+ Disjunction Experiments

These are the results of trial runs during implementation of fully disjunctive LVs. Summary: about 40% speedup can be achieved simply by representing the same problem differently.

Snapshots were taken at these different steps in the implementation:

1. old behaviour (only relevance is computed by level)
2. rewrite `lastresort' constraints so that they only apply to ROOT edges (to see whether this hurts perfomance)
3. integrate shallow optimizer that detects label accesses qualified by label at the top level
4. integrate fully recursive optimizer that understands almost all qualifications
5. fix serious bug in version 4 that should have hurt performance but miraculously didn't

-- KilianAFoth - 17 Jan 2003