Smallfu666 commented on issue #12607:
URL: https://github.com/apache/gluten/issues/12607#issuecomment-5212494970

   I picked this up but haven't been able to reproduce the mismatch on current 
main.
   
   - Direct `CAST(<double> AS DECIMAL(18,2))` matches vanilla Spark on the 
tie-boundary values, including `0.275` and `nextDown(0.275)` (both give the 
same result as Spark).
   - A grid of `CAST(c_0/c_1 AS DECIMAL(18,2))` over 6400 double-division pairs 
(Spark 3.5 + Velox, vanilla vs Gluten in separate JVMs) shows **0 mismatches**.
   
   The report doesn't include the actual `c_0`/`c_1` values, the data source, 
or versions, so I can't pin the cause. Could you share a minimal reproducer — 
ideally a `VALUES`-based SQL with the exact `c_0`/`c_1`, the Spark/Gluten/Velox 
versions, and whether the columns come from a Parquet/ORC scan (plus 
`spark.sql.ansi.enabled` / `spark.sql.decimalOperations.allowPrecisionLoss`)?
   
   Unassigning myself so someone with a reproducer can pick it up.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to