On Sat, Mar 31, 2018 at 11:51 AM, Gunnar Morling <gunnar@hibernate.org> wrote:
That's an interesting one; it was originally worded exactly like Matt suggests, but then we changed into the current version: https://hibernate.atlassian.net/browse/BVAL-678.

I'm sympathetic to the current behaviour of that TCK test and to adjusting the spec wording accordingly. Only you, Guillaume, didn't seem to like that alternative as per our discussion on June 26th, in the Hibernate Validator chat room. Perhaps you remember what made you dislike it back then?

Ah, yes, I remember now.

The paragraph is:
  • If the constraint carries neither the Unwrapping.Unwrap nor the Unwrapping.Skip payload:
    • If there is exactly one maximally-specific type-compliant value extractor and this extractor is marked with @UnwrapByDefault, this extractor is applied;
    • Otherwise, no value extractor is applied.
The initial wording "exactly one maximally-specific type-compliant value extractor marked with @UnwrapByDefault" was not accurate because it was not clear if we should determine the "maximally-specific type-compliant value extractor" first and then check if it is @UnwrapByDefault. This is important because, for a given type parameter, we want to get the maximally specific value extractor first and then we want to check if it is marked with @UnwrapByDefault. We don't want to get the maximally specific (value extractor marked with @UnwrapByDefault).

We also miss a mention of the case when there are 2 maximally-specific type-compliant value extractors marked with @UnwrapByDefault.

I think a wording more in line with our intentions would be:
  • We determine the maximally-specific type-compliant value extractors.
    • If there are no matching value extractors or none is marked with @UnwrapByDefault, no value extractor is applied.
    • If exactly one is marked with @UnwrapByDefault, this very extractor is applied.
    • If more than one are marked with @UnwrapByDefault, a ConstraintDeclarationException is raised.
I would lean towards implementing it this way in the implementations and fix it for BV 2.1.

-- 
Guillaume