Fix #1757 and #2105: honor global default inclusion setting

[bwaldvogel]

This fixes a regression that was introduced with 75098b7.

[cowtowncoder]

I don't think this is correct. Could you please explain the logic for change?

[bwaldvogel]

In 75098b7 you introduced the new flag _useRealPropertyDefaults which indicates if POJO property defaults are used to decide if a value should be suppressed for (de-)serialization.

As you mention in a comment, there are 3 possible levels of default inclusion definition/overrides:
(a) global
(b) per-type
(c) per-property

As far as I understand, the calculation of _useRealPropertyDefaults only takes (b) into account. Thus, a globally configured default inclusion (a) doesn’t work (anymore) and leads to the behavior change that was reported in #1757 and #2105. It is also shown by the new test case in my pull request.

My suggested fix sets _useRealPropertyDefaults to the merged combination of (a)+(b) instead of just using (b).

My suggested change doesn’t break an existing test in jackson-databind.
Do you have a specific situation in mind where you think this change is not correct?

[bwaldvogel]

@cowtowncoder: Thanks for clarifying your point of view.

Sorry for the misunderstanding on my side. I didn’t expect that this behavior change was actually intended in a bug-fix release (2.8.3) and I didn’t notice that the documentation was changed in 2.7.0.

It would be nice if there would be a unit test for it and a hint in the release notes.