What's the reason for going through avro? Validation? I can see for Hive and Trino, we start from a DDL as a String, we parse this string and we create a Avro schema, then feed the schema to

schema_util

. I can replicate the same for Iceberg, but navigate the Iceberg type tree instead of parsing a String. But I'd like to understand why we have (or should) go through Avro?