RishavTiwari25 opened a new pull request, #26258:
URL: https://github.com/apache/datafusion/pull/26258
## Which issue does this PR close?
- Closes #13850.
## Rationale for this change
Multiplying an interval by a number, which PostgreSQL and DuckDB both
support, fails in DataFusion:
```sql
> SELECT interval '1 day' * 2;
Error during planning: Cannot get result type for temporal operation
Interval(MonthDayNano) * Int64: ...
Invalid interval arithmetic operation: Interval(MonthDayNano) *
Interval(MonthDayNano)
> SELECT interval '1 hour' * 3.5;
Arrow error: Invalid argument error: Invalid interval arithmetic operation:
Interval(MonthDayNano) * Float64
```
Dividing by an integer (`interval '1 day' / 2`) fails the same way, while
`interval '1 hour' / 1.5` already works.
The arrow side of #13850 (apache/arrow-rs#6335) is available in the arrow 60
release DataFusion uses: the kernels support `Interval * Int64`,
`Interval(MonthDayNano) * Float64` and `Interval(MonthDayNano) / Float64`. Two
things on the DataFusion side still got in the way:
- `temporal_math_coercion` coerced the numeric operand to
`Interval(MonthDayNano)`, which arrow rejects as `Interval * Interval`.
- Result type probing and physical evaluation used `mul_wrapping`, but
arrow's interval kernels only implement the checked `mul`.
## What changes are included in this PR?
- A new `interval_number_coercion` handles `interval * number`, `number *
interval` and `interval / number`: the interval becomes
`Interval(MonthDayNano)`, an integer factor becomes `Int64` (exact
multiplication), and any other numeric factor or divisor becomes `Float64`. The
old `Interval * number => Interval * Interval` arms are removed, so `interval +
2` now fails with the generic "cannot coerce" error instead of a cast error.
- `BinaryTypeCoercer::get_result` probes multiplication with `mul`, and
`BinaryExpr` evaluates `interval * number` with `mul`, since the interval
kernels do not implement the wrapping variant.
## What is the testing strategy for this PR?
- `datetime/arith_interval_double.slt`: the cases that were documented as
failing now have expected results.
- New `datetime/arith_interval_integer.slt` covers integer factors on both
sides, division by an integer, `NULL` operands, column inputs, and the result
type.
- Existing `datafusion-expr-common` and `datafusion-physical-expr` tests.
## Are there any user-facing changes?
`interval * integer`, `integer * interval`, `interval * float`, `float *
interval` and `interval / number` are now supported and return
`Interval(MonthDayNano)`.
I used an AI coding assistant while working on this change; I have read the
diff and the tests and can walk through them.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
--
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]
