[jfmc]

Many times the algorithm that you are implementing requires a precise data flow that is not reversible, so using traditional arithmetic (is/2) is better for catching errors.

On the other hand CLP(FD) is not new at all (it is very popular for constraint programming).

[rscho]

Why would it be better for error handling? If you'll be using unidirectional flow only, then the point is moot. But using clp is arguably better IMO, allowing type and range checks while allowing relational execution.