hsiang-c commented on code in PR #18205: URL: https://github.com/apache/datafusion/pull/18205#discussion_r2462587584
########## datafusion/spark/src/function/math/abs.rs: ########## @@ -0,0 +1,1136 @@ +// 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. + +use crate::function::error_utils::{ + invalid_arg_count_exec_err, unsupported_data_type_exec_err, +}; +use arrow::array::*; +use arrow::datatypes::DataType; +use arrow::datatypes::*; +use datafusion_common::DataFusionError::ArrowError; +use datafusion_common::{internal_err, DataFusionError, Result, ScalarValue}; +use datafusion_expr::{ + ColumnarValue, ScalarFunctionArgs, ScalarUDFImpl, Signature, Volatility, +}; +use std::any::Any; +use std::sync::Arc; + +/// Spark-compatible `abs` expression +/// <https://spark.apache.org/docs/latest/api/sql/index.html#abs> +/// +/// Returns the absolute value of input +/// Returns NULL if input is NULL, returns NaN if input is NaN. +/// +/// Differences with DataFusion abs: +/// - Spark's ANSI-compliant dialect, when off (i.e. `spark.sql.ansi.enabled=false`), taking absolute values on minimal values of signed integers returns the value as is. DataFusion's abs throws "DataFusion error: Arrow error: Compute error" on arithmetic overflow +/// - Spark's abs also supports ANSI interval types: YearMonthIntervalType and DayTimeIntervalType. DataFusion's abs doesn't. +/// +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkAbs { + signature: Signature, +} + +impl Default for SparkAbs { + fn default() -> Self { + Self::new() + } +} + +impl SparkAbs { + pub fn new() -> Self { + Self { + signature: Signature::user_defined(Volatility::Immutable), + } + } +} + +impl ScalarUDFImpl for SparkAbs { + fn as_any(&self) -> &dyn Any { + self + } + + fn name(&self) -> &str { + "abs" + } + + fn signature(&self) -> &Signature { + &self.signature + } + + fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> { + Ok(arg_types[0].clone()) + } + + fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> { + spark_abs(&args.args) + } + + fn coerce_types(&self, arg_types: &[DataType]) -> Result<Vec<DataType>> { + if arg_types.len() > 2 { + return Err(invalid_arg_count_exec_err("abs", (1, 2), arg_types.len())); + } + match &arg_types[0] { + DataType::Null + | DataType::UInt8 + | DataType::UInt16 + | DataType::UInt32 + | DataType::UInt64 + | DataType::Int8 + | DataType::Int16 + | DataType::Int32 + | DataType::Int64 + | DataType::Float32 + | DataType::Float64 + | DataType::Decimal128(_, _) + | DataType::Decimal256(_, _) + | DataType::Interval(IntervalUnit::YearMonth) + | DataType::Interval(IntervalUnit::DayTime) => Ok(vec![arg_types[0].clone()]), + other => { + if other.is_numeric() { + Ok(vec![DataType::Float64]) + } else { + Err(unsupported_data_type_exec_err( + "abs", + "Numeric Type or Interval Type", + &arg_types[0], + )) + } + } + } + } +} + +macro_rules! legacy_compute_op { + ($ARRAY:expr, $FUNC:ident, $TYPE:ident, $RESULT:ident) => {{ + let n = $ARRAY.as_any().downcast_ref::<$TYPE>(); + match n { + Some(array) => { + let res: $RESULT = + arrow::compute::kernels::arity::unary(array, |x| x.$FUNC()); + Ok(res) + } + _ => Err(DataFusionError::Internal(format!( + "Invalid data type for abs" + ))), + } + }}; +} + +macro_rules! ansi_compute_op { + ($ARRAY:expr, $FUNC:ident, $TYPE:ident, $RESULT:ident, $NATIVE:ident, $FROM_TYPE:expr) => {{ + let n = $ARRAY.as_any().downcast_ref::<$TYPE>(); + match n { + Some(array) => { + match arrow::compute::kernels::arity::try_unary(array, |x| { + if x == $NATIVE::MIN { + Err(arrow::error::ArrowError::ArithmeticOverflow( + $FROM_TYPE.to_string(), + )) + } else { + Ok(x.$FUNC()) + } + }) { + Ok(res) => Ok(ColumnarValue::Array( + Arc::<PrimitiveArray<$RESULT>>::new(res), + )), + Err(_) => Err(arithmetic_overflow_error($FROM_TYPE)), + } + } + _ => Err(DataFusionError::Internal("Invalid data type".to_string())), + } + }}; +} + +fn arithmetic_overflow_error(from_type: &str) -> DataFusionError { + ArrowError( + Box::from(arrow::error::ArrowError::ComputeError(format!( + "arithmetic overflow from {from_type}", + ))), + None, + ) +} + +pub fn spark_abs(args: &[ColumnarValue]) -> Result<ColumnarValue, DataFusionError> { + if args.len() > 2 { + return internal_err!("abs takes at most 2 arguments, but got: {}", args.len()); + } + + let fail_on_error = if args.len() == 2 { + match &args[1] { + ColumnarValue::Scalar(ScalarValue::Boolean(Some(fail_on_error))) => { + *fail_on_error + } + _ => { + return internal_err!( + "The second argument must be boolean scalar, but got: {:?}", + args[1] + ); + } + } + } else { + false + }; + + match &args[0] { + ColumnarValue::Array(array) => match array.data_type() { + DataType::Null + | DataType::UInt8 + | DataType::UInt16 + | DataType::UInt32 + | DataType::UInt64 => Ok(args[0].clone()), + DataType::Int8 => { + if !fail_on_error { + let result = + legacy_compute_op!(array, wrapping_abs, Int8Array, Int8Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } else { + ansi_compute_op!(array, abs, Int8Array, Int8Type, i8, "Int8") + } + } + DataType::Int16 => { + if !fail_on_error { + let result = + legacy_compute_op!(array, wrapping_abs, Int16Array, Int16Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } else { + ansi_compute_op!(array, abs, Int16Array, Int16Type, i16, "Int16") + } + } + DataType::Int32 => { + if !fail_on_error { + let result = + legacy_compute_op!(array, wrapping_abs, Int32Array, Int32Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } else { + ansi_compute_op!(array, abs, Int32Array, Int32Type, i32, "Int32") + } + } + DataType::Int64 => { + if !fail_on_error { + let result = + legacy_compute_op!(array, wrapping_abs, Int64Array, Int64Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } else { + ansi_compute_op!(array, abs, Int64Array, Int64Type, i64, "Int64") + } + } + DataType::Float32 => { + let result = legacy_compute_op!(array, abs, Float32Array, Float32Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } + DataType::Float64 => { + let result = legacy_compute_op!(array, abs, Float64Array, Float64Array); + Ok(ColumnarValue::Array(Arc::new(result?))) + } + DataType::Decimal128(precision, scale) => { + if !fail_on_error { + let result = legacy_compute_op!( + array, + wrapping_abs, + Decimal128Array, + Decimal128Array + )?; + let result = + result.with_data_type(DataType::Decimal128(*precision, *scale)); + Ok(ColumnarValue::Array(Arc::new(result))) + } else { + // Need to pass precision and scale from input, so not using ansi_compute_op + let input = array.as_any().downcast_ref::<Decimal128Array>(); + match input { + Some(i) => { + match arrow::compute::kernels::arity::try_unary(i, |x| { + if x == i128::MIN { + Err(arrow::error::ArrowError::ArithmeticOverflow( + "Decimal128".to_string(), + )) + } else { + Ok(x.abs()) + } + }) { + Ok(res) => Ok(ColumnarValue::Array(Arc::< + PrimitiveArray<Decimal128Type>, + >::new( + res.with_data_type(DataType::Decimal128( + *precision, *scale, + )), + ))), + Err(_) => Err(arithmetic_overflow_error("Decimal128")), + } + } + _ => Err(DataFusionError::Internal( + "Invalid data type".to_string(), + )), + } + } + } + DataType::Decimal256(precision, scale) => { + if !fail_on_error { + let result = legacy_compute_op!( + array, + wrapping_abs, + Decimal256Array, + Decimal256Array + )?; + let result = + result.with_data_type(DataType::Decimal256(*precision, *scale)); + Ok(ColumnarValue::Array(Arc::new(result))) + } else { + // Need to pass precision and scale from input, so not using ansi_compute_op + let input = array.as_any().downcast_ref::<Decimal256Array>(); + match input { + Some(i) => { + match arrow::compute::kernels::arity::try_unary(i, |x| { + if x == i256::MIN { + Err(arrow::error::ArrowError::ArithmeticOverflow( + "Decimal256".to_string(), + )) + } else { + Ok(x.wrapping_abs()) // i256 doesn't define abs() method + } + }) { + Ok(res) => Ok(ColumnarValue::Array(Arc::< + PrimitiveArray<Decimal256Type>, + >::new( + res.with_data_type(DataType::Decimal256( + *precision, *scale, + )), + ))), + Err(_) => Err(arithmetic_overflow_error("Decimal256")), + } + } + _ => Err(DataFusionError::Internal( + "Invalid data type".to_string(), + )), + } + } + } + DataType::Interval(unit) => match unit { + IntervalUnit::YearMonth => { + let result = legacy_compute_op!( + array, + wrapping_abs, + IntervalYearMonthArray, + IntervalYearMonthArray + )?; + let result = result.with_data_type(DataType::Interval(*unit)); + Ok(ColumnarValue::Array(Arc::new(result))) + } + IntervalUnit::DayTime => { + let result = legacy_compute_op!( + array, + wrapping_abs, + IntervalDayTimeArray, + IntervalDayTimeArray + )?; + let result = result.with_data_type(DataType::Interval(*unit)); + Ok(ColumnarValue::Array(Arc::new(result))) + } + IntervalUnit::MonthDayNano => internal_err!( + "MonthDayNano is not a supported Interval unit for Spark ABS" + ), + }, + dt => internal_err!("Not supported datatype for Spark ABS: {dt}"), + }, + ColumnarValue::Scalar(sv) => { + match sv { + ScalarValue::Null + | ScalarValue::UInt8(_) + | ScalarValue::UInt16(_) + | ScalarValue::UInt32(_) + | ScalarValue::UInt64(_) => Ok(args[0].clone()), + ScalarValue::Int8(a) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => { + Ok(ColumnarValue::Scalar(ScalarValue::Int8(Some(abs_val)))) + } + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Int8(Some(v)))) + } else { + Err(arithmetic_overflow_error("Int8")) + } + } + }) + .unwrap(), + ScalarValue::Int16(a) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => { + Ok(ColumnarValue::Scalar(ScalarValue::Int16(Some(abs_val)))) + } + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Int16(Some(v)))) + } else { + Err(arithmetic_overflow_error("Int16")) + } + } + }) + .unwrap(), + ScalarValue::Int32(a) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => { + Ok(ColumnarValue::Scalar(ScalarValue::Int32(Some(abs_val)))) + } + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Int32(Some(v)))) + } else { + Err(arithmetic_overflow_error("Int32")) + } + } + }) + .unwrap(), + ScalarValue::Int64(a) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => { + Ok(ColumnarValue::Scalar(ScalarValue::Int64(Some(abs_val)))) + } + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Int64(Some(v)))) + } else { + Err(arithmetic_overflow_error("Int64")) + } + } + }) + .unwrap(), + ScalarValue::Float32(a) => Ok(ColumnarValue::Scalar( + ScalarValue::Float32(a.map(|x| x.abs())), + )), + ScalarValue::Float64(a) => Ok(ColumnarValue::Scalar( + ScalarValue::Float64(a.map(|x| x.abs())), + )), + ScalarValue::Decimal128(a, precision, scale) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => Ok(ColumnarValue::Scalar( + ScalarValue::Decimal128(Some(abs_val), *precision, *scale), + )), + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Decimal128( + Some(v), + *precision, + *scale, + ))) + } else { + Err(arithmetic_overflow_error("Decimal128")) + } + } + }) + .unwrap(), + ScalarValue::Decimal256(a, precision, scale) => a + .map(|v| match v.checked_abs() { + Some(abs_val) => Ok(ColumnarValue::Scalar( + ScalarValue::Decimal256(Some(abs_val), *precision, *scale), + )), + None => { + if !fail_on_error { + // return the original value + Ok(ColumnarValue::Scalar(ScalarValue::Decimal256( + Some(v), + *precision, + *scale, + ))) + } else { + Err(arithmetic_overflow_error("Decimal256")) + } + } + }) + .unwrap(), + ScalarValue::IntervalYearMonth(a) => { + let result = a.map(|v| v.wrapping_abs()); + Ok(ColumnarValue::Scalar(ScalarValue::IntervalYearMonth( + result, + ))) + } + ScalarValue::IntervalDayTime(a) => { + let result = a.map(|v| v.wrapping_abs()); + Ok(ColumnarValue::Scalar(ScalarValue::IntervalDayTime(result))) + } + + dt => internal_err!("Not supported datatype for Spark ABS: {dt}"), + } + } + } +} + +#[cfg(test)] +mod tests { + use super::*; + use datafusion_common::cast::{ + as_decimal128_array, as_decimal256_array, as_float32_array, as_float64_array, + as_int16_array, as_int32_array, as_int64_array, as_int8_array, + as_interval_dt_array, as_interval_ym_array, as_uint64_array, + }; + + fn with_fail_on_error<F: Fn(bool) -> Result<()>>(test_fn: F) { + for fail_on_error in [true, false] { + let _ = test_fn(fail_on_error); + } + } + + // Unsigned types, return as is + #[test] + fn test_abs_u8_scalar() { + with_fail_on_error(|fail_on_error| { + let args = ColumnarValue::Scalar(ScalarValue::UInt8(Some(u8::MAX))); + let fail_on_error_arg = + ColumnarValue::Scalar(ScalarValue::Boolean(Some(fail_on_error))); + match spark_abs(&[args, fail_on_error_arg]) { + Ok(ColumnarValue::Scalar(ScalarValue::UInt8(Some(result)))) => { + assert_eq!(result, u8::MAX); + Ok(()) + } + Err(e) => { + if fail_on_error { + assert!( + e.to_string().contains("ARITHMETIC_OVERFLOW"), + "Error message did not match. Actual message: {e}" + ); + Ok(()) + } else { + panic!("Didn't expect error, but got: {e:?}") + } + } + _ => unreachable!(), + } + }); + } Review Comment: @Jefffrey You're right, thanks for the feedback. -- 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]
