strategy type instantiations (=
strategies) and adjuster compositions. Finally, class graphs freeze
the programs.