Hi Boaz:

back to your remark of yesterday about a more adaptive programming style.

If we have the strategies

from A ... to B

and

from C ... to D

and in the class graph C is not mentioned at all,
it depends on what is in the adaptive program whether the
situation is fatal.

If the adaptive program contains a call to a method of C
(presumably a method related to the traversal from C to D)
then the Java compiler will stop. Therefore we should make
the situation fatal at the design level.

However, if C is not used at all in the adaptive program,
then a warning would
be appropriate since the Java compiler will not complain
(since no code is produced by the strategy "from C to D"
and its use).

You propose a style of AP where the class graph determines the parts
of the strategy which are relevant and so far Demeter has promoted
an approach where ALL the information in the strategies
must be reflected in the class graph.

So you promote a PARTIAL STRATEGY style while so far we used an
ALL STRATEGY style. The PARTIAL STRATEGY style is certainly more 
flexible since we can reuse the same adaptive program with more class graphs.
The class graph determines which parts of the adaptive program are relevant.

I think you are right and we should only give a warning when
a strategy edge defines 0 paths in the class graphs. If the additional
flexibility becomes confusing we can always go back and strengthen
the level of the error messages.

What do the experts think about this? Any strong opinions?

-- Karl