hutiefang76 commented on PR #12592: URL: https://github.com/apache/seatunnel/pull/12592#issuecomment-6252036150
Updated in de8f9d02c. The five points now stand as follows: 1. Pre-epoch aliases still use the legacy plain `getTimestamp` getter. A real-file source regression pins pre-epoch DST-gap and Gregorian-cutover behavior in four JVM zones through both `table_path` and `query`. Those read limitations remain deferred; the description and EN/ZH docs explicitly separate them from the sink fix. 2. The WARN was already gated on an actual pre-epoch alias value and contains no driver version. Ordinary post-epoch alias rows do not trigger it. 3. Alias classification is now computed once per ResultSet and rebuilt when the ResultSet identity changes. The bounded cache is transient; tests cover switching back, absent metadata and serialization. Calendar stays local because it is mutable. 4. `reconvertTimestampType` already preserves scale: 0 → TIMESTAMP_S, 1–3 → TIMESTAMP_MS, 4–6 → TIMESTAMP, 7–9 → TIMESTAMP_NS. Existing automatic-DDL and real-sink tests cover this mapping and nanosecond storage; it no longer silently maps scale 9 to plain TIMESTAMP. 5. The Javadoc now states that detection uses metadata type names, is independent of driver version and will not automatically switch to a future typed getter. On JDK 17, all 114 DuckDB tests passed with zero failures/errors/skips, along with scoped Spotless and root `./mvnw -q -DskipTests verify`. Root verify skips tests. The MySQL container E2E has not been run locally, and the PR description now says so explicitly. Remote Build remains a separate readiness requirement. <sub>Drafted and posted by an AI assistant for @hutiefang76.</sub> -- 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]
