field detection regression between Jackson 2.8 and 2.9

[hgschmie]

Using this test code: https://github.com/airlift/airlift/blob/master/json/src/test/java/io/airlift/json/TestJsonModule.java#L101-L110

which explicitly tests that in an object mapper that has all autodetection disabled: https://github.com/airlift/airlift/blob/master/json/src/main/java/io/airlift/json/ObjectMapperProvider.java#L107-L130

This test passes with Jackson 2.8. With Jackson 2.9, it reports

[ERROR] testFieldDetection(io.airlift.json.TestJsonModule) Time elapsed: 0.007 s <<< FAILURE!
java.lang.AssertionError: Sets differ: expected [make, model, year, purchased, nameList] but got [make, model, year, purchased, notes, nameList]
at io.airlift.json.TestJsonModule.testFieldDetection(TestJsonModule.java:109)

So the field, that is supposed to not-detected by Jackson is detected by 2.9 (but not by 2.8). I tested with 2.8.1 (passes), 2.8.11 (passes), 2.9.0 (fails) and 2.9.4 (fails).

Looking through the release notes I did not find anything that would explain this behavior (or how to turn it off), so I consider that an actual regression.

[cowtowncoder]

Ok sounds legit. 2.9.5 does have one bug fix, #1947, and I wonder if that might be related or same.

[cowtowncoder]

Looking at code, I am wondering if I am missing something, regarding base visibility settings.
If the intent is to disable introspection of fields, default settings need to be changed, as public fields are detected by default. So I was expecting to see something do that, but test seems to use quite basic ObjectMapper. Since a DI framework is being used it is quite possible I am just missing that wiring which could explain the difference.

As to changes: no planned changes were made to change the logic between 2.8 and 2.9.

[hgschmie]

Will test as soon as 2.9.5 comes out. Are snapshots available from somewhere? (oss?)

[cowtowncoder]

@hgschmie Sonatype snapshots should have auto-built 2.9.5-SNAPSHOT. Although there have been some issues with dependencies (jackson-base and/or jackson-parent did not get travis-built from commits), so may require some local building.

[hgschmie]

I can confirm that 2.9.5-SNAPSHOT for core and databind (these are the deps that I have replaced) fix the problem. So this seems to be the problem described in #1947 .

[mbasmanova]

I ran into this issue as well. When will 2.9.5 release become available?

[cowtowncoder]

Planning to release 2.9.5 before end of the month, possibly over the weekend.

[hgschmie]

Already wrote above that the fix in #1947 fixes the problem for me. So

[cowtowncoder]

@hgschmie ah sorry. Somehow missed your update, didn't see it. Thanks!

[cowtowncoder]

Fixed by 2.9.5.