DanielLeens commented on PR #13070: URL: https://github.com/apache/seatunnel/pull/13070#issuecomment-6594435038
Addressed in ba7318ed9b for [review 5746417219](https://github.com/apache/seatunnel/pull/13070#pullrequestreview-5746417219). **Issue 1 (deny-list retypes `vector` in expressions and after keywords):** fixed at two levels. - `MySqlAntlrDdlParser#parse` keeps Debezium's original `MySqlLexer` for every statement, unless the text mentions `vector` and fails to parse with that lexer. The check runs before the tree walk, so it has no side effect on `Tables`. Any DDL that parses today, on any MySQL version, takes exactly today's path, whatever the heuristic does. - Inside a statement that does need it, the deny-list is replaced by a positive check. `vector` is retyped only after a column name in `( name`, `, name`, `ADD name`, `[ADD|MODIFY] COLUMN name` or `CHANGE [COLUMN] old new`. Index, key, constraint, partition and alter-clause keywords never count as a name. - Your statements are now regressions in `testParseVectorColumnTypeWithoutCapturingVectorNames`: `ADD COLUMN doubled INT AS (vector * 2)`, `ADD CONSTRAINT ck CHECK (vector > 0)`, `ORDER BY vector`, `CREATE DATABASE vector` and `ADD COLUMN extra INT, DROP vector`. They assert the column stays `INT`, or is dropped. A mixed `CREATE TABLE mixed (vector INT, v VECTOR(2), CHECK (vector > 0))` asserts both columns. **Issue 2 (0.40.2 → 0.40.7):** done. The jar matches the Maven Central SHA-1, all classes are major version 52, and `zstd-jni` is the same 1.5.0-2 compile dependency. Its API compared with 0.40.2 differs only in `TransactionPayloadEventData` uncompressed size (`int` → `long`), which neither Debezium 1.9.8 nor SeaTunnel calls. `BufferedSocketInputStream` and `deserializeVector` are unchanged. The POM comment records why 0.40.x is used. **Issue 3 (no MySQL 5.x execution):** not added in this PR. The MySQL CDC IT lane already used ~134 of 180 minutes on a1f31ae, and a full 5.7 suite would add about 30 minutes. The PR description now states that 5.x compatibility of the new client rests on source analysis only. **Issue 4 (TiDB `VECTOR`):** `TiDBCatalog#buildColumn` keeps rejecting `VECTOR` with the same COMMON-17 error as before, covered by `TiDBTableOptionsTest#testTiDBCatalogStillRejectsVectorColumn`. **Docs:** both MySQL-CDC pages note the little-endian float32 layout (4 bytes per dimension, not `FLOAT_VECTOR`). Both JDBC MySQL source tables list `VECTOR` → `BYTES`. That JDBC path has no IT in this PR, as the description says. The description's zstd-jni line is corrected. Only Spotless ran locally. Compile and tests rely on the fork Build for ba7318ed9b. -- 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]
