Hi Doug and Johan:

I think we should switch as soon as possible
to the new compilation algorithm for the following reasons:

It changes the semantics and we don't want semantic changes
between beta and the final release. The already implemented method generation
approach should be available as an option for line strategies.
Selectable by the configuration object.

Only the new semantics/algorithm gives us a general solution even
for operations such as intersection and merge.

We have now thought long enough about the new semantics/algorithm
that I don't expect improvements over the next 2 years to the basic
concepts.  It is the right time to implement.

It is good to implement the algorithm while Boaz is still here.

Since the new algorithm is very much in the spirit of TAO (the
traversal graph is interpreted by the generic Traversal function), 
I think that Johan should implement the new algorithm as soon as possible?

=============

I think that the Fox implementation should be transformed into the new
algorithm.

The paper works with a special kind of flat class graphs,
called simple class graphs. I think it would be good
to use simple, or at least flat class graphs in the new implementation
since it simplifies everything without too much cost, I think.
What is your take on this?

-- Karl