UnionType Items in Array Type

[mladerman]

Hi there! I've hit an issue where encoding an ArrayType that has a UnionType for its items fails.

{
  "namespace": "example.avro",
  "type": "record",
  "name": "Example",
  "fields": [
    {
        "name": "values",
        "type":
            {
                "type": "array",
                "items": ["int", "string"]
            }
    }
  ]
}

Here's the example running this

var Avro = require('avsc');
var schema = Avro.parse('./example.avsc');
var data = {values: ["test"]}
schema.toBuffer(data);
TypeError: Object.keys called on non-object
    at Function.keys (native)
    at UnionType._write (avsc/lib/types.js:826:19)
    at ArrayType._write (avsc/lib/types.js:1367:19)
    at RecordType.writeExample [as _write] (eval at <anonymous> (avsc/lib/types.js:1678:10), <anonymous>:3:6)
    at RecordType.Type.toBuffer (avsc/lib/types.js:264:8)

Interestingly enough, this succeeds when passing in an empty array for values.

I've tested the same schema with the official python avro implementation to confirm that my schema is valid. Please let me know if there is any more information that I should provide!

[mtth]

Hey! This is related to how decoded unions are represented. The default UnionType expects its values to be wrapped inside objects (similarly to their JSON representation). So for your example, the value ({values: ['test']}) you are passing should be transformed to {values: [{'string': 'test'}]}.

This is required to correctly serialize all union values. Here's an example of an ambiguous case otherwise:

var type = avro.parse(['int', 'float']);
var buf = type.toBuffer(2); // Should 2 be serialized as an integer or a float?

The python implementation can cause corruption (and also requires a linear scan of branches during serialization, which is slow). If you're curious, you can take a look at #16 for more context.

Not all unions lead to ambiguous cases though so I'm actually working on adding an option to represent decoded union values without the wrapping object when possible.

[mladerman]

Ah, perfect, thank you!