datafusion_physical_expr/
scalar_function.rs

// 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.

//! Declaration of built-in (scalar) functions.
//! This module contains built-in functions' enumeration and metadata.
//!
//! Generally, a function has:
//! * a signature
//! * a return type, that is a function of the incoming argument's types
//! * the computation, that must accept each valid signature
//!
//! * Signature: see `Signature`
//! * Return type: a function `(arg_types) -> return_type`. E.g. for sqrt, ([f32]) -> f32, ([f64]) -> f64.
//!
//! This module also has a set of coercion rules to improve user experience: if an argument i32 is passed
//! to a function that supports f64, it is coerced to f64.

use std::any::Any;
use std::fmt::{self, Debug, Formatter};
use std::hash::{Hash, Hasher};
use std::sync::Arc;

use crate::physical_expr::{down_cast_any_ref, physical_exprs_equal};
use crate::PhysicalExpr;

use arrow::datatypes::{DataType, Schema};
use arrow::record_batch::RecordBatch;
use arrow_array::Array;
use datafusion_common::{internal_err, DFSchema, Result, ScalarValue};
use datafusion_expr::interval_arithmetic::Interval;
use datafusion_expr::sort_properties::ExprProperties;
use datafusion_expr::type_coercion::functions::data_types_with_scalar_udf;
use datafusion_expr::{expr_vec_fmt, ColumnarValue, Expr, ScalarUDF};

/// Physical expression of a scalar function
pub struct ScalarFunctionExpr {
    fun: Arc<ScalarUDF>,
    name: String,
    args: Vec<Arc<dyn PhysicalExpr>>,
    return_type: DataType,
    nullable: bool,
}

impl Debug for ScalarFunctionExpr {
    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
        f.debug_struct("ScalarFunctionExpr")
            .field("fun", &"<FUNC>")
            .field("name", &self.name)
            .field("args", &self.args)
            .field("return_type", &self.return_type)
            .finish()
    }
}

impl ScalarFunctionExpr {
    /// Create a new Scalar function
    pub fn new(
        name: &str,
        fun: Arc<ScalarUDF>,
        args: Vec<Arc<dyn PhysicalExpr>>,
        return_type: DataType,
    ) -> Self {
        Self {
            fun,
            name: name.to_owned(),
            args,
            return_type,
            nullable: true,
        }
    }

    /// Get the scalar function implementation
    pub fn fun(&self) -> &ScalarUDF {
        &self.fun
    }

    /// The name for this expression
    pub fn name(&self) -> &str {
        &self.name
    }

    /// Input arguments
    pub fn args(&self) -> &[Arc<dyn PhysicalExpr>] {
        &self.args
    }

    /// Data type produced by this expression
    pub fn return_type(&self) -> &DataType {
        &self.return_type
    }

    pub fn with_nullable(mut self, nullable: bool) -> Self {
        self.nullable = nullable;
        self
    }

    pub fn nullable(&self) -> bool {
        self.nullable
    }
}

impl fmt::Display for ScalarFunctionExpr {
    fn fmt(&self, f: &mut Formatter) -> fmt::Result {
        write!(f, "{}({})", self.name, expr_vec_fmt!(self.args))
    }
}

impl PhysicalExpr for ScalarFunctionExpr {
    /// Return a reference to Any that can be used for downcasting
    fn as_any(&self) -> &dyn Any {
        self
    }

    fn data_type(&self, _input_schema: &Schema) -> Result<DataType> {
        Ok(self.return_type.clone())
    }

    fn nullable(&self, _input_schema: &Schema) -> Result<bool> {
        Ok(self.nullable)
    }

    fn evaluate(&self, batch: &RecordBatch) -> Result<ColumnarValue> {
        let inputs = self
            .args
            .iter()
            .map(|e| e.evaluate(batch))
            .collect::<Result<Vec<_>>>()?;

        // evaluate the function
        let output = self.fun.invoke_batch(&inputs, batch.num_rows())?;

        if let ColumnarValue::Array(array) = &output {
            if array.len() != batch.num_rows() {
                // If the arguments are a non-empty slice of scalar values, we can assume that
                // returning a one-element array is equivalent to returning a scalar.
                let preserve_scalar = array.len() == 1
                    && !inputs.is_empty()
                    && inputs
                        .iter()
                        .all(|arg| matches!(arg, ColumnarValue::Scalar(_)));
                return if preserve_scalar {
                    ScalarValue::try_from_array(array, 0).map(ColumnarValue::Scalar)
                } else {
                    internal_err!("UDF returned a different number of rows than expected. Expected: {}, Got: {}",
                            batch.num_rows(), array.len())
                };
            }
        }
        Ok(output)
    }

    fn children(&self) -> Vec<&Arc<dyn PhysicalExpr>> {
        self.args.iter().collect()
    }

    fn with_new_children(
        self: Arc<Self>,
        children: Vec<Arc<dyn PhysicalExpr>>,
    ) -> Result<Arc<dyn PhysicalExpr>> {
        Ok(Arc::new(
            ScalarFunctionExpr::new(
                &self.name,
                Arc::clone(&self.fun),
                children,
                self.return_type().clone(),
            )
            .with_nullable(self.nullable),
        ))
    }

    fn evaluate_bounds(&self, children: &[&Interval]) -> Result<Interval> {
        self.fun.evaluate_bounds(children)
    }

    fn propagate_constraints(
        &self,
        interval: &Interval,
        children: &[&Interval],
    ) -> Result<Option<Vec<Interval>>> {
        self.fun.propagate_constraints(interval, children)
    }

    fn dyn_hash(&self, state: &mut dyn Hasher) {
        let mut s = state;
        self.name.hash(&mut s);
        self.args.hash(&mut s);
        self.return_type.hash(&mut s);
        // Add `self.fun` when hash is available
    }

    fn get_properties(&self, children: &[ExprProperties]) -> Result<ExprProperties> {
        let sort_properties = self.fun.output_ordering(children)?;
        let children_range = children
            .iter()
            .map(|props| &props.range)
            .collect::<Vec<_>>();
        let range = self.fun().evaluate_bounds(&children_range)?;

        Ok(ExprProperties {
            sort_properties,
            range,
        })
    }
}

impl PartialEq<dyn Any> for ScalarFunctionExpr {
    /// Comparing name, args and return_type
    fn eq(&self, other: &dyn Any) -> bool {
        down_cast_any_ref(other)
            .downcast_ref::<Self>()
            .map(|x| {
                self.name == x.name
                    && physical_exprs_equal(&self.args, &x.args)
                    && self.return_type == x.return_type
            })
            .unwrap_or(false)
    }
}

/// Create a physical expression for the UDF.
pub fn create_physical_expr(
    fun: &ScalarUDF,
    input_phy_exprs: &[Arc<dyn PhysicalExpr>],
    input_schema: &Schema,
    args: &[Expr],
    input_dfschema: &DFSchema,
) -> Result<Arc<dyn PhysicalExpr>> {
    let input_expr_types = input_phy_exprs
        .iter()
        .map(|e| e.data_type(input_schema))
        .collect::<Result<Vec<_>>>()?;

    // verify that input data types is consistent with function's `TypeSignature`
    data_types_with_scalar_udf(&input_expr_types, fun)?;

    // Since we have arg_types, we dont need args and schema.
    let return_type =
        fun.return_type_from_exprs(args, input_dfschema, &input_expr_types)?;

    Ok(Arc::new(
        ScalarFunctionExpr::new(
            fun.name(),
            Arc::new(fun.clone()),
            input_phy_exprs.to_vec(),
            return_type,
        )
        .with_nullable(fun.is_nullable(args, input_dfschema)),
    ))
}
datafusion_physical_expr/scalar_function.rs

datafusion_physical_expr/
scalar_function.rs
