Re: [eclipse-users] ic(sets) bug

Malcolm Ryan wrote:
> I've discovered a problem with ic(sets):
> 
> [eclipse 9]: #([3,2,1], L).
> 
> No (0.00s cpu)
> [eclipse 10]: #([1,2,3], L).
> 
> L = 3
> Yes (0.00s cpu)

[3,2,1] is not a _set_ in the sense of the library.
The solver works on the domain of the sorted, duplicate-free
lists of integers.  If you give anything else to the constraints,
the behaviour is not defined.  That behaviour is pervasive, e.g.

?- sameset([1,2,3],[3,2,1]).
No.

and I wouldn't like to change anything.  It simply means that all
the set-constants you use in your code must be well-formed, so if
you have unsorted lists, use sort/2 first to make a proper set.
You could argue that the constraints should raise errors instead
of failing, but adding sortedness checks everywhere would be quite
expensive.

This should probably be clarified in the documentation. Also,
ic_sets:is_solver_type/1 should probably check for sortedness.


-- Joachim