Inconsistent behaviour with nulls for records

[joca-bt]

Forbidding nulls works correctly with classes but doesn't with records. Moreover, the behaviour with records changes if we use @JsonProperty.

Let's take the following ObjectMapper definition:

ObjectMapper objectMapper = JsonMapper.builder()
    .defaultSetterInfo(JsonSetter.Value.construct(Nulls.FAIL, Nulls.FAIL))
    .withCoercionConfigDefaults(config -> config.setCoercion(CoercionInputShape.String, CoercionAction.Fail))
    .build();

When testing nulls with a class everything works as expected:

public class Test {
    private String fieldName;

    public void setFieldName(String fieldName) {
        this.fieldName = fieldName;
    }
}

objectMapper.readValue("{ \"fieldName\": \"value\" }", Test.class); // ok (no exception is thrown)
objectMapper.readValue("{ \"fieldName\": null }", Test.class); // ok, throws InvalidNullException as expected
objectMapper.readValue("{ }", Test.class); // ok (no exception is thrown)

When testing the same class but adding @JsonProperty to it everything still works as expected:

public class Test {
    private @JsonProperty("field_name") String fieldName;

    public void setFieldName(String fieldName) {
        this.fieldName = fieldName;
    }
}

objectMapper.readValue("{ \"field_name\": \"value\" }", Test.class); // ok (no exception is thrown)
objectMapper.readValue("{ \"field_name\": null }", Test.class); // ok, throws InvalidNullException as expected
objectMapper.readValue("{ }", Test.class); // ok (no exception is thrown)

When testing nulls with a record null hating doesn't seem to be working:

public record Test(String fieldName) {}

objectMapper.readValue("{ \"fieldName\": \"value\" }", Test.class); // ok (no exception is thrown)
objectMapper.readValue("{ \"fieldName\": null }", Test.class); // not ok, was expecting a InvalidNullException to be thrown but it wasn't
objectMapper.readValue("{ }", Test.class); // ok (no exception is thrown)

When testing the same record but adding @JsonProperty to it null hating starts working but something else breaks:

public record Test(@JsonProperty("field_name") String fieldName) {}

objectMapper.readValue("{ \"field_name\": \"value\" }", Test.class); // ok (no exception is thrown)
objectMapper.readValue("{ \"field_name\": null }", Test.class); // ok, throws InvalidNullException as expected
objectMapper.readValue("{ }", Test.class); // not ok, throws InvalidNullException when it shouldn't

[JooHyukKim]

Issues #2974 #2970 might answer your question.

Btw, I failed to recreate the "Inconsistent behavior" at this PR #3848. Could you check again your test cases, @joca-bt please? I will tag you.

[joca-bt]

Here's a simple project replicating it.

test2.zip

[JooHyukKim]

Here's a simple project replicating it.
test2.zip

Can you post github link instead and delete comment with zip file link? For security concerns. Or even better check the PR #3848 I mentioned.

[joca-bt]

You can inspect the zip without running it if you have security concerns. Best, João

…

On Wed, 29 Mar 2023, 04:35 Kim, Joo Hyuk, ***@***.***> wrote: Here's a simple project replicating it. test2.zip <https://github.com/FasterXML/jackson-databind/files/11092234/test2.zip> can you post github link instead and delete comment with zip file link? For security concerns. — Reply to this email directly, view it on GitHub <#3847 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJ44FSDEFPYGPWZLSXWBU3W6ONYZANCNFSM6AAAAAAWKGHJVI> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[yihtserns]

@joca-bt From the attached project, I see that you're using version 2.14.2. I think this issue has been fixed in 2.15.0 by #3724, which is why #3848 wasn't able to reproduce the issue.

You can verify this by testing with com.fasterxml.jackson.core:jackson-databind:2.15.0-rc2 (don't forget to also upgrade any other Jackson dependency to the same version, to avoid any unnecessary problems).

[cowtowncoder]

@joca-bt if you are able to verify this against 2.15.0-rc2 please let us know so I can close this (or keep open if issue remains).

[joca-bt]

I can test it in the weekend. Where can I get the snapshot jar from?

[cowtowncoder]

@joca-bt Not snapshot, pre-release candidate (2.15.0-rc2), from Maven Central. :)

[joca-bt]

Ran the test cases in the zip I shared against 2.15.0-rc:

• Fixes line 27 objectMapper.readValue("{ \"fieldName\": null }", Test3.class); // should throw but doesn't. It now throws an InvalidNullException.
• Breaks line 28 objectMapper.readValue("{ }", Test3.class);. It now throws an InvalidNullException when it shouldn't.
• Line 32 remains the same (broken) objectMapper.readValue("{ }", Test4.class); // throws but shouldn't. Still throws an InvalidNullException when it shouldn't.

Overall, there are still some issues with records.

[yihtserns]

Right I totally missed the { } scenario.

Actually your class is not identical to your record, e.g.:

public record Test(String fieldName) { }

...is actually equivalent to:

public class Test {
  private String fieldName;
  public Test(String fieldName) { this.fieldName = fieldName) }
}

...which when fed with empty hash { }, will also throw InvalidNullException.

---

The javadoc for Nulls & JsonSetter both mentioned "explicit null value". Does that mean there is a config to differentiate between the input containing <field name>: null vs input not having the field (i.e. absent)? 🤔

[cowtowncoder]

@yihtserns Missing value (absent) is generally differently handled from explicit null -- former causes no action wrt setters for example. But in case of Creators (constructors/factory methods) it needs to be handled so there can be a difference. For deserialization purposes JsonDeserializer does have separate methods (getNullValue(), getAbsentValue() to allow different coercions. This gets mostly tricky with things like Optional<...>.

[cowtowncoder]

I think this issue probably needs to be broken up into separate ones, targeting what is not apparently working with 2.15.0-rc2.
I am not sure all cases are working incorrectly, but they are easier to handle via separate issues usually as there may be separate root causes.

[yihtserns]

Created #3870 for "fail explicit null + constructor creator + absent value = InvalidNullException".

[yihtserns]

Just to repeat myself, a Class equivalent of Record is not one with setter, but this:

public class Test {
    private final String fieldName;

    @ConstructorProperties({"fieldName"})
    public Test(String fieldName) {
        this.fieldName = fieldName;
    }

    public String getFieldName() {
        return this.fieldName;
    }
}

(There were some behavioural discrepancies between a Class with constructor creator vs Records, but they are now mostly aligned in 2.15.0.)

...which will have this behaviour:

objectMapper.readValue("{ \"fieldName\": \"value\" }", Test.class); // no exception thrown
objectMapper.readValue("{ \"fieldName\": null }", Test.class); // throws InvalidNullException
objectMapper.readValue("{ }", Test.class); // throws InvalidNullException

Knowing that, this issue is really about the behavioural difference between a POJO with constructor creator vs setter, i.e. #3870.