kosiew commented on code in PR #20594: URL: https://github.com/apache/datafusion/pull/20594#discussion_r2923398430
########## datafusion/sqllogictest/test_files/spark/math/floor.slt: ########## @@ -21,22 +21,138 @@ # For more information, please see: # https://github.com/apache/datafusion/issues/15914 +# Tests for Spark-compatible floor function. +# Spark semantics differ from DataFusion's built-in floor in two ways: +# 1. Return type: Spark returns Int64 for float/integer inputs; +# DataFusion returns the same float type (e.g. floor(1.5::DOUBLE) -> DOUBLE in DF, BIGINT in Spark) +# 2. Decimal precision: Spark adjusts precision to (p - s + 1) for Decimal128(p, s) with scale > 0; +# DataFusion preserves the original precision and scale +# +# Example: SELECT floor(1.50::DECIMAL(10,2)) +# Spark: returns Decimal(9, 0) value 1 +# DataFusion: returns Decimal(10, 2) value 1.00 + ## Original Query: SELECT floor(-0.1); ## PySpark 3.5.5 Result: {'FLOOR(-0.1)': Decimal('-1'), 'typeof(FLOOR(-0.1))': 'decimal(1,0)', 'typeof(-0.1)': 'decimal(1,1)'} -#query -#SELECT floor(-0.1::decimal(1,1)); +query R +SELECT floor(-0.1::decimal(1,1)); +---- +-1 ## Original Query: SELECT floor(3.1411, -3); ## PySpark 3.5.5 Result: {'floor(3.1411, -3)': Decimal('0'), 'typeof(floor(3.1411, -3))': 'decimal(4,0)', 'typeof(3.1411)': 'decimal(5,4)', 'typeof(-3)': 'int'} +## TODO: 2-argument floor(value, scale) is not yet implemented #query #SELECT floor(3.1411::decimal(5,4), -3::int); Review Comment: Also add explicit type assertions to the SLT coverage (for example with `typeof(...)`) since the main semantic difference from DataFusion is the result type, not just the numeric value. ########## datafusion/spark/src/function/math/floor.rs: ########## @@ -0,0 +1,263 @@ +// 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 std::any::Any; +use std::sync::Arc; + +use arrow::array::{AsArray, Decimal128Array}; +use arrow::compute::cast; +use arrow::datatypes::{DataType, Decimal128Type, Float32Type, Float64Type, Int64Type}; +use datafusion_common::utils::take_function_args; +use datafusion_common::{Result, exec_err}; +use datafusion_expr::{ + ColumnarValue, ScalarFunctionArgs, ScalarUDFImpl, Signature, Volatility, +}; + +/// Spark-compatible `floor` expression +/// <https://spark.apache.org/docs/latest/api/sql/index.html#floor> +/// +/// Differences with DataFusion floor: +/// - Spark's floor returns Int64 for float and integer inputs; DataFusion preserves +/// the input type (Float32→Float32, Float64→Float64, integers coerced to Float64) +/// - Spark's floor on Decimal128(p, s) returns Decimal128(p−s+1, 0), reducing scale +/// to 0; DataFusion preserves the original precision and scale +/// - Spark only supports Decimal128; DataFusion also supports Decimal32/64/256 +/// - Spark does not check for decimal overflow; DataFusion errors on overflow +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkFloor { + signature: Signature, +} + +impl Default for SparkFloor { + fn default() -> Self { + Self::new() + } +} + +impl SparkFloor { + pub fn new() -> Self { + Self { + signature: Signature::numeric(1, Volatility::Immutable), + } + } +} + +impl ScalarUDFImpl for SparkFloor { + fn as_any(&self) -> &dyn Any { + self + } + + fn name(&self) -> &str { + "floor" + } + + fn signature(&self) -> &Signature { + &self.signature + } + + fn return_type(&self, arg_types: &[DataType]) -> Result<DataType> { + match &arg_types[0] { + DataType::Decimal128(p, s) if *s > 0 => { + let new_p = ((*p as i64) - (*s as i64) + 1).clamp(1, 38) as u8; + Ok(DataType::Decimal128(new_p, 0)) + } + DataType::Decimal128(p, s) => Ok(DataType::Decimal128(*p, *s)), + _ => Ok(DataType::Int64), + } + } + + fn invoke_with_args(&self, args: ScalarFunctionArgs) -> Result<ColumnarValue> { + let return_type = args.return_type().clone(); + spark_floor(&args.args, &return_type) + } +} + +fn spark_floor(args: &[ColumnarValue], return_type: &DataType) -> Result<ColumnarValue> { + let input = match take_function_args("floor", args)? { + [ColumnarValue::Scalar(value)] => value.to_array()?, Review Comment: Looks like every scalar invocation will incur this round trip: scalar→array→result‑array→scalar pattern Consider following the built-in `floor`/`ceil` pattern for scalar inputs so scalar calls can stay scalar and avoid the extra array round-trip. ########## datafusion/spark/src/function/math/floor.rs: ########## @@ -0,0 +1,263 @@ +// 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 std::any::Any; +use std::sync::Arc; + +use arrow::array::{AsArray, Decimal128Array}; +use arrow::compute::cast; +use arrow::datatypes::{DataType, Decimal128Type, Float32Type, Float64Type, Int64Type}; +use datafusion_common::utils::take_function_args; +use datafusion_common::{Result, exec_err}; +use datafusion_expr::{ + ColumnarValue, ScalarFunctionArgs, ScalarUDFImpl, Signature, Volatility, +}; + +/// Spark-compatible `floor` expression +/// <https://spark.apache.org/docs/latest/api/sql/index.html#floor> +/// +/// Differences with DataFusion floor: +/// - Spark's floor returns Int64 for float and integer inputs; DataFusion preserves +/// the input type (Float32→Float32, Float64→Float64, integers coerced to Float64) +/// - Spark's floor on Decimal128(p, s) returns Decimal128(p−s+1, 0), reducing scale +/// to 0; DataFusion preserves the original precision and scale +/// - Spark only supports Decimal128; DataFusion also supports Decimal32/64/256 +/// - Spark does not check for decimal overflow; DataFusion errors on overflow +#[derive(Debug, PartialEq, Eq, Hash)] +pub struct SparkFloor { + signature: Signature, +} + +impl Default for SparkFloor { + fn default() -> Self { + Self::new() + } +} + +impl SparkFloor { + pub fn new() -> Self { + Self { + signature: Signature::numeric(1, Volatility::Immutable), Review Comment: ~How does this implement Spark's floor(expr, scale)?~ > ## TODO: 2-argument floor(value, scale) is not yet implemented Signature::numeric(1) is a wider surface than spark_floor's supported data_types. -- 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]
