andygrove opened a new issue, #4489:
URL: https://github.com/apache/datafusion-comet/issues/4489

   ## Describe the bug
   
   Spark 4.0+ supports collated `StringType` (e.g. `STRING COLLATE 
UTF8_LCASE`). `CometCast.isSupported` matches both source and target string 
types via `case (DataTypes.StringType, _) => ...` and `case (_, 
DataTypes.StringType) => ...`, where `DataTypes.StringType` is the singleton 
default-collation `StringType` instance.
   
   Scala pattern equality means a non-default-collation `StringType` instance 
should NOT match `DataTypes.StringType`, and the cast would fall through to the 
default `unsupported(...)` branch and fall back to Spark. This appears to be 
the intended (safe) behaviour, but it is implicit: there is no 
`isStringCollationType` guard like the other string-touching serdes use 
(`arrays.scala::CometArrayIntersect`, 
`QueryPlanSerde::supportedScalarSortElementType`). And there is no test that 
asserts the fallback for `CAST(c AS STRING COLLATE UTF8_LCASE)` and `CAST(c AS 
STRING)` when `c` is collated.
   
   If the equality ever yields true (e.g. via a future refactor of 
`StringType.equals`), Comet would silently route collated-string casts through 
the byte-oriented native path, producing incorrect results for downstream 
collation-aware comparisons / aggregations / hashing.
   
   Surfaced by the cast audit (collection PR queue). Tracked under the umbrella 
#2190 for Spark 4.0 collation support.
   
   ## Expected behavior
   
   Either:
   
   1. Add an explicit `isStringCollationType` guard in `CometCast.isSupported` 
so the fallback is declared rather than relying on Scala pattern semantics, OR
   2. Add tests asserting that `CAST` to/from a non-default `StringType` 
collation falls back to Spark and does not run native.
   
   ## Additional context
   
   - Comet matrix: `CometCast.scala`, lines around the `case 
(DataTypes.StringType, _)` and `case (_, DataTypes.StringType)` matches.
   - Umbrella issue: #2190 (`[Spark 4.0] Add string collation support`).
   


-- 
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