How to handle nested AVRO Schemas: Union, several Files

nestedAvroTest.java

public void nestedSchemaTest() throws IOException, NoSuchAlgorithmException {

  Schema.Parser parser = new Schema.Parser();
  
  // nested schemas as union
  parser.parse(new File("schemaUnion.avsc"));
  
  // load dependent first, then "derived" schema
  parser.parse(new File("schema1.avsc"));
  parser.parse(new File("schema2.avsc"));

  System.out.println(parser.getTypes());
  
  System.out.println(parser.getTypes().get("person").toString(true));
  System.out.println(parser.getTypes().get("person2").toString(true));
}

schema1.avsc

{
  "type": "record",
  "name": "Address2",
  "fields": [
    {
      "name": "streetaddress",
      "type": "string"
    },
    {
      "name": "city",
      "type": "string"
    }
  ]
}

schema2.avsc

{
  "type": "record",
  "name": "person2",
  "fields": [
    {
      "name": "firstname",
      "type": "string"
    },
    {
      "name": "lastname",
      "type": "string"
    },
    {
      "name": "address",
      "type": "Address2"
    }
  ]
}
]

schemaUnion.avsc

[
  {
    "type": "record",
    "name": "Address",
    "fields": [
      {
        "name": "streetaddress",
        "type": "string"
      },
      {
        "name": "city",
        "type": "string"
      }
    ]
  },
  {
    "type": "record",
    "name": "person",
    "fields": [
      {
        "name": "firstname",
        "type": "string"
      },
      {
        "name": "lastname",
        "type": "string"
      },
      {
        "name": "address",
        "type": "Address"
      }
    ]
  }
]

Is there any dynamic way to load the dependent first, then "derived" schema

At least I haven't found a "AVRO-native" way to do this. Some ideas:

• You can define everything in one place so you only have "one" schema with the substructure defined on first usage
• Wrap the parser with some exception-handling to get the unresolved dependencies, resolve and retry. Maybe quick and dirty, but too much random trial-and-error for me t.b.h.
• more clever solution: preload and parse the schemas with some king of JSON-parser, build a list/tree of the dependencies and generate a loading order for the schema files by traversing this tree