[bv-dev] Distinguishing cross-parameter and generic constraints

[Emmanuel Bernard]

On the subject, Gunnar and I were discussing three valid options to
define a constraint as cross-parameter constraint:

## @CrossParametersConstraint

(first off I hate the name, can't we find a better one?)

    @Constraint(validatedBy=ObjectScriptAssert.class)
    @CrossParametersConstraint(validatedBy=ObjectScriptAssert.class)
    public @interface ScriptAssert { ... }

## @Constraint for both

    @Constraint(validatedBy=ObjectScriptAssert.class
                validatedByForCrossParameters=ObjectScriptAssert.class)
    public @interface ScriptAssert { ... }

## @Validates on ConstraintValidator


    @Constraint(validatedBy=ObjectScriptAssert.class)
    public @interface ScriptAssert { ... }

    @Validates({CROSS_PARAMETERS, ANNOTATED_ELEMENTS})
    public class ObjectScriptAssert implements
        ConstraintValidator<ScriptAssert,Object[]> { ... }

## Thoughts

I think I like @CrossParametersConstraints more than @Constraint for
both as it makes things explicit and keep the simple use case as is.

But I am intrigued by the @Validates option. In many ways, it's similar
to how we resolve the right constraint validator based on the type. The
only difference is that the @Validates annotation refines how Object[]
should be interpreted. It "fits better" in my opinion.

My concern around the latest option is that it forces us to be a
subclass of ConstraintViolation for the foreseeable future including
when we reopen the type-safe options we explored in 
http://beanvalidation.org/proposals/BVAL-232/

Thoughts and comments?

Emmanuel


On Tue 2013-01-15 15:14, Gunnar Morling wrote:
> 2013/1/15 Emmanuel Bernard <emmanuel at hibernate.org>
> 
> > I can live with that but we must make it crystal clear that in case of
> > ambiguity, we raise an exception. Otherwise, someone changing the method
> > sig might have some nasty surprises.
> >
> 
> And which of the options
> 
>     @CrossParameterConstraint(validatedBy=MyCrossParameterValidator.class)
>     @Constraint(validatedBy=MyGenericValidator.class)
>     public @interface MyConstraint { ... }
> 
> vs.
> 
>     @Constraint(
>         validatedBy=MyGenericValidator.class,
>         crossParameterValidatorType=MyCrossParameterValidator.class
>     )
>     public @interface MyConstraint { ... }
> 
> would you prefer?
> 
> Personally I'd lean towards the latter, since there would be exactly one
> meta-annotation denoting an annotation as constraint annotation type, with
> all attributes for the required configuration.
> 
> If following the 2nd approach, we should IMO add a default clause to
> validatedBy() (set to the empty array), allowing for
> the convenient definition of cross-parameter only constraints. According to
> [1] this change would be binary-compatible.
> 
> But this is no strong preference, I could also live with the two separate
> annotations.
> 
> --Gunnar
> 
> [1]
> http://wiki.eclipse.org/Evolving_Java-based_APIs_2#Evolving_API_interfaces_-_API_methods

> _______________________________________________
> beanvalidation-dev mailing list
> beanvalidation-dev at lists.jboss.org
> https://lists.jboss.org/mailman/listinfo/beanvalidation-dev