Copilot commented on code in PR #22605: URL: https://github.com/apache/datafusion/pull/22605#discussion_r3320529943
########## datafusion/spark/src/function/math/pow.rs: ########## @@ -0,0 +1,195 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! Spark-compatible `pow` / `power` function. +//! +//! Unlike the default DataFusion (PostgreSQL) implementation, Spark returns +//! `Infinity` for `pow(0, <negative>)` rather than raising an error. + +use std::sync::Arc; + +use arrow::array::{ArrayRef, Float64Array}; +use arrow::datatypes::DataType; + +use datafusion_common::utils::take_function_args; +use datafusion_common::{Result, ScalarValue}; +use datafusion_expr::{ + ColumnarValue, Documentation, ScalarFunctionArgs, ScalarUDFImpl, Signature, +}; +use datafusion_functions::math::power::PowerFunc; + +/// Spark-compatible implementation of `pow` / `power`. +/// +/// Behavioural difference from the DataFusion default: +/// - `pow(0, <negative>)` → `Infinity` (IEEE 754 / Spark semantics) +/// The default raises `"zero raised to a negative power is undefined"` to +/// match PostgreSQL. +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkPow { + inner: PowerFunc, + aliases: Vec<String>, +} + +impl Default for SparkPow { + fn default() -> Self { + Self::new() + } +} + +impl SparkPow { + pub fn new() -> Self { + Self { + inner: PowerFunc::new(), + // SparkPow is named "pow"; expose "power" as an alias so that + // both names resolve to Spark semantics when this crate is active. + aliases: vec!["power".to_string()], + } + } +} + +impl ScalarUDFImpl for SparkPow { + fn name(&self) -> &str { + "pow" + } + + fn aliases(&self) -> &[String] { + &self.aliases + } + + fn signature(&self) -> &Signature { + self.inner.signature() + } + + fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> { + self.inner.return_type(arg_types) + } + + fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> { + // Only Float64 needs the Spark override. + // Decimal / integer paths are delegated to the standard PowerFunc which + // already handles them correctly (decimal can't represent Infinity anyway). + if !matches!(args.args[0].data_type(), DataType::Float64) { + return self.inner.invoke_with_args(args); Review Comment: The Float64 override gate only checks the first argument’s type, but the implementation later assumes *both* base and exponent are Float64 (downcasts with `expect`). If base is Float64 but exponent is a different numeric type (possible with coercions/mixed inputs), this can panic. Consider checking both arg types up-front (base and exponent) and delegating to `inner` unless both are Float64, or explicitly casting the exponent to Float64 before downcasting. ########## datafusion/sqllogictest/test_files/spark/math/pow.slt: ########## @@ -22,6 +22,27 @@ # https://github.com/apache/datafusion/issues/15914 ## Original Query: SELECT pow(2, 3); -## PySpark 3.5.5 Result: {'pow(2, 3)': 8.0, 'typeof(pow(2, 3))': 'double', 'typeof(2)': 'int', 'typeof(3)': 'int'} -#query -#SELECT pow(2::int, 3::int); +## PySpark 3.5.5 Result: {'pow(2, 3)': 8.0, 'typeof(pow(2, 3))': 'double'} +query R +SELECT pow(2::int, 3::int); +---- +8 Review Comment: This SLT expectation appears inconsistent with the documented PySpark result (`8.0` as a double). With `query R`, results are regex-matched per value; `8` typically won’t match `8.0` unless written as a regex that allows the decimal part. If the engine returns a double-formatted value, update the expected output to match (e.g., `8.0`) or use a regex that allows either representation (e.g., `8(\\.0)?`) so the test matches Spark’s double semantics. ########## datafusion/spark/src/function/math/pow.rs: ########## @@ -0,0 +1,195 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! Spark-compatible `pow` / `power` function. +//! +//! Unlike the default DataFusion (PostgreSQL) implementation, Spark returns +//! `Infinity` for `pow(0, <negative>)` rather than raising an error. + +use std::sync::Arc; + +use arrow::array::{ArrayRef, Float64Array}; +use arrow::datatypes::DataType; + +use datafusion_common::utils::take_function_args; +use datafusion_common::{Result, ScalarValue}; +use datafusion_expr::{ + ColumnarValue, Documentation, ScalarFunctionArgs, ScalarUDFImpl, Signature, +}; +use datafusion_functions::math::power::PowerFunc; + +/// Spark-compatible implementation of `pow` / `power`. +/// +/// Behavioural difference from the DataFusion default: +/// - `pow(0, <negative>)` → `Infinity` (IEEE 754 / Spark semantics) +/// The default raises `"zero raised to a negative power is undefined"` to +/// match PostgreSQL. +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkPow { + inner: PowerFunc, + aliases: Vec<String>, +} + +impl Default for SparkPow { + fn default() -> Self { + Self::new() + } +} + +impl SparkPow { + pub fn new() -> Self { + Self { + inner: PowerFunc::new(), + // SparkPow is named "pow"; expose "power" as an alias so that + // both names resolve to Spark semantics when this crate is active. + aliases: vec!["power".to_string()], + } + } +} + +impl ScalarUDFImpl for SparkPow { + fn name(&self) -> &str { + "pow" + } + + fn aliases(&self) -> &[String] { + &self.aliases + } + + fn signature(&self) -> &Signature { + self.inner.signature() + } + + fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> { + self.inner.return_type(arg_types) + } + + fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> { + // Only Float64 needs the Spark override. + // Decimal / integer paths are delegated to the standard PowerFunc which + // already handles them correctly (decimal can't represent Infinity anyway). + if !matches!(args.args[0].data_type(), DataType::Float64) { + return self.inner.invoke_with_args(args); + } + + let num_rows = args.number_rows; + + // ── Scalar × Scalar fast path ──────────────────────────────────────── + // Pattern-match on the slice to avoid any ownership issues. + if let [ + ColumnarValue::Scalar(ScalarValue::Float64(b)), + ColumnarValue::Scalar(ScalarValue::Float64(e)), + ] = args.args.as_slice() + { + // b and e are &Option<f64>; Option<f64> is Copy. + let result = (*b).zip(*e).map(|(b, e)| { + // Spark: 0^negative = +Infinity (covers both 0.0 and -0.0) + if b == 0.0 && e < 0.0 { + f64::INFINITY + } else { + b.powf(e) + } + }); + return Ok(ColumnarValue::Scalar(ScalarValue::Float64(result))); + } + + // ── Array path ─────────────────────────────────────────────────────── + let [base, exponent] = take_function_args(self.name(), &args.args)?; + + let base_arr: ArrayRef = base.to_array(num_rows)?; + let exp_arr: ArrayRef = exponent.to_array(num_rows)?; Review Comment: Unit tests currently cover only the scalar×scalar fast path. The array path (and mixed scalar/array cases) is newly introduced behavior and should be exercised to prevent regressions, especially around `0 ^ negative` and null handling. Add tests that pass `ColumnarValue::Array` inputs (and ideally a mixed scalar/array case) and assert `Infinity` for base=0 and exponent<0. ########## datafusion/spark/src/function/math/pow.rs: ########## @@ -0,0 +1,195 @@ +// Licensed to the Apache Software Foundation (ASF) under one +// or more contributor license agreements. See the NOTICE file +// distributed with this work for additional information +// regarding copyright ownership. The ASF licenses this file +// to you under the Apache License, Version 2.0 (the +// "License"); you may not use this file except in compliance +// with the License. You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, +// software distributed under the License is distributed on an +// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY +// KIND, either express or implied. See the License for the +// specific language governing permissions and limitations +// under the License. + +//! Spark-compatible `pow` / `power` function. +//! +//! Unlike the default DataFusion (PostgreSQL) implementation, Spark returns +//! `Infinity` for `pow(0, <negative>)` rather than raising an error. + +use std::sync::Arc; + +use arrow::array::{ArrayRef, Float64Array}; +use arrow::datatypes::DataType; + +use datafusion_common::utils::take_function_args; +use datafusion_common::{Result, ScalarValue}; +use datafusion_expr::{ + ColumnarValue, Documentation, ScalarFunctionArgs, ScalarUDFImpl, Signature, +}; +use datafusion_functions::math::power::PowerFunc; + +/// Spark-compatible implementation of `pow` / `power`. +/// +/// Behavioural difference from the DataFusion default: +/// - `pow(0, <negative>)` → `Infinity` (IEEE 754 / Spark semantics) +/// The default raises `"zero raised to a negative power is undefined"` to +/// match PostgreSQL. +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkPow { + inner: PowerFunc, + aliases: Vec<String>, +} + +impl Default for SparkPow { + fn default() -> Self { + Self::new() + } +} + +impl SparkPow { + pub fn new() -> Self { + Self { + inner: PowerFunc::new(), + // SparkPow is named "pow"; expose "power" as an alias so that + // both names resolve to Spark semantics when this crate is active. + aliases: vec!["power".to_string()], + } + } +} + +impl ScalarUDFImpl for SparkPow { + fn name(&self) -> &str { + "pow" + } + + fn aliases(&self) -> &[String] { + &self.aliases + } + + fn signature(&self) -> &Signature { + self.inner.signature() + } + + fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> { + self.inner.return_type(arg_types) + } + + fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> { + // Only Float64 needs the Spark override. + // Decimal / integer paths are delegated to the standard PowerFunc which + // already handles them correctly (decimal can't represent Infinity anyway). + if !matches!(args.args[0].data_type(), DataType::Float64) { + return self.inner.invoke_with_args(args); + } + + let num_rows = args.number_rows; + + // ── Scalar × Scalar fast path ──────────────────────────────────────── + // Pattern-match on the slice to avoid any ownership issues. + if let [ + ColumnarValue::Scalar(ScalarValue::Float64(b)), + ColumnarValue::Scalar(ScalarValue::Float64(e)), + ] = args.args.as_slice() + { + // b and e are &Option<f64>; Option<f64> is Copy. + let result = (*b).zip(*e).map(|(b, e)| { + // Spark: 0^negative = +Infinity (covers both 0.0 and -0.0) + if b == 0.0 && e < 0.0 { + f64::INFINITY + } else { + b.powf(e) + } + }); + return Ok(ColumnarValue::Scalar(ScalarValue::Float64(result))); + } + + // ── Array path ─────────────────────────────────────────────────────── + let [base, exponent] = take_function_args(self.name(), &args.args)?; + + let base_arr: ArrayRef = base.to_array(num_rows)?; + let exp_arr: ArrayRef = exponent.to_array(num_rows)?; + + let base_f64 = base_arr + .as_any() + .downcast_ref::<Float64Array>() + .expect("base must be Float64Array"); + let exp_f64 = exp_arr + .as_any() + .downcast_ref::<Float64Array>() + .expect("exponent must be Float64Array"); + + // b.powf(e) follows IEEE 754: 0.0_f64.powf(-1.0) == f64::INFINITY. + // No explicit guard needed — the default Rust behaviour is exactly + // what Spark requires. + let result: Float64Array = base_f64 + .iter() + .zip(exp_f64.iter()) + .map(|(b, e)| match (b, e) { + (Some(b), Some(e)) => { + if b == 0.0 && e < 0.0 { + Some(f64::INFINITY) + } else { + Some(b.powf(e)) + } + } Review Comment: The comment says no explicit guard is needed for `0 ^ negative`, but the code still includes an explicit guard. This is internally inconsistent and could confuse future maintainers. Either remove the guard (since `powf` already yields `Infinity`) or update the comment to explain why the explicit branch is intentionally retained (e.g., clarity/consistency with scalar path). -- 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]
