voonhous commented on PR #19687:
URL: https://github.com/apache/hudi/pull/19687#issuecomment-5413560990

   Head `fab55f1948ca`, three commits on `8389e2646ad3`:
   
   - **Spark 4.0**: `VariantType` is an `AtomicType` on 4.0 and batch reads are 
disabled for variant only on 4.1+, so the default COW/MOR base read was 
vectorized through Spark's own `ParquetReadSupport` and never reached the guard 
(Spark failed it with the column-less `WRONG_NUM_FIELDS`). 
`Spark40ParquetReader.read` now rejects off the footer before either reader; a 
`PushVariantIntoScan` rewrite struct is rejected on shape.
   - **Hive guards**: legs for the 3-level `list`/`element` layout, map values 
and a realtime split with the file group reader disabled; native log names skip 
the `.parquet` gate.
   - **Bootstrap**: the footer-schema read also shifted every data column by 
one on partitioned tables whose external files omit the partition column 
(legacy merge path); pinned by the partitioned leg, and a non-nullable evolved 
column now fails naming it.
   - **Sort columns**: the error names the unorderable member (`'s.m'`, 
`'arr[]'`).
   - **Tests**: legacy `BaseFileOnlyRelation` guard pinned in 
`TestLegacyParquetReadPath`; nested row-writer leg reads through both parquet 
readers; shared guard fixtures; duplicate legs dropped.
   


-- 
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]

Reply via email to