Problem serializing `JsonMappingException` due to addition of non-ignored `processor` property

[jecaplan]

I notice that, e.g., JsonMappingException has several bidirectional links in internal fields like _processor.

Why not annotate those fields with @JsonIgnore? You've already got the Java serialization "transient" on them.

One symptom of not eating your own dog food here is that AWS SWF Flow Framework is broken. If it attempts to deserialize the result of an activity and that attempt fails with a JsonMappingException, the framework then attempts to serialize the resulting DataConverterException via Jackson...which fails with a different JsonMappingException (StackOverflowError) because of the embedded JsonMappingException.

I see no reason why a serde framework should not be able to serde itself.

[cowtowncoder]

Since there is transient (and since field is marked as protected) it seems superfluous to also add @JsonIgnore: former alone should prevent usage. If this is not enough for some reason, I am not against adding annotations; and certainly exceptions produced should be Jackson-serializable.
In perfect world deserialization should also work, although that is bit trickier due to usually polymorphic nature of exception hierarchies.

If there is an actual problem (which sounds like there is), a reproducible test case would be helpful, especially to guard against future regressions. But also to make sure I understand the specific problem.

[jecaplan]

This example is with Jackson 2.8.0 I believe (whatever the latest AWS SDK is using).

public class TestJackson {
    public static class NoSerdeConstructor {
        private String strVal;
        public String getVal() { return strVal; }
        public NoSerdeConstructor( String strVal ) {
            this.strVal = strVal;
        }
    }
    @Test
    public void test() throws IOException {
        ObjectMapper mpr = new ObjectMapper();

//      // fails: shouldn't this pass?
//      assertFalse( mpr.canDeserialize( mpr.constructType( NoSerdeConstructor.class ) ) );

        try {
            mpr.readValue( "{ \"val\": \"foo\" }", NoSerdeConstructor.class );
        } catch ( JsonMappingException excCannotConstructInstance ) {
            // throws a different JsonMappingException: Infinite recursion
            mpr.writeValueAsString( excCannotConstructInstance );
        }
    }
}

[cowtowncoder]

Thank you! I'll see if it occurs with 2.7 as well (most likely does); original description makes more sense with code now.

Note on canDeserialize: test is approximate and can not currently catch all problem cases. This particular case should be possible to detect (since there are no alternate means to construct), but JsonDeserializer API does not have means to expose the problem as is. The only cases that would report "can not deserialize" are the ones where either no deserializer is found for type, or construction of deserializer fails with exception. So in practice neither canSerialize() nor canDeserialize() is particular useful.

[cowtowncoder]

Ah yes. You are right about need for @JsonIgnore: I forgot that transient keyword does not (by default) propagate to matching getter.

Quite odd that this wasn't reported earlier for 2.7, but better late than never.

[jecaplan]

I have a pull request with regression test open if that's of any use to you.

On Sep 12, 2016 10:46 PM, "Tatu Saloranta" notifications@github.com wrote:

Ah yes. You are right about need for @JsonIgnore: I forgot that transient
keyword does not (by default) propagate to matching getter.

Quite odd that this wasn't reported earlier for 2.7, but better late than
never.

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1368 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAM9dCvdMG1K_Yv_wHr8sDINWogK58USks5qpjicgaJpZM4J6EKy
.

[cowtowncoder]

@jecaplan it is indeed.