[Regression in 54.0.0]. Decimal cast to smaller precision gives invalid (off-by-one) result in some cases #7069
Description
Describe the bug
See practical examples in apache/datafusion#14450.
Somehow a regression caused by #6833.
To Reproduce
Adding this into arrow-cast/src/cast/mod.rs:
#[test]
fn test_decimal_to_decimal() {
let array = vec![Some(520)];
let array = create_decimal_array(array, 4, 2).unwrap();
let input_type = DataType::Decimal128(4, 2);
let output_type = DataType::Decimal128(3, 2);
assert!(can_cast_types(&input_type, &output_type));
let options = CastOptions {
safe: false,
..Default::default()
};
let result = cast_with_options(&array, &output_type, &options);
assert_eq!(result.unwrap().as_primitive::<Decimal128Type>().value(0), 520);
}
passes on 84dba34, but fails on eb7ab83 with
assertion `left == right` failed
left: 521
right: 520
Somehow something in the code adds an extra 1 to the value?
Expected behavior
Test passes
Additional context
Fyi @himadripal @andygrove @findepi @viirya @tustvold as you were involved in that earlier change. I tried to look at it but don't understand how it'd go wrong myself, maybe you know?