andygrove opened a new pull request, #4678:
URL: https://github.com/apache/datafusion-comet/pull/4678

   ## Which issue does this PR close?
   
   Part of #4673.
   
   ## Rationale for this change
   
   The serde framework intends that an expression declares whether it can be 
accelerated through `getSupportLevel`, with the central dispatcher in 
`QueryPlanSerde` translating `Incompatible`/`Unsupported` into the appropriate 
fallback. Several aggregate serdes instead made these decisions inside 
`convert` by calling `withFallbackReason` directly, even though the decisions 
are statically determinable from the expression.
   
   ## What changes are included in this PR?
   
   Move static support decisions out of `convert` and into `getSupportLevel` 
for the aggregate expressions:
   
   - `Min`, `Max`: unsupported data type and the strict-floating-point gate, 
via a shared `AggSerde.minMaxSupportLevel` helper.
   - `Average`, `Sum`, `BitAndAgg`, `BitOrAgg`, `BitXorAgg`: unsupported data 
type.
   - `BloomFilterAggregate`: unsupported child data type (the static type 
filter that was embedded in the convert guard).
   
   This is behavior-preserving: every case that previously fell back to Spark 
continues to fall back. The remaining `withFallbackReason` calls in this file 
are child-reason rollups, including the `serializeDataType` rollups gated on 
`dataType.isEmpty`, which are the intended use inside `convert`.
   
   ## How are these changes tested?
   
   Covered by existing tests in `CometAggregateSuite` (80 tests covering 
min/max/sum/avg/bit aggregates and decimal/float paths) and the 
`bloom_filter_agg` test in `CometExecSuite`, which exercise both the supported 
and fallback paths.
   


-- 
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