At work we have a use case to read Avro records from Kafka, and to write them to one of several BigQuery tables (based on compatibility). This means that there is no single schema (with all data) that is compatible with all records we'll be reading.
In order to allow this, we'd like to be able to use the BinaryMessageDecoder without any read schema, and have it use the write schema of each record as the read schema for that record.
Pull request will follow.
Pull request submitted.