Jackson 2.8.11 release for Scala module

[cowtowncoder]

I hope this should be easier than 2.9. So, 2.8.11 is out:

https://github.com/FasterXML/jackson/wiki/Jackson-Release-2.8.11

and it'd be good to get Scala module out too. This is planned to be the last full 2.8.x release, closing the branch. So should perhaps simplify things a bit (I forget if there's any scala version deprecation involved here, perhaps not).

[mbknor]

hmm.. This one also fails with the same error as #357

Is it something wrong with my machine?

....

We need to figure this out..

[cowtowncoder]

Hmmh. It could be one of backported fixes. While nothing looks particularly suspicious, it could be either one of:

• Nested type arguments doesn't work with polymorphic types jackson-databind#1604
• Jackson-databind caches plain map deserializer and use it even map has @JsonDeserializer jackson-databind#1807

One thing you could try locally is to build against 2.8.10 and see if that'd work.
I would guess it does, which wouldn't help a lot. If it is easy enough to build against 2.8.11-SNAPSHOT could also try checking out various points in 2.8 to rule out PRs.
But I guess fundamentally, assuming it is side-effect of one of the fixes, something would probably have to change in the test. Or, in worst case, scala module... (worst just because it's not very easy to see how things work, sometimes).

[mbknor]

I have not tracked the issue down yet, but this is what I've found out so far:

I've been using the 2.8-branch for testing - I guess there are less changes between 2.8.10 and 2.8.11.

The problem is this line:

jackson-module-scala/src/main/scala/com/fasterxml/jackson/module/scala/deser/UntypedObjectDeserializerModule.scala

Line 29 in fd3055b

_mapDeser = ctxt.findRootValueDeserializer(

In 2.8.10, _mapDeser ends up with SortedMapDeserializer
In 2.8.11, it ends up with SeqDeserializer.

I have not figured out why yet, but I write it here for info/starting point

[cowtowncoder]

Ok. I don't know if change is problematic (it may well be), but for context that "untyped" deserializer is for use with java.lang.Object as nominal type (most often due to raw types / missing generic declaration). And override is needed, I think, to change mapping from JSON Object into Scala Map, not JDK Map (and perhaps ditto for Java List to Scala List).

[mbknor]

I found it the change causing the problem:

FasterXML/jackson-databind@4675896#diff-71da9745111f01bc93519d574c2f9fabL416

if (baseType.isInterface()) {
    // The scala-test was relying on this line.
    newType = baseType.refine(subclass, tb, null, new JavaType[] { baseType });
} else {
    newType = baseType.refine(subclass, tb, baseType, NO_TYPES);
}

For 2.8.10, we end up with isMapLike == true, but for 2.8.11 we end up with isCollectionLike == true which breaks down the line..

I'm not sure if I have found a regression or not.. What do you think?

[cowtowncoder]

Hmmh. This could also be related to the fact that Scala module is not necessarily supporting type refinement (to support collection- / map-like) in the right place: things have changed a little.
So another place to look might be TypeModifier, to ensure it recognizes the need.

New code in this specific place is, I think, (more) correct. But I don't quite understand negative ramifications on Scala side.

Getting CollectionLikeType and MapLikeType right is important, since code in other areas relies on that.

[mbknor]

2.8.11 released