andygrove opened a new issue, #4492:
URL: https://github.com/apache/datafusion-comet/issues/4492
## Describe the bug
Spark's `spark.sql.legacy.castComplexTypesToString.enabled` configuration
switches the format used by `CAST(<struct|array|map> AS STRING)`:
- Default (`false`): structs render as `{a, b, c}`, arrays as `[1, 2, 3]`,
nulls as `null`.
- Legacy (`true`): structs render as `[a, b, c]`, arrays as `[1, 2, 3]`,
nulls render as `""` (empty string).
Comet's native struct/array-to-string formatter hard-codes the default-mode
brackets and uses `cast_options.null_string = "null"`. The
`legacy_cast_complex_to_string` flag is not currently plumbed through the cast
proto, so users who set
`spark.sql.legacy.castComplexTypesToString.enabled=true` will see Comet output
differ from Spark for `CAST(<complex> AS STRING)`.
Surfaced by the cast audit (collection PR queue).
## Steps to reproduce
```sql
SET spark.sql.legacy.castComplexTypesToString.enabled = true;
SELECT CAST(struct(1, 2, null) AS STRING);
-- Spark legacy: [1,2,]
-- Comet: {1, 2, null}
```
## Expected behavior
Either:
1. Plumb the conf through the Cast proto and switch the native formatter on
`legacy_cast_complex_to_string`, OR
2. Downgrade `(StructType|ArrayType|MapType, StringType)` casts to
`Incompatible(Some(...))` when the conf is enabled.
## Additional context
- Native impl: `native/spark-expr/src/conversion_funcs/cast.rs`
(struct/array string formatters)
- Comet matrix: `CometCast.canCastToString`
- Spark conf: `SQLConf.LEGACY_COMPLEX_TYPES_TO_STRING`
--
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]
