[jira] [Resolved] (FLINK-2447) TypeExtractor returns wrong type info when a Tuple has two fields of the same POJO type

["Fabian Hueske (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/FLINK-2447?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Fabian Hueske resolved FLINK-2447.
----------------------------------
    Resolution: Fixed

Fixed in 0.10 with 5546a1efabcc9c5f500abdb4e38a5cc05d35980a
Fixed in 0.9.1 with ec3b983276e85e0baed98ffa8acf6709f20032de

> TypeExtractor returns wrong type info when a Tuple has two fields of the same POJO type
> ---------------------------------------------------------------------------------------
>
>                 Key: FLINK-2447
>                 URL: https://issues.apache.org/jira/browse/FLINK-2447
>             Project: Flink
>          Issue Type: Bug
>          Components: Java API
>            Reporter: Gabor Gevay
>            Assignee: Timo Walther
>             Fix For: 0.10, 0.9.1
>
>
> Consider the following code:
> DataSet<FooBarPojo> d1 = env.fromElements(new FooBarPojo());
> 		DataSet<Tuple2<FooBarPojo, FooBarPojo>> d2 = d1.map(new MapFunction<FooBarPojo, Tuple2<FooBarPojo, FooBarPojo>>() {
> 			@Override
> 			public Tuple2<FooBarPojo, FooBarPojo> map(FooBarPojo value) throws Exception {
> 				return null;
> 			}
> 		});
> where FooBarPojo is the following type:
> public class FooBarPojo {
> 	public int foo, bar;
> 	public FooBarPojo() {}
> }
> This should print a tuple type with two identical fields:
> Java Tuple2<PojoType<FooBarPojo, fields = [bar: Integer, foo: Integer]>, PojoType<FooBarPojo, fields = [bar: Integer, foo: Integer]>>
> But it prints the following instead:
> Java Tuple2<PojoType<FooBarPojo, fields = [bar: Integer, foo: Integer]>, GenericType<FooBarPojo>>
> Note, that this problem causes some co-groups in Gelly to crash with "org.apache.flink.api.common.InvalidProgramException: The pair of co-group keys are not compatible with each other" when the vertex ID type is a POJO, because the second field of the Edge type gets to be a generic type, but the POJO gets recognized in the Vertex type, and getNumberOfKeyFields returns different numbers for the POJO and the generic type.
> The source of the problem is the mechanism in TypeExtractor that would detect recursive types (see the "alreadySeen" field in TypeExtractor), as it mistakes the second appearance of FooBarPojo with a recursive field.
> Specifically the following happens: createTypeInfoWithTypeHierarchy starts to process the Tuple2<FooBarPojo, FooBarPojo> type, and in line 434 it calls itself for the first field, which proceeds into the privateGetForClass case which correctly detects that it is a POJO, and correctly returns a PojoTypeInfo; but in the meantime in line 1191, privateGetForClass adds PojoTypeInfo to "alreadySeen". Then the outer createTypeInfoWithTypeHierarchy approaches the second field, goes into privateGetForClass, which mistakenly returns a GenericTypeInfo, as it thinks in line 1187, that a recursive type is being processed.
> (Note, that if we comment out the recursive type detection (the lines that do their thing with the alreadySeen field), then the output is correct.)



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)