adriangb opened a new pull request, #22027:
URL: https://github.com/apache/datafusion/pull/22027
## Which issue does this PR close?
- Closes #.
## Rationale for this change
`Interval::mul`, `Interval::div`, `Interval::intersect`, `Interval::union`,
and `Interval::contains` all asserted that both operands had identical data
types. This caused internal errors during interval propagation for ordinary SQL
queries that mix `Decimal128` precisions/scales — for example `numeric /
count(*)`, where the operands resolve to `Decimal128(38, 10)` and
`Decimal128(20, 0)`:
```
Internal error: Assertion failed: dt.clone() == rhs_type.clone()
(left: Decimal128(38, 10), right: Decimal128(20, 0)):
Intervals must have the same data type for division
```
`Interval::add` and `Interval::sub` already used `BinaryTypeCoercer` to find
a common arithmetic type; this PR brings `mul` and `div` in line. Once
`mul`/`div` coerce, the result of an arithmetic op fed into `intersect` (e.g.
by the CP solver in `cp_solver.rs`) may have a different type than the child
interval, so `intersect`/`union`/`contains` are also relaxed to coerce via
`comparison_coercion` (matching the existing pattern in
`Interval::contains_value`).
## What changes are included in this PR?
- Replace the type-equality asserts in `Interval::mul` and `Interval::div`
with a new `coerce_operands` helper that uses
`BinaryTypeCoercer::get_result_type` and casts both intervals to the common
type when they differ.
- Replace the type-equality asserts in `Interval::intersect`,
`Interval::union`, and `Interval::contains` with a new `coerce_for_comparison`
helper that uses `comparison_coercion` and casts both intervals to the common
type when they differ.
- Update the doc comments on the affected methods to document the new
coercion behavior.
The same-type fast path is preserved (no allocation/cast when both intervals
already share a type), so this should be a no-op for existing call sites.
## Are these changes tested?
Yes:
- New unit test `test_mul_div_mismatched_operand_types` in
`interval_arithmetic.rs` exercising `Decimal128(38, 10)` / `Decimal128(20, 0)`
and `Decimal128 × Int64` for both `mul` and `div`.
- New regression query in `decimal.slt` exercising the customer-shape
`numeric / bigint` division through the optimizer's interval-propagation path.
Without this fix the SLT query fails with an internal error; with it, it
returns rows.
- Existing `datafusion-expr-common`, `datafusion-physical-expr`,
`datafusion-physical-plan`, and full `sqllogictests` suites all pass.
## Are there any user-facing changes?
Queries that previously hit `Internal error: Intervals must have the same
data type ...` (or the equivalent for intersect/union/contains) during
interval/stats propagation will now succeed. No public API changes — these
methods kept their signatures; the documented preconditions are relaxed.
🤖 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]
