linhongliu-db opened a new pull request, #59889:
URL: https://github.com/apache/spark/pull/59889
### What changes were proposed in this pull request?
Fix decimal remainder when the integral quotient exceeds the 39-digit math
context even though the remainder is representable. Keep the existing
context-based calculation first and retry with exact remainder only if it
throws `ArithmeticException`.
### Why are the changes needed?
`Decimal.%` can throw `ArithmeticException: Division impossible` for a large
dividend and a small divisor. For example:
```sql
SELECT
CAST('99999999999999999999999999999999999999' AS DECIMAL(38, 0)) %
CAST('0.0007' AS DECIMAL(4, 4));
```
The representable result is `0.0004`. The failure comes from the
intermediate quotient's precision, rather than the remainder.
Trying the existing operation first preserves its successful results and
extreme-scale behavior. The fallback is limited to the failed arithmetic call;
decimal construction and the existing zero-divisor handling are unchanged.
### Does this PR introduce _any_ user-facing change?
Yes. Decimal `%` and `pmod` succeed for inputs whose representable remainder
previously failed because of the intermediate quotient limit. Existing
successful calculations continue to use the same operation.
### How was this patch tested?
Added regressions cover large quotients, both operand signs, quotient
precision boundaries, `%` and `pmod`, scale preservation, zero divisors, and
negative and extreme scales. They also cover inputs where the existing
operation succeeds but exact division cannot materialize its quotient, and
cases where both attempts fail.
`DecimalSuite` and `ArithmeticExpressionSuite`: all 77 tests passed with
SBT. The new large-quotient regressions reproduced `Division impossible` before
the production change. SQL API and Catalyst test Scala style checks passed.
### Was this patch authored or co-authored using generative AI tooling?
Generated-by: Codex CLI 0.164.0 (GPT-6)
--
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]
