Support deserializing [] as null or empty collection when the java type is a not an object

[fabienrenaud]

In PHP, it is possible to associate arrays. Those associates arrays can then be json encoded and turn to be JSON objects.
But when the associate array is empty, for various processing reasons, json_encode won't be able to tell that this particular empty associate array is an object, so it encodes it as [].
Then all hell breaks in Java because deserialize [] into a non array class doesn't work.

Task: Support that such as it deserialize to null when the java type is an non array object, and as empty collection when the java type is a collection.

[cowtowncoder]

So, would it be correct to say that for Maps, and perhaps POJOs, following JSON:

[ ]

should behave as a special case that would be equivalent to

{ }

If so, this could make sense, and could perhaps be added as a feature like, say, DeserializationFeature.ACCEPT_EMPTY_ARRAY_AS_EMPTY_OBJECT.
This is bit tricky for couple of reasons (Type Id handling, for example, can use array wrapping; and detection of non-empty arrays in streaming manner needs care); but it might be doable otherwise.

[fabienrenaud]

Maps and pojos are expected to be represented as objects in JSON. But in case they are empty, PHP may say

[]

So, yes, the proposed option makes sense :)
"EMPTY_OBJECT" thought would require some more information about how it is handle for different types... For any non array objects and non collection java type, i would use null. For collection java types, i would use the matching empty collection.

[cowtowncoder]

For collection objects that would already work. I guess use of null might be safest option for POJOs.

I loathe PHP. This kind of nonsense reminds me of why.

[fabienrenaud]

Do you think it could be supported in v2.5?
No pressure here, I just wonder when I should plan to stop hacking around that "bug" in my code :)

[cowtowncoder]

@fabienrenaud Yes I think so; I think that this is both high priority and reasonable to implement. But if latter turns out to be false I will add an update. So, yes, I hope to get this in by 2.5.

[cowtowncoder]

Turns out it was quite easy, so it's done.

[fabienrenaud]

Awesome :)
Any idea when 2.5 will be in maven repos?

[cowtowncoder]

2.5.0 has not been released, and there is no date set for starting with release candidates. So I think it will take a while. But perhaps it could get released in October.