Default Values for Optionals Or Reject All Nulls

[NTCoding]

Hello,

I think your library is great but it doesn't quite work for one Scala use case. When there is a property of type Option I want it to set a default value of none even when FAIL_ON_MISSING_CREATOR properties is true.

I would like to do this conventionally without having to apply any annotations to the code. Could you give me advice on the best way to implement this and I will do the work and send a pull request.

Something in here seems sensible:

 protected Object _findMissing(SettableBeanProperty prop) throws JsonMappingException
    {
        // First: do we have injectable value?
        Object injectableValueId = prop.getInjectableValueId();
        if (injectableValueId != null) {
            return _context.findInjectableValue(prop.getInjectableValueId(),
                    prop, null);
        }
        // Second: required?
        if (prop.isRequired()) {
            throw _context.mappingException("Missing required creator property '%s' (index %d)",
                    prop.getName(), prop.getCreatorIndex());
        }
        if (_context.isEnabled(DeserializationFeature.FAIL_ON_MISSING_CREATOR_PROPERTIES)) {
            throw _context.mappingException("Missing creator property '%s' (index %d); DeserializationFeature.FAIL_ON_MISSING_CREATOR_PROPERTIES enabled",
                    prop.getName(), prop.getCreatorIndex());
        }
        // Third: default value
        JsonDeserializer<Object> deser = prop.getValueDeserializer();
        return deser.getNullValue(_context);
    }

Perhaps a new service "defaultCreatorValueProvider" or similar?

If there is a better way, or even an existing way, please let me know.

Thanks

[cowtowncoder]

Definition of FAIL_ON_MISSING_CREATOR is pretty clear -- fail if value is missing from incoming JSON. I don't see why its semantics should be changed.

But the other case of allowing missing properties should handle default value for Option value correctly; similar to how primitive values are handled (they can not be set to null). As long as deserializer used for Option defines non-null return value for getEmptyValue(...), which is called to get value to pass to creator method.

[NTCoding]

Hi,

I understand that you are a Java developer and perhaps are not familiar with monads and optionals (apologies if that is condescending). However, an Option[T] is explicitly stating that the type is optional - part of the type's DNA is that there may be no value for it - and hence a missing constructor value means it should default to a value of None.

It is explained in more detail here: FasterXML/jackson-module-scala#203

If an object has two properties - one is optional and one is not - the type system is expressing what is mandatory - the optional values. Is there any way to do this currently - if the value is missing throw an exception unless it is an Option?

In the "other case" you mention, then it would not fail if a value was missing from a non-optional type and the value would default to null. Would you consider a pull request with a new deserialization option: "REJECT_ALL_NULL_VALUES" or are you completely against adding new functionality that helps to support Scala optionals?

[NTCoding]

I've added a pull request to demonstrate another possible way of solving this problem without interfering with the fail on creator properties. Please let me know what you think: #990

I'm happy to put in more effort if you give me guidance.

Thanks

[cowtowncoder]

@NTCoding keep in mind Jackson knows nothing about Scala, and generally does not care.
That Scala has additional semantics does not really matter -- null is a null, and Optional is NOT null. There's no two ways about that. Instead Optional's (and other referential types, ReferentialType as of 2.6) do have concept of "absent", and that is separated.

I hope to have a look at the proposal in near future, but I am bit backlogged due to major rewrite of type handling to resolve long-standing problems with generic type resolution (wrt type aliasing, mostly).

[NTCoding]

Thanks for the feedback. I guess the code is open source so I can publish my own fork for now.

[cowtowncoder]

Replaced with #1402, which I think would provide all the pieces needed. Scala module might need something additional (and definitely needs to support this); but from jackson-databind perspective I think this should be sufficient. Planning to get implemented for 2.9.