I finally managed to commit my (huge) changes to the implementation of BDSs. The main novelty are:
- shortest-path reduction no longer compromises shortest-path closure (note: I still have to implement the extraction of constraints for the case of a dbm in reduced form);
- affine images, preimages and generalized affine images have been rewritten almost from scratch; they should be less buggy than before (this is definitely the case for the case of generalized images); at the very least, the current code is no longer relying upon unreasonable assumptions on the kind of coefficients and the sign of denominators.
Some tests will now fail: I strongly suspect that for some of them there is no real bug, but rather the "known_result" was wrong.
In summary, there still is plenty of things to do.
Ciao, Enea.