NullPointerException when schema uses field aliases

[Ryan Holmes]

I'm getting an NPE when writing an Avro file with one schema and reading
back with a slightly modified schema. The only difference between the writer
and the reader schemas is that the reader renames a single field and adds an
alias. I'm just testing Avro's schema evolution features, so the code is
indeed as trivial as it appears. Here's the stack trace I get when reading:

Exception in thread "main" java.lang.NullPointerException
at org.apache.avro.Schema$Field.hashCode(Schema.java:421)
at java.util.AbstractList.hashCode(AbstractList.java:527)
at org.apache.avro.Schema$RecordSchema.hashCode(Schema.java:602)
at
org.apache.avro.io.parsing.ValidatingGrammarGenerator$LitS.hashCode(ValidatingGrammarGenerator.java:133)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator$LitS2.hashCode(ResolvingGrammarGenerator.java:461)
at java.util.HashMap.get(HashMap.java:300)
at
org.apache.avro.io.parsing.ResolvingGrammarGenerator.resolveRecords(ResolvingGrammarGenerator.java:197)
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:76)
at org.apache.avro.io.ResolvingDecoder.<init>(ResolvingDecoder.java:46)
at
org.apache.avro.generic.GenericDatumReader.getResolver(GenericDatumReader.java:93)
at
org.apache.avro.generic.GenericDatumReader.read(GenericDatumReader.java:103)
at org.apache.avro.file.DataFileStream.next(DataFileStream.java:198)
at org.apache.avro.file.DataFileStream.next(DataFileStream.java:185)
...


The writer schema:

{ "type": "record",
  "name": "Message",
  "namespace": "com.edmunds.avro.data",
  "fields" : [
    {"name": "id", "type": "long"},
    {"name": "time", "type": "long"},
    {"name": "body", "type": ["string", "null"]}
  ]
}


The reader schema:

{ "type": "record",
  "name": "Message",
  "namespace": "com.edmunds.avro.data",
  "fields" : [
    {"name": "messageId", "type": "long", "aliases": ["id"]},
    {"name": "time", "type": "long"},
    {"name": "body", "type": ["string", "null"]}
  ]
}


Here's the writing code:

    private void writeMessages() {

        File file = getAvroFile();

        if(file.exists()) {
            file.delete();
        }

        DatumWriter<Message> writer = new
SpecificDatumWriter<Message>(Message.class);
        DataFileWriter<Message> dataFileWriter = new
DataFileWriter<Message>(writer);
        try {
            dataFileWriter.create(Message.SCHEMA$, file);
            for(Message m : createMessages()) {
                dataFileWriter.append(m);
                System.out.println("Wrote message: " + objToString(m));
            }
            dataFileWriter.close();
        } catch (IOException e) {
            e.printStackTrace();
        }
    }


And the reading code:

    private void readMessages() {

        File file = getAvroFile();
        DatumReader<Message> reader = new
SpecificDatumReader<Message>(Message.class);

        DataFileReader<Message> dataFileReader;
        try {
            dataFileReader = new DataFileReader<Message>(file, reader);
        } catch (IOException e) {
            e.printStackTrace();
            return;
        }

        while(dataFileReader.hasNext()) {
            Message m = dataFileReader.next();
            System.out.println("Read message: " + objToString(m));
        }
    }



I'm using Avro 1.4.1 and the avro-maven-plugin (updated to Avro 1.4.1) to
generate the Message class(es). My code works fine when I use the exact same
schema for the reader and writer. What am I doing wrong?