[hibernate-dev] [Bean Validation] Type-safe Validator and resolution algorithm

Today, ConstraintValidator accepts Object and the implementation is responsible for handling the casting and raise an exception if the type is not supported. The idea is to get type-safe validator and a discovery mechanism to associate a validator to a given runtime type being validated. The resolution is not 100% complete, you might end up with: - no valid validator - no way to decide between two or more validators and get exceptions at runtime. I would really like feedback on it as we could arrange some of these rules to not fail (at least in the latter case): - do we want this type-safe + auto resolution algorithm strategy or should we stick with the untyped solution - do we want to fail in case of ambiguity or rather choose the first compatible provider? Personally I find the feature quite elegant and the metadata can be used by tools to warn a user if a constraint does not match the static type it is applied on. API and declaration A more typesafe proposal would be: public interface ConstraintValidator<A extends Annotation, T> { void initialize(A annotation); boolean isValid(T object, ConstraintValidationContext context); } public class StringSizeValidator implements ConstraintValidator<Size, String> { ... } public class CollectionSizeValidator implements ConstraintValidator<Size, Collection> { ... } public class MapSizeValidator implements ConstraintValidator<Size, Map> { ... } public class ArraySizeValidator implements ConstraintValidator<Size, Object[]> { ... } @Constraint(validatedBy={ StringSizeValidator.class, CollectionSizeValidator.class, MapSizeValidator.class, ArraySizeValidator.class} ) public @interface Size { ... } We then need to decide at runtime, which validator needs to be executed. Here is a proposed algorithm heavily based on the Java Language Specification (boy it's hard to read it). Resolution algorithm The type T of a ConstraintValidator must not make direct use of generics itself. This is because the algorithm compare the actual type at runtime and such notion would be lost. So you can do ConstraintValidator<Size, Collection> but not ConstraintValidator<Size, Collection<String>>. For a given runtime object t of type T about to be validated for a given constraint: if t is null the first validator in the array of available validators is used. otherwise A validator accepting U is said to be compliant with the type T if T is a subtype of U according to the JLS chapter 4.10. If no compliant validator is found for T, a UnexpectedTypeForConstraint exception is raised. If a validator is taking the responsibility to dispatch types (ie ConstraintValidator<Size,Object>), it must use the same exception in the same circumstances. If one compliant validator is found, it is used. If more than one compliant validator is found, the one with the most specific parameter is used. The informal intuition is that one parameter is more specific than another if any assignment handled by the first parameter could be passed on to the other one without a compile-time type error. Of a given type T, a compliant validator accepting U is considered maximally specific if no other T compliant validator accepting V is such as V is a strict subtype of U. String subtype is defined by JSL chapter 4.10. If more than one maximally specific validator is found, a AmbiguousValidatorException is raised. WDYT?