[Help needed] Fix custom null serialization

[marcospassos]

@cowtowncoder implementing a null custom serialization for an internal library, I stumbled upon a bug that prevents serialized null from being deserialized (which should affect other packages, such as JSR-353). For now, this PR only includes a test that reproduces the problem.

Error message:

com.fasterxml.jackson.databind.exc.MismatchedInputException: No content to map due to end-of-input

Could you give me some guidance on how to fix it?

[cowtowncoder]

I think this is not right: should instead check that current token is JsonToken.VALUE_NULL: parser already points to the first token of value to deserialize.

[cowtowncoder]

It'd be good to maybe file issue separately, link to/from this PR.
There's one small problem with custom deserializer, not sure if it matters.

[marcospassos]

It'd be good to maybe file issue separately, link to/from this PR.

I'll open a PR.

I just pushed a new commit, but it doesn't seem to fix the issue.

[marcospassos]

Additional info: the same does not happen when serializing/deserializing to/from JSON.

[cowtowncoder]

Ok. Thank you for verifying.

I think this could be related to one oddity wrt Avro handling of nulls: unless null value is part of union, it "does not exist": that is, it does not take any space. Value of a field with only null type is always null, and nothing is encoded.
This is fine in theory, but presents unique problem at root-level reader: it is not easy to know if stream has ended or not -- just because there are no bytes left does not necessarily mean we are done as null may still be implicitly available!
(I had to fix one bug earlier related to this).

[marcospassos]

Dammit, now I see why I couldn't find an easy way to fix it myself without consuming your limited time. But it seems a critical bug to fix, as it can break things seriously. Again, let me know if I can help you fix it.

[cowtowncoder]

@marcospassos Unfortunately I do not really have time to work on this now.