Do not add constraints of the form s <= 'a <= t, for monomorphic alpha's, just check that the constraint holds for all substitution and replace it with a universal Sat or Unsat. This prevent the algorithm from introducing dummy substitutions later on of the form 'a <- 'a which interact badly with type cleaning (replacing covariant/contra-variant occurences by Empty/Any).

Fix conditionaly compiled code, that contained type errors.

Fix a regression where polymorphic variables would not be chosen in favor of monomorphic ones during the 'single' phase of constraint generation.

Improve the debugging of the unify step of the tallying. Fix a bug where substitutions were applied in the wrong order, yielding partially empty types.

Change the ordering use for types during constraint generation with one which is compatible with subtyping, when types are comparable.