zhangyue19921010 opened a new issue, #11333: URL: https://github.com/apache/paimon/issues/11333
### Search before asking - [x] I searched the issues and found no matching report. ### Paimon version 2.2-SNAPSHOT, master at `61f9981bc3b265da12446bf48eb0db441fb2da86`. ### Compute Engine Java file-format API, JDK 11. ### Minimal reproduce step Write an Avro file with a string-keyed map using `GenericDatumWriter`, then read it as Paimon `MAP<STRING, STRING>`: - A map with Avro value type `"string"` fails when the requested Paimon value type is nullable. - A map with Avro value type `["string", "null"]` also fails. The generic nullable-union reader already supports this ordering. The regression includes non-null and null values, empty maps, multiple records and a following string field. It also passes decoded maps directly to a Paimon Avro writer with a different value-nullability declaration, using only non-null values when the destination requires them, and reads the output with Avro's `GenericDatumReader`. ### What doesn't meet your expectations? These valid inputs should read and rewrite correctly. The optimized map reader assumes the requested nullability and null-first encoding match the file, which can cause malformed-length or EOF errors. Raw-byte rewriting with different value nullability produces an invalid destination encoding. ### Anything else? Use the existing schema-aware reader for noncanonical value layouts. Keep byte-backed maps for compatible layouts, and check their encoded value nullability before reusing bytes during a write. ### Are you willing to submit a PR? - [x] I'm willing to submit a PR! -- 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]
