Various ScalarValue numeric method fixes & refactors (especially decimal)

datafusion/common/src/scalar/consts.rs

@@ -17,6 +17,9 @@
 
 // Constants defined for scalar construction.
 
+use arrow::datatypes::{Decimal32Type, Decimal64Type, Decimal128Type, DecimalType};
+use arrow::datatypes::{Decimal256Type, i256};
+
 // Next F16 value above π (upper bound)
 pub(super) const PI_UPPER_F16: half::f16 = half::f16::from_bits(0x4249);
 
@@ -54,3 +57,63 @@ pub(super) const NEGATIVE_FRAC_PI_2_LOWER_F32: f32 =
 // Next f64 value below -π/2 (lower bound)
 pub(super) const NEGATIVE_FRAC_PI_2_LOWER_F64: f64 =
     (-std::f64::consts::FRAC_PI_2).next_down();
+
+// Generate lookup table for 1 values of decimals (1, 10, 100, etc.)
+macro_rules! decimal_ones_lut {
+    () => {{
+        let mut values = [1; _];
+        let mut i = 1;
+        while i < values.len() {
+            values[i] = values[i - 1] * 10;
+            i += 1;
+        }
+        values
+    }};
+}
+
+// 1, 10, 100 values meant to be indexed by scale. We omit handling for MAX_SCALE
+// itself (we don't go to MAX_SCALE + 1) since we can't represent a 1 value at
+// that scale.
+pub(super) const DECIMAL32_ONES: [i32; Decimal32Type::MAX_SCALE as usize] =
+    decimal_ones_lut!();
+pub(super) const DECIMAL64_ONES: [i64; Decimal64Type::MAX_SCALE as usize] =
+    decimal_ones_lut!();
+pub(super) const DECIMAL128_ONES: [i128; Decimal128Type::MAX_SCALE as usize] =
+    decimal_ones_lut!();
+pub(super) const DECIMAL256_ONES: [i256; Decimal256Type::MAX_SCALE as usize] = {
+    // This code was generated by codex and frankly I don't know how it works,
+    // but the test below verifies it outputs the correct values so ¯\_(ツ)_/¯
+    //
+    // This is mainly a shortcut for not needing to manually list out each value
+    // anyway.
+    //
+    // TODO: simplify this after https://github.com/apache/arrow-rs/pull/10363
+    //       lands upstream
+    let mut values = [i256::ONE; _];
+    let mut i = 1;
+    while i < values.len() {
+        let (low, high) = values[i - 1].to_parts();
+        let low_product = (low as u64 as u128) * 10;
+        let high_product = (low >> 64) * 10 + (low_product >> 64);
+        let low = ((high_product as u64 as u128) << 64) | low_product as u64 as u128;
+        let carry = (high_product >> 64) as i128;
+        values[i] = i256::from_parts(low, high * 10 + carry);
+        i += 1;
+    }
+    values
+};
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn test_ensure_correct_decimal256_ones() {
+        for (scale, val) in DECIMAL256_ONES.iter().enumerate() {
+            let zeros = "0".repeat(scale);
+            let num = "1".to_string() + &zeros;
+            let num = i256::from_string(&num).unwrap();
+            assert_eq!(num, *val, "{scale}");
+        }
+    }
+}