Mario Eberhard created AVRO-2236:
------------------------------------
Summary: Java Avro Default Value restrictions to first union type
leaks to usage of record types
Key: AVRO-2236
URL: https://issues.apache.org/jira/browse/AVRO-2236
Project: Avro
Issue Type: Bug
Components: java
Affects Versions: 1.8.2
Reporter: Mario Eberhard
Attachments: avrodefaulttest.zip
Default values are restricted to being of the same type as the first type in a
union.
{code:java}
{
"type": "record",
"name": "ComplexValue",
"fields":
[
{
"name": "value",
"type":
[
"null",
"long"
],
"default": null
}
]
}
{code}
This works as documented. However, the restriction also applies to default
values of this record type:
{code:java}
{
"type": "record",
"name": "ExampleRecord",
"namespace": "com.example",
"fields":
[
{
"name": "value1",
"type":
{
"type": "record",
"name": "ComplexValue",
"fields":
[
{
"name": "value",
"type":
[
"null",
"long"
],
"default": null
}
]
}
},
{
"name": "value2",
"type": "ComplexValue",
"default":
{
"value": 15
}
}
]
}
{code}
In this case the record "ExampleRecord" has a field "value2" of type
"ComplexType". This field is not optional but has a default to be able to read
instance where this field is not present.
During deserialization the following error is thrown:
{code:java}
org.apache.avro.AvroTypeException: Non-null default value for null type: 15
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.encode(ResolvingGrammarGenerator.java:413)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.encode(ResolvingGrammarGenerator.java:365)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.encode(ResolvingGrammarGenerator.java:335)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.getBinary(ResolvingGrammarGenerator.java:307)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.resolveRecords(ResolvingGrammarGenerator.java:285)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.generate(ResolvingGrammarGenerator.java:118)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.generate(ResolvingGrammarGenerator.java:50)
at org.apache.avro.io.ResolvingDecoder.resolve(ResolvingDecoder.java:85)
at org.apache.avro.io.ResolvingDecoder.<init>(ResolvingDecoder.java:49)
at
org.apache.avro.io.DecoderFactory.resolvingDecoder(DecoderFactory.java:307)
at
org.apache.avro.generic.GenericDatumReader.getResolver(GenericDatumReader.java:128)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:143)
at
com.example.SerializationTest.lambda$getDeserializer$1(SerializationTest.java:75)
at
com.example.SerializationTest.valueReadWithCorrectDefaultValue(SerializationTest.java:41)
{code}
I would argue that a concrete instance of "ComplexValue" with a specific value
should be allowed in this case. I see no reason why the default restriction of
the underlying schema should even apply. My guess is, that this is an
unintended consequence of code reuse in the java client.
I added an example gradle project as attachment. Run the tests to reproduce the
above example.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
