mbutrovich opened a new pull request, #7699: URL: https://github.com/apache/datafusion-comet/pull/7699
## Which issue does this PR close? Closes #7694. ## Rationale for this change See #7694. ## What changes are included in this PR? - `spark_decimal_div_internal` rescales the operands to the result scale instead of one digit past it, so the `i128` and `BigInt` branches each divide once per row. `/` rounds half away from zero from the remainder, and `div` truncates. - With one digit fewer in the rescale exponents, some inputs take the other branch, as #7694 describes. Both branches compute the same quotient. - `benches/decimal_div.rs` gets cases with Spark's result types: `/` on `Decimal(10, 4)` by `Decimal(10, 3)`, `div` at scale 0, the `BigInt` branch, and integer `div` on `Decimal(19, 0)` with small and with full-range values. Arrow's decimal `div` is the control, since this PR doesn't touch it. ## How are these changes tested? The tests are in `native/spark-expr/src/math_funcs/div.rs`. `test_decimal_div_matches_exact_quotient` compares every row of `/` and `div` with an exact `BigInt` quotient across scale combinations on both branches. It covers both signs, exact halves, and values at the precision bound. It also passes against `main`, so it checks that the rewrite changes no result. `test_integral_divide_overflow_through_kernel` checks the ANSI long-overflow error on both branches. ## Benchmarks TODO # AI usage disclosure I wrote this PR with help from Claude but understand and support the core approach, implementation, and test coverage. -- 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]
