JingsongLi commented on PR #10409: URL: https://github.com/apache/paimon/pull/10409#issuecomment-6081284114
Independent validation of the existing [P1] Python-to-Java Vortex compatibility finding: **release gate BLOCK** on this head. I ran the actual mixed-table write/read boundary locally with JDK 8, Arrow 19 and the current Java JNI 0.73.0: | Python writer | Compact option | Java single-vector table read | Java multi-vector table read | | --- | --- | --- | --- | | vortex-data 0.87.0 | omitted/default false | ERROR: `Encoding not found in registry: 1` | same error | | vortex-data 0.70.0 (previous pin) | omitted/default false | PASS | PASS | | vortex-data 0.87.0 | true | ERROR: `Encoding not found in registry: 1` | same error | The tests were `JavaPyVortexE2ETest#testJavaReadVectorDedicatedFile` and `#testJavaReadMultiVectorDedicatedFile`, with `run.e2e.tests=true`; none were skipped. The Python-only checks passed (125 tests and 22 subtests, covering the changed readers/writers, schema reuse, Variant and HDFS paths), as did project Flake8. Thus Python round trips alone do not establish interoperability. The dependency upgrade changes default writes, so the compact feature flag cannot protect existing Java consumers. Please verify a compatible Java reader or backward-compatible Python output before merging, retaining both old-file and new-file mixed tests. Updating compatible readers before enabling new-format writers is necessary for a rolling rollout; reverting the Python dependency alone does not rewrite files already committed in the new encoding. I did not find a separate additional regression in the new slicing/schema-cache/compact routing logic. This adds an executable old/new control to the existing finding rather than another duplicate issue. -- 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]
