[AVRO-361] Specific Compiler fails to handle union with two fixed branches

• Type: Bug
• Status: Closed
• Priority: Blocker
• Resolution: Fixed
• Fix Version/s: 1.3.0
• Component/s: java
• Hadoop Flags: Reviewed
• Assignee: Scott Carey
• Reporter: Scott Carey
• Created: 21/Jan/10 02:18
• Updated: 01/Mar/10 17:09
• Resolved: 02/Feb/10 06:51

Description

The following record fails to compile with the specific compiler:

{"name": "ipAddr", "type": "record", "fields":[
  {"name": "addr", "type": [
    {"name": "IPv6", "type": "fixed", "size": 16 },
    {"name": "IPv4", "type": "fixed", "size": 4 }]
  }
 ]
}   

The stack trace is:

org.apache.avro.AvroRuntimeException: Ambiguous union: [{"type":"fixed","name":"IPv6","size":16},{"type":"fixed","name":"IPv4","size":4}]
	at org.apache.avro.Schema$UnionSchema.<init>(Schema.java:613)
	at org.apache.avro.Schema.parse(Schema.java:874)
	at org.apache.avro.Schema.parse(Schema.java:825)
	at org.apache.avro.Schema.parse(Schema.java:709)
	at org.apache.avro.specific.SpecificCompiler.compileSchema(SpecificCompiler.java:111)\

This is on trunk:

$ svn info
Path: .
URL: http://svn.apache.org/repos/asf/hadoop/avro/trunk/lang/java
Repository Root: http://svn.apache.org/repos/asf
Repository UUID: 13f79535-47bb-0310-9956-ffa450edef68
Revision: 901469

The code for UnionSchema in Schema.java has this constructor:

public UnionSchema(List<Schema> types) {
      super(Type.UNION);
      this.types = types;
      int seen = 0;
      for (Schema type : types) {                 // check legality of union
        switch (type.getType()) {
        case UNION: 
          throw new AvroRuntimeException("Nested union: "+this);
        case RECORD:
          if (type.getName() != null)
            continue;
        default:
          int mask = 1 << type.getType().ordinal();
          if ((seen & mask) != 0)
            throw new AvroRuntimeException("Ambiguous union: "+this);
          seen |= mask;
        }
      }
    }

That allows only one member of any type other than RECORD. The spec says:

Unions may not contain more than one schema with the same
type, except for the named types record, fixed and enum.

The code above does not adhere to this.

I am attaching a patch for only this code, but a unit test with a test schema that has two records, two fixed, and two enum in it as well as one of each of the unnamed types is probably necessary as well. I am not yet familiar with the test infrastructure.
I am also not familiar with what else this may impact.

Issue Links

• relates to: AVRO-362 Add test to ensure Python implementation handles Union schema with two fixed types of different names