MaxGekk opened a new pull request, #56805:
URL: https://github.com/apache/spark/pull/56805
### What changes were proposed in this pull request?
Extend common-type resolution so two `TIME(p)` operands of differing
fractional-seconds precision resolve to `TIME(max(p1, p2))`:
- `TypeCoercionHelper.findWiderDateTimeType` now widens two `TIME` operands
to the larger precision. Because both the default (`TypeCoercion`) and ANSI
(`AnsiTypeCoercion`) `findTightestCommonType` implementations delegate to this
single method, the change covers UNION/INTERSECT/EXCEPT,
COALESCE/CASE/IF/NULLIF, GREATEST/LEAST, IN lists/subqueries, and array/map
literals.
- `Cast.canANSIStoreAssign` allows lossless TIME widening (smaller -> larger
precision) as a silent store assignment, while narrowing stays
explicit-CAST-only. `UpCastRule.canUpCast` already keeps cross-precision TIME
explicit-only, mirroring the nanosecond-timestamp precedent.
Cross-family pairs (TIME vs DATE / TIMESTAMP) remain incomparable.
### Why are the changes needed?
ANSI SQL (ISO/IEC 9075-2, result-type rule for a set of comparable datetime
types) requires that across datetime operands of differing fractional-seconds
precision the result type is the datetime type with the largest
fractional-seconds precision. Previously `findWiderDateTimeType` returned
`None` for any pair involving `TIME`, so set and conditional operations over
mixed `TIME(p)` raised an analysis error and required an explicit `CAST`. The
widening cast (`TIME(p1) -> TIME(p2)` via `truncateTimeToPrecision`) already
existed, so this was purely a common-type-resolution gap.
### Does this PR introduce _any_ user-facing change?
Yes, within the unreleased TIME type (`@Unstable`). Set and conditional
operations over mixed `TIME(p)` now resolve `TIME(max(p))` without an explicit
cast; values from the smaller-precision side are widened losslessly. Inserting
a smaller-precision TIME value into a larger-precision TIME column now
succeeds, while the reverse (narrowing) still requires an explicit cast.
For example:
```sql
SELECT typeof(c), c FROM (
SELECT '12:34:56.789' :: TIME(3) AS c
UNION ALL SELECT '01:02:03.456789' :: TIME(6)) ORDER BY c;
-- time(6) 01:02:03.456789
-- time(6) 12:34:56.789
```
### How was this patch tested?
- Added type-coercion unit tests in `TypeCoercionSuite` and
`AnsiTypeCoercionSuite` (widening to max precision, equal precision, and
TIME-vs-DATE/TIMESTAMP returning `None`).
- Added a store-assignment / up-cast contract test in `CastSuiteBase`.
- Added golden-file coverage in `time.sql` for UNION, COALESCE, CASE,
NULLIF, GREATEST/LEAST, array/map, IN, and store-assignment INSERTs (widening
succeeds, narrowing fails without an explicit cast).
- Ran the affected catalyst suites and the `time.sql` `SQLQueryTestSuite`.
### Was this patch authored or co-authored using generative AI tooling?
Generated-by: Cursor (Claude Opus 4.8)
--
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]
