voonhous opened a new issue, #20670: URL: https://github.com/apache/hudi/issues/20670
### Problem The Trino connector never relates `HoodieSchema` to Trino types in one place. Column types come from the Hive type strings in the metastore (`HoodieSchema` to Hive type through Hive sync, then Hive type to Trino type through Trino's Hive translation), while the merge path works on records typed by `HoodieSchema` that hold Trino objects. Code that needs both sides has to know both hops. Two current examples: - #20648: the merge buffer's spill serializer has to know that UUID and VECTOR arrive as `varbinary` and BLOB and VARIANT as rows, because that is how Hive sync declares them. - #20012: `CREATE TABLE` adds its own Trino to `HoodieSchema` conversion and rejects UUID and TIME only because the metastore has no such types. ### Proposal Add `TrinoTypeMapper` to `hudi-trino`: a plain class that maps each `HoodieSchemaType` to exactly one Trino type and rejects the types Trino cannot hold. It is the Trino entry of the type plane in RFC-104 (#18661, Workstream D), introduced without the shared interface so that connector work can start using it; implementing the interface later is a hook-up. Scope of the first PR: - `toTypeSystem(HoodieSchema)` only, with tests for every `HoodieSchemaType`. - No caller in the connector. Column types still come from the metastore. Follow-ups, each on its own: - The reverse direction (Trino type to `HoodieSchema`), folding in the converter from #20012. - Moving the connector's column types onto the mapper. This changes what Trino reports for UUID, TIME, TIMESTAMP and VECTOR columns, so it needs gating. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
