Yes, this looks like a bug in GenericData#validate(). It should use
GenericData#resolveUnion().
Please file an issue in Jira. If you are able, attach a patch that
includes a test.
Thanks,
Doug
On Tue, Dec 9, 2014 at 12:30 PM, Jeffrey Mullins (BLOOMBERG/ BOSTON)
<jmullin...@bloomberg.net> wrote:
> Hi,
>
> When invoking GenericData.get().validate() with a union schema containing
> multiple record schemas an IndexOutOfBoundsException is possible.
>
> The exception results under the following conditions:
> 1) Union contains multiple record schemas
> 2) The first record schema in the union contains more fields than
> subsequent record schemas
> 3) validate() is invoked with the union schema and a record of one of the
> subsequent record schemas.
>
> Below is a simple unit test which reproduces the exception.
>
> @Test
> public void testValidateUnion() {
> Schema type1Schema = SchemaBuilder.record("Type1")
> .fields()
> .requiredString("myString")
> .requiredInt("myInt")
> .endRecord();
>
> Schema type2Schema = SchemaBuilder.record("Type2")
> .fields()
> .requiredString("myString")
> .endRecord();
>
> Schema unionSchema = SchemaBuilder.unionOf()
> .type(type1Schema).and().type(type2Schema)
> .endUnion();
>
> GenericRecord record = new GenericData.Record(type2Schema);
> record.put("myString", "myValue");
> GenericData.get().validate(unionSchema, record);
> }
>
> I'm happy to submit a JIRA for this issue, but thought I'd check in with the
> mailing list before doing so.
>
> Thanks,
> Jeff
