sunchao opened a new issue, #5500:
URL: https://github.com/apache/datafusion-comet/issues/5500

   ### Describe the bug
   
   `CometObjectHashAggregate` returns `None` when Comet shuffle is disabled 
without attaching a fallback reason. This is a pre-existing diagnostics gap 
identified while reviewing #5421, not a regression from that PR's 
aggregate-buffer repair.
   
   The shuffle guard in [the reviewed 
source](https://github.com/apache/datafusion-comet/blob/cfa311307b8430e7f70a9744122f2aa7bdd30c55/spark/src/main/scala/org/apache/spark/sql/comet/operators.scala#L2024-L2036)
 bypasses `doConvert` and does not call `withFallbackReason`. The strict 
unexplained-fallback check can therefore report an unannotated decline, or a 
reason collected from another node can obscure this particular cause.
   
   ### Steps to reproduce
   
   Add a focused serde/planner regression for `ObjectHashAggregateExec`, such 
as grouped `collect_list` over an eligible Parquet input with 
`spark.sql.execution.useObjectHashAggregateExec=true` and 
`spark.comet.shuffle.enabled=false`. Check the shuffle-disabled guard directly 
and the resulting extended-explain reason; run the regression with 
`spark.comet.explain.fallback.strict.enabled` both enabled and disabled.
   
   The missing annotation is verified from the guard's source. A separate 
end-to-end strict-mode reproduction has not been run for this issue.
   
   ### Expected behavior
   
   The deliberate decline should carry an accurate shuffle-disabled reason, 
without changing aggregate execution eligibility or masking an unrelated 
missing explanation.
   
   ### Additional context
   
   The [review 
discussion](https://github.com/apache/datafusion-comet/pull/5421#issuecomment-5429161669)
 explicitly separates this issue from the buffer-compatibility fix. #5499 
tracks stale fallback-reason provenance; this issue covers a specific missing 
reason at its source.
   


-- 
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