Getter that returns an abstract collection breaks a delegating `@JsonCreator`

[2is10]

Using version 2.9.2 or 2.9.8 (and probably many other versions), the following test throws:

JsonMappingException: Cannot find a deserializer for non-concrete Collection type [collection type; class java.util.AbstractSet, contains [simple type, class java.lang.Integer]]

    @Test
    public void testDeserializeValSet() throws IOException {
        new ObjectMapper().readValue("[1,2,3]", ValSet.class);
    }

    static class ValSet {
        @JsonValue
        private final AbstractSet<Integer> vals;

        @JsonCreator(mode = Mode.DELEGATING)
        ValSet(Collection<Integer> vals) {
            this.vals = new HashSet<>(vals);
        }

        public AbstractSet<Integer> getVals() {
            return vals;
        }
    }

This is a spurious error since the @JsonCreator constructor only requires Collection<Integer>, not AbstractSet<Integer>.

Some workarounds:

• Delete getVals().
• Make getVals() non-public.
• Move @JsonValue to getVals().
• Add @JsonIgnore to getVals().
• Add @JsonAutoDetect(getterVisibility = Visibility.NONE) to ValSet.

[cowtowncoder]

@2is10 thank you for reporting this: with quick glance this does seem like a bug: for deserialization, the highest precedence information should be constructor parameter which declares expected type of Collection<Integer>, and whatever type serialization side has (getter) should not override this (not @JsonValue, which should not even come to play here).

[cowtowncoder]

Ok. So, one thing I fixed first is that for some reason there were no fallback mappings for AbstractSet (or AbstractList or AbstractMap). There should be -- same ones used for Set, List and Map. Doing that works around this issue.

But I'll see if I can also find why type detection prefers AbstractSet here: while it is more refined, it is not from the most specific accessor to use for deserialization.
It is possible this is related to one known problem with Creator, creator properties being separate set from "regular" properties (and unification occurring too late in the process), and if so, that can't be solved before 3.0. But it is not yet clear if that is the case or not.

[cowtowncoder]

Hmmh. Ok, so accessor problem comes via "setterless" property, which is concept wherein "getter" may be used as sort-of setter if nothing else is available. And since what we have here is delegating constructor (and NOT setter-based), there is attempt to find deserializer for theoretical use for property vals...

This is not optimal, but I don't think I can solve that for time being, so I will let the preliminary fix take care of short-term problem here. But I'll see if I can write a separate failing test for this problem.

[cowtowncoder]

Created #2252 as followup for this issue.

[2is10]

Thanks @cowtowncoder!

Note: In my actual code, the abstract Set class that triggered the problem was Guava's ImmutableSet, so I'm personally still invested in a solution to #2252. (I didn’t want to introduce a dependency on jackson-datatypes-collections/guava as a workaround.)