waynexia commented on code in PR #2516:
URL: https://github.com/apache/arrow-datafusion/pull/2516#discussion_r879107316
##########
datafusion/physical-expr/src/aggregate/sum.rs:
##########
@@ -262,98 +249,83 @@ fn sum_decimal_with_diff_scale(
}
}
+macro_rules! downcast_arg {
+ ($ARG:expr, $NAME:expr, $ARRAY_TYPE:ident) => {{
+ $ARG.as_any().downcast_ref::<$ARRAY_TYPE>().ok_or_else(|| {
+ DataFusionError::Internal(format!(
+ "could not cast {} to {}",
+ $NAME,
+ type_name::<$ARRAY_TYPE>()
+ ))
+ })?
+ }};
+}
+
+macro_rules! union_arrays {
+ ($LHS: expr, $RHS: expr, $DTYPE: expr, $ARR_DTYPE: ident, $NAME: expr) =>
{{
+ let lhs_casted = &cast(&$LHS.to_array(), $DTYPE)?;
+ let rhs_casted = &cast(&$RHS.to_array(), $DTYPE)?;
+ let lhs_prim_array = downcast_arg!(lhs_casted, $NAME, $ARR_DTYPE);
+ let rhs_prim_array = downcast_arg!(rhs_casted, $NAME, $ARR_DTYPE);
+
+ let chained = lhs_prim_array
+ .iter()
+ .chain(rhs_prim_array.iter())
+ .collect::<$ARR_DTYPE>();
+
+ Arc::new(chained)
+ }};
+}
+
pub(crate) fn sum(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue>
{
- Ok(match (lhs, rhs) {
- (ScalarValue::Decimal128(v1, p1, s1), ScalarValue::Decimal128(v2, p2,
s2)) => {
+ let result = match (lhs.get_datatype(), rhs.get_datatype()) {
+ (DataType::Decimal(p1, s1), DataType::Decimal(p2, s2)) => {
let max_precision = p1.max(p2);
- if s1.eq(s2) {
- // s1 = s2
- sum_decimal(v1, v2, max_precision, s1)
- } else if s1.gt(s2) {
- // s1 > s2
- sum_decimal_with_diff_scale(v1, v2, max_precision, s1, s2)
- } else {
- // s1 < s2
- sum_decimal_with_diff_scale(v2, v1, max_precision, s2, s1)
+
+ match (lhs, rhs) {
+ (
+ ScalarValue::Decimal128(v1, _, _),
+ ScalarValue::Decimal128(v2, _, _),
+ ) => {
+ Ok(if s1.eq(&s2) {
+ // s1 = s2
+ sum_decimal(v1, v2, &max_precision, &s1)
+ } else if s1.gt(&s2) {
+ // s1 > s2
+ sum_decimal_with_diff_scale(v1, v2, &max_precision,
&s1, &s2)
+ } else {
+ // s1 < s2
+ sum_decimal_with_diff_scale(v2, v1, &max_precision,
&s2, &s1)
+ })
+ }
+ _ => Err(DataFusionError::Internal(
+ "Internal state error on sum decimals ".to_string(),
+ )),
}
}
- // float64 coerces everything to f64
- (ScalarValue::Float64(lhs), ScalarValue::Float64(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::Float32(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::Int64(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::Int32(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::Int16(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::Int8(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
+ (DataType::Float64, _) | (_, DataType::Float64) => {
+ let data: ArrayRef =
+ union_arrays!(lhs, rhs, &DataType::Float64, Float64Array,
"f64");
+ sum_batch(&data, &arrow::datatypes::DataType::Float64)
}
- (ScalarValue::Float64(lhs), ScalarValue::UInt64(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::UInt32(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::UInt16(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- (ScalarValue::Float64(lhs), ScalarValue::UInt8(rhs)) => {
- typed_sum!(lhs, rhs, Float64, f64)
- }
- // float32 has no cast
- (ScalarValue::Float32(lhs), ScalarValue::Float32(rhs)) => {
- typed_sum!(lhs, rhs, Float32, f32)
- }
- // u64 coerces u* to u64
- (ScalarValue::UInt64(lhs), ScalarValue::UInt64(rhs)) => {
- typed_sum!(lhs, rhs, UInt64, u64)
+ (DataType::Float32, _) | (_, DataType::Float32) => {
+ let data: ArrayRef =
+ union_arrays!(lhs, rhs, &DataType::Float32, Float32Array,
"f32");
Review Comment:
What about just cut all this coercions logic? I've investigate all the
occurrences of `sum()`, it's only used to accumulate aggregator state in `sum`,
`sum_distinct` and `average` where the operand's type of `sum()` is
deterministic. And `sum()` is an internal function (`pub(crate)`), API change
of this function is acceptable.
I try to remove all the match arms that with different operand types and
only fail two cases (`sum_distinct_i32_with_nulls` and
`sum_distinct_u32_with_nulls`). I think this is acceptable. And I find the
min/max calculator already applied this.
About how to achieve calculate operator over different types, I think we can
extract our coercion rule to something like
```rust
fn coercion(lhs: DataType, rhs: DataType) -> DataType {}
```
And cast both operands to the result type before calculation.
--
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]
