RFC on FD.reified.sumCN

[Raphael Collet]

Christian Schulte wrote:
> 
> Well, I think Tobias fix was quite accurate in the sense that propagation
> has become correct. Furthermore, implemnting true nonlinear reified
> equations is quite painful while it usefulness is really limited. For me it
> is not so clear that doing anything better than that is worth the effort.

I understand.  I really deplore the absence of a comment like: "the
propagator is buggy in the non-linear case."  It costs nothing, and it
can save time for your peers!  I will add such a comment then.

> Another option is to go for expansion in the first place and save all the
> trouble.

I have thought about using the same trick as in FD.reified.sumACN
(exerpt from FD.oz):

      proc {GenSumACR IV DV Rel D B}
         NegRel = NegRelTable.Rel
      in
         {FdBool B}
         thread
            or B=1 {FdpSumAC IV DV Rel    D}
            [] B=0 {FdpSumAC IV DV NegRel D}
            end
         end
      end

Maybe that's a bit heavy, but at least it does some pruning. 
FD.reified.sumCN really is problematic, IMO.  For instance,
{FD.reified.sumCN [1] [X X] '>:' 0 1} does no pruning at all!  This
behavior is quite unexpected, and it makes the programmer think there's
something wrong in her program (that's how I found it).


Thanks a lot, Christian!  I appreciate that you replied so quick.

Cheers,
raph