array with union records throws exception

[gaborbernat]

Reproduce with an example:

use avro_rs::Schema;

fn main() {
    // let raw = fs::read_to_string("/Users/bgabor8/git/dqc/data-benchmark/trans-3.1.avsc").unwrap();
    // let schema = Schema::parse_str(&raw).unwrap();
    // println!("{:?}", schema);

    let raw = r#"
    {
      "name": "Data",
      "type": "record",
      "fields": [
        {
          "name": "sources",
          "type": {
            "type": "array",
            "items": [
              {
                "name": "Url",
                "type": "record",
                "fields": [{ "name": "url", "type": "string" }]
              },
              {
                "name": "Source",
                "type": "record",
                "fields": [{ "name": "data", "type": "string"}]
              }
            ]
          }
        }
      ]
    }
    "#;
    Schema::parse_str(raw).unwrap();
}

value: ParseSchemaError("Unions cannot contain duplicate types")

For reference the python avro behaviour (demonstrating that the construction is valid):

import json
from io import BytesIO

from avro.io import BinaryEncoder, DatumWriter, DatumReader, BinaryDecoder
from avro.schema import parse

json_schema_raw = {
    "name": "Data",
    "type": "record",
    "fields": [
        {
            "name": "sources",
            "type": {
                "type": "array",
                "items": [
                    {"name": "Url", "type": "record", "fields": [{"name": "url", "type": "string"}]},
                    {"name": "Source", "type": "record", "fields": [{"name": "data", "type": "string"}]},
                ],
            },
        }
    ],
}
schema = parse(json.dumps(json_schema_raw))

writer = DatumWriter()
with BytesIO() as serialized_binary:
    binary_encoder = BinaryEncoder(serialized_binary)
    writer.write_record(schema, {"sources": [{"url": "a"}, {"data": "b"}]}, binary_encoder)
    raw = serialized_binary.getvalue()

print(raw)

reader = DatumReader()
with BytesIO(raw) as binary_input:
    binary_decoder = BinaryDecoder(binary_input)
    record = reader.read_record(schema, schema, binary_decoder)
print(record)

b'\x04\x00\x02a\x02\x02b\x00'
{'sources': [{'url': 'a'}, {'data': 'b'}]}

The issue seems to be here https://github.com/flavray/avro-rs/blob/master/src/schema.rs#L330, while iterating through the schema records the SchemaKind::from(schema) identifies both Url and Source from above as Schema::Null, hence the duplicate error.