datafusion 4.0.0

// 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.

//! Defines physical expressions that can evaluated at runtime during query execution

use std::any::Any;
use std::convert::TryFrom;
use std::sync::Arc;

use crate::error::{DataFusionError, Result};
use crate::physical_plan::{Accumulator, AggregateExpr, PhysicalExpr};
use crate::scalar::ScalarValue;
use arrow::compute;
use arrow::datatypes::DataType;
use arrow::{
    array::{
        ArrayRef, Float32Array, Float64Array, Int16Array, Int32Array, Int64Array,
        Int8Array, UInt16Array, UInt32Array, UInt64Array, UInt8Array,
    },
    datatypes::Field,
};

use super::format_state_name;

/// SUM aggregate expression
#[derive(Debug)]
pub struct Sum {
    name: String,
    data_type: DataType,
    expr: Arc<dyn PhysicalExpr>,
    nullable: bool,
}

/// function return type of a sum
pub fn sum_return_type(arg_type: &DataType) -> Result<DataType> {
    match arg_type {
        DataType::Int8 | DataType::Int16 | DataType::Int32 | DataType::Int64 => {
            Ok(DataType::Int64)
        }
        DataType::UInt8 | DataType::UInt16 | DataType::UInt32 | DataType::UInt64 => {
            Ok(DataType::UInt64)
        }
        DataType::Float32 => Ok(DataType::Float32),
        DataType::Float64 => Ok(DataType::Float64),
        other => Err(DataFusionError::Plan(format!(
            "SUM does not support type \"{:?}\"",
            other
        ))),
    }
}

impl Sum {
    /// Create a new SUM aggregate function
    pub fn new(expr: Arc<dyn PhysicalExpr>, name: String, data_type: DataType) -> Self {
        Self {
            name,
            expr,
            data_type,
            nullable: true,
        }
    }
}

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

    fn field(&self) -> Result<Field> {
        Ok(Field::new(
            &self.name,
            self.data_type.clone(),
            self.nullable,
        ))
    }

    fn state_fields(&self) -> Result<Vec<Field>> {
        Ok(vec![Field::new(
            &format_state_name(&self.name, "sum"),
            self.data_type.clone(),
            self.nullable,
        )])
    }

    fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
        vec![self.expr.clone()]
    }

    fn create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
        Ok(Box::new(SumAccumulator::try_new(&self.data_type)?))
    }
}

#[derive(Debug)]
struct SumAccumulator {
    sum: ScalarValue,
}

impl SumAccumulator {
    /// new sum accumulator
    pub fn try_new(data_type: &DataType) -> Result<Self> {
        Ok(Self {
            sum: ScalarValue::try_from(data_type)?,
        })
    }
}

// returns the new value after sum with the new values, taking nullability into account
macro_rules! typed_sum_delta_batch {
    ($VALUES:expr, $ARRAYTYPE:ident, $SCALAR:ident) => {{
        let array = $VALUES.as_any().downcast_ref::<$ARRAYTYPE>().unwrap();
        let delta = compute::sum(array);
        ScalarValue::$SCALAR(delta)
    }};
}

// sums the array and returns a ScalarValue of its corresponding type.
pub(super) fn sum_batch(values: &ArrayRef) -> Result<ScalarValue> {
    Ok(match values.data_type() {
        DataType::Float64 => typed_sum_delta_batch!(values, Float64Array, Float64),
        DataType::Float32 => typed_sum_delta_batch!(values, Float32Array, Float32),
        DataType::Int64 => typed_sum_delta_batch!(values, Int64Array, Int64),
        DataType::Int32 => typed_sum_delta_batch!(values, Int32Array, Int32),
        DataType::Int16 => typed_sum_delta_batch!(values, Int16Array, Int16),
        DataType::Int8 => typed_sum_delta_batch!(values, Int8Array, Int8),
        DataType::UInt64 => typed_sum_delta_batch!(values, UInt64Array, UInt64),
        DataType::UInt32 => typed_sum_delta_batch!(values, UInt32Array, UInt32),
        DataType::UInt16 => typed_sum_delta_batch!(values, UInt16Array, UInt16),
        DataType::UInt8 => typed_sum_delta_batch!(values, UInt8Array, UInt8),
        e => {
            return Err(DataFusionError::Internal(format!(
                "Sum is not expected to receive the type {:?}",
                e
            )))
        }
    })
}

// returns the sum of two scalar values, including coercion into $TYPE.
macro_rules! typed_sum {
    ($OLD_VALUE:expr, $DELTA:expr, $SCALAR:ident, $TYPE:ident) => {{
        ScalarValue::$SCALAR(match ($OLD_VALUE, $DELTA) {
            (None, None) => None,
            (Some(a), None) => Some(a.clone()),
            (None, Some(b)) => Some(b.clone() as $TYPE),
            (Some(a), Some(b)) => Some(a + (*b as $TYPE)),
        })
    }};
}

pub(super) fn sum(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue> {
    Ok(match (lhs, rhs) {
        // 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)
        }
        (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)
        }
        (ScalarValue::UInt64(lhs), ScalarValue::UInt32(rhs)) => {
            typed_sum!(lhs, rhs, UInt64, u64)
        }
        (ScalarValue::UInt64(lhs), ScalarValue::UInt16(rhs)) => {
            typed_sum!(lhs, rhs, UInt64, u64)
        }
        (ScalarValue::UInt64(lhs), ScalarValue::UInt8(rhs)) => {
            typed_sum!(lhs, rhs, UInt64, u64)
        }
        // i64 coerces i* to u64
        (ScalarValue::Int64(lhs), ScalarValue::Int64(rhs)) => {
            typed_sum!(lhs, rhs, Int64, i64)
        }
        (ScalarValue::Int64(lhs), ScalarValue::Int32(rhs)) => {
            typed_sum!(lhs, rhs, Int64, i64)
        }
        (ScalarValue::Int64(lhs), ScalarValue::Int16(rhs)) => {
            typed_sum!(lhs, rhs, Int64, i64)
        }
        (ScalarValue::Int64(lhs), ScalarValue::Int8(rhs)) => {
            typed_sum!(lhs, rhs, Int64, i64)
        }
        e => {
            return Err(DataFusionError::Internal(format!(
                "Sum is not expected to receive a scalar {:?}",
                e
            )))
        }
    })
}

impl Accumulator for SumAccumulator {
    fn update_batch(&mut self, values: &[ArrayRef]) -> Result<()> {
        let values = &values[0];
        self.sum = sum(&self.sum, &sum_batch(values)?)?;
        Ok(())
    }

    fn update(&mut self, values: &[ScalarValue]) -> Result<()> {
        // sum(v1, v2, v3) = v1 + v2 + v3
        self.sum = sum(&self.sum, &values[0])?;
        Ok(())
    }

    fn merge(&mut self, states: &[ScalarValue]) -> Result<()> {
        // sum(sum1, sum2) = sum1 + sum2
        self.update(states)
    }

    fn merge_batch(&mut self, states: &[ArrayRef]) -> Result<()> {
        // sum(sum1, sum2, sum3, ...) = sum1 + sum2 + sum3 + ...
        self.update_batch(states)
    }

    fn state(&self) -> Result<Vec<ScalarValue>> {
        Ok(vec![self.sum.clone()])
    }

    fn evaluate(&self) -> Result<ScalarValue> {
        Ok(self.sum.clone())
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::physical_plan::expressions::col;
    use crate::{error::Result, generic_test_op};
    use arrow::datatypes::*;
    use arrow::record_batch::RecordBatch;

    #[test]
    fn sum_i32() -> Result<()> {
        let a: ArrayRef = Arc::new(Int32Array::from(vec![1, 2, 3, 4, 5]));
        generic_test_op!(
            a,
            DataType::Int32,
            Sum,
            ScalarValue::from(15i64),
            DataType::Int64
        )
    }

    #[test]
    fn sum_i32_with_nulls() -> Result<()> {
        let a: ArrayRef = Arc::new(Int32Array::from(vec![
            Some(1),
            None,
            Some(3),
            Some(4),
            Some(5),
        ]));
        generic_test_op!(
            a,
            DataType::Int32,
            Sum,
            ScalarValue::from(13i64),
            DataType::Int64
        )
    }

    #[test]
    fn sum_i32_all_nulls() -> Result<()> {
        let a: ArrayRef = Arc::new(Int32Array::from(vec![None, None]));
        generic_test_op!(
            a,
            DataType::Int32,
            Sum,
            ScalarValue::Int64(None),
            DataType::Int64
        )
    }

    #[test]
    fn sum_u32() -> Result<()> {
        let a: ArrayRef =
            Arc::new(UInt32Array::from(vec![1_u32, 2_u32, 3_u32, 4_u32, 5_u32]));
        generic_test_op!(
            a,
            DataType::UInt32,
            Sum,
            ScalarValue::from(15u64),
            DataType::UInt64
        )
    }

    #[test]
    fn sum_f32() -> Result<()> {
        let a: ArrayRef =
            Arc::new(Float32Array::from(vec![1_f32, 2_f32, 3_f32, 4_f32, 5_f32]));
        generic_test_op!(
            a,
            DataType::Float32,
            Sum,
            ScalarValue::from(15_f32),
            DataType::Float32
        )
    }

    #[test]
    fn sum_f64() -> Result<()> {
        let a: ArrayRef =
            Arc::new(Float64Array::from(vec![1_f64, 2_f64, 3_f64, 4_f64, 5_f64]));
        generic_test_op!(
            a,
            DataType::Float64,
            Sum,
            ScalarValue::from(15_f64),
            DataType::Float64
        )
    }

    fn aggregate(
        batch: &RecordBatch,
        agg: Arc<dyn AggregateExpr>,
    ) -> Result<ScalarValue> {
        let mut accum = agg.create_accumulator()?;
        let expr = agg.expressions();
        let values = expr
            .iter()
            .map(|e| e.evaluate(batch))
            .map(|r| r.map(|v| v.into_array(batch.num_rows())))
            .collect::<Result<Vec<_>>>()?;
        accum.update_batch(&values)?;
        accum.evaluate()
    }
}