Should we support annotating JSON instances with the results of the evaluation?

[fdutton]

One of the big differences between Draft 6 and Draft 7 is the description of how a Validator should annotate JSON instances in the presence of possibly short-circuiting operations like oneOf, anyOf and if-then-else. Later drafts rely more heavily on this notion of annotating each each node in the JSON instance. For example, the behavior of the unevaluatedItems keyword in Draft 2020-12 is defined in terms of how the prefixItems, items and contains keywords annotated a particular node in the JSON instance.

The main driver for this change is the acknowledgement that the application code may be in a better position to determine how to address non-conforming data than the simple pass-fail results provided by a Validator that only uses assertions (like this library currently provides).

The major benefit we gain from this approach is that reasoning about the validity of a JSON instance becomes simpler. However, this simplicity has a cost in that it requires:

• A custom Jackson JsonNodeFactory that produce an annotatable version of each subclass of JsonNode
• Subclasses for each JsonNode to accommodate annotations
• Excluding annotations from the Jackson serialization mechanism