[AVRO-1017] org.apache.avro.generic.GenericData.resolveUnion does not deal correctly with a union of two records with (different) namespace definitions

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.6.1
• Component/s: java
• Assignee: Unassigned
• Reporter: Joseph Adler
• Created: 06/Feb/12 22:20
• Updated: 06/Feb/12 23:33

Description

Suppose that you define two different avro objects:

{"name":"A", "namespace":"com.something", "type":"record", ...} {"name":"B", "namespace":"com.anotherthing", "type":"record", ...}

Then define a union of the two:

{"name":"bothAAndB", "type":[

{"name":"A", "namespace":"com.something", "type":"record", ...}

,

{"name":"B", "namespace":"com.anotherthing", "type":"record", ...}

]}

If you write an object of type A or B into a GenericRecord and try to use org.apache.avro.generic.GenericDatumWriter to store the object, you will get an org.apache.avro.UnresolvedUnionException exception.

Very specifically, it looks like org.apache.avro.UnionSchema.getIndexNamed is being called by org.apache.avro.generic.GenericDatumWriter.resolveUnion with the short name of the record, but that the org.apache.avro.UnionSchema object contains the full names of the objects. So, org.apache.avro.UnionSchema.getIndexNamed doesn't find a match and Avro fails to find an object in the union that matches the object in the datum.

So... I'd like to see one of two fixes: (1) an informative error message telling me that I can't form a union of records from different namespaces, or (2) a successful lookup of the schema and record writing.