You are viewing a plain text version of this content. The canonical link for it is here.
Posted to github@arrow.apache.org by "alamb (via GitHub)" <gi...@apache.org> on 2023/05/11 11:43:12 UTC
[GitHub] [arrow-datafusion] alamb commented on a diff in pull request #6276: feat: support bitwise and boolean aggregate functions
alamb commented on code in PR #6276:
URL: https://github.com/apache/arrow-datafusion/pull/6276#discussion_r1191024150
##########
datafusion/expr/src/type_coercion/aggregates.rs:
##########
@@ -395,6 +440,17 @@ pub fn avg_sum_type(arg_type: &DataType) -> Result<DataType> {
}
}
+pub fn is_bit_and_or_xor_support_arg_type(arg_type: &DataType) -> bool {
+ matches!(
+ arg_type,
+ arg_type if NUMERICS.contains(arg_type)
+ )
Review Comment:
Is there any reason to use a `matches!` here? it seems like this could just be `NUMERICS.contains(arg_type)` 🤔
##########
datafusion/core/tests/sqllogictests/test_files/aggregate.slt:
##########
@@ -1390,6 +1429,63 @@ as values
('2021-01-01T05:11:10.432', 'Row 3');
+statement ok
Review Comment:
Given postgres also runs these queries what do you think about also adding them into a `pg_compat` test in https://github.com/apache/arrow-datafusion/tree/main/datafusion/core/tests/sqllogictests/test_files/pg_compat?
##########
datafusion/core/tests/sqllogictests/test_files/aggregate.slt:
##########
@@ -1390,6 +1429,63 @@ as values
('2021-01-01T05:11:10.432', 'Row 3');
+statement ok
+create table bit_aggregate_functions (
+ c1 SMALLINT NOT NULL,
+ c2 SMALLINT NOT NULL,
+ c3 SMALLINT,
+)
+as values
+ (5, 10, 11),
+ (33, 11, null),
+ (9, 12, null);
+
+# query_bit_and
+query III
+SELECT bit_and(c1), bit_and(c2), bit_and(c3) FROM bit_aggregate_functions
+----
+1 8 11
Review Comment:
I double checked this is consistent with postgres 👍
##########
docs/source/user-guide/expressions.md:
##########
@@ -216,6 +216,11 @@ Unlike to some databases the math functions in Datafusion works the same way as
| approx_median(expr) | Calculates an approximation of the median for `expr`. |
| approx_percentile_cont(expr, percentile) | Calculates an approximation of the specified `percentile` for `expr`. |
| approx_percentile_cont_with_weight(expr, weight_expr, percentile) | Calculates an approximation of the specified `percentile` for `expr` and `weight_expr`. |
+| bit_and(expr) | Computes the bitwise AND of all non-null input values for `expr`. |
Review Comment:
Thank you for also updating the documentation
##########
datafusion/physical-expr/src/aggregate/bool_and_or.rs:
##########
@@ -0,0 +1,643 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{ArrayRef, BooleanArray},
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+
+fn bool_and(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.false_count() == 0)
+}
+
+fn bool_or(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.true_count() != 0)
+}
+
+// Bool and/Bool or aggregation can take Dictionary encode input but always produces unpacked
+// (aka non Dictionary) output. We need to adjust the output data type to reflect this.
+// The reason bool and/bool or aggregate produces unpacked output because there is only one
+// bool and/bool or value per group; there is no needs to keep them Dictionary encode
+fn bool_and_or_aggregate_data_type(input_type: DataType) -> DataType {
+ if let DataType::Dictionary(_, value_type) = input_type {
+ *value_type
+ } else {
+ input_type
+ }
+}
+
+// returns the new value after bool_and/bool_or with the new values, taking nullability into account
+macro_rules! typed_bool_and_or_batch {
+ ($VALUES:expr, $ARRAYTYPE:ident, $SCALAR:ident, $OP:ident) => {{
+ let array = downcast_value!($VALUES, $ARRAYTYPE);
+ let delta = $OP(array);
+ Ok(ScalarValue::$SCALAR(delta))
+ }};
+}
+
+// bool_and/bool_or the array and returns a ScalarValue of its corresponding type.
+macro_rules! bool_and_or_batch {
+ ($VALUES:expr, $OP:ident) => {{
+ match $VALUES.data_type() {
+ DataType::Boolean => {
+ typed_bool_and_or_batch!($VALUES, BooleanArray, Boolean, $OP)
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "Bool and/Bool or is not expected to receive the type {e:?}"
+ )));
+ }
+ }
+ }};
+}
+
+/// dynamically-typed bool_and(array) -> ScalarValue
+fn bool_and_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_and)
+}
+
+/// dynamically-typed bool_or(array) -> ScalarValue
+fn bool_or_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_or)
+}
+
+// bool_and/bool_or of two scalar values.
+macro_rules! typed_bool_and_or {
+ ($VALUE:expr, $DELTA:expr, $SCALAR:ident, $OP:ident) => {{
Review Comment:
Given that you added `ScalarValue::bitand` etc, maybe you could also add `ScalarValue::and` ? I think the overall code complexity would be the same but then the logic wouldn't be hidden in macros inside the aggregator
A similar comment applies to ScalarValue:or etc
##########
datafusion/physical-expr/src/aggregate/bool_and_or.rs:
##########
@@ -0,0 +1,643 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{ArrayRef, BooleanArray},
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+
+fn bool_and(array: &BooleanArray) -> Option<bool> {
Review Comment:
👍 these are good implementations
##########
datafusion/physical-expr/src/aggregate/bool_and_or.rs:
##########
@@ -0,0 +1,643 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{ArrayRef, BooleanArray},
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+
+fn bool_and(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.false_count() == 0)
+}
+
+fn bool_or(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.true_count() != 0)
+}
+
+// Bool and/Bool or aggregation can take Dictionary encode input but always produces unpacked
+// (aka non Dictionary) output. We need to adjust the output data type to reflect this.
+// The reason bool and/bool or aggregate produces unpacked output because there is only one
+// bool and/bool or value per group; there is no needs to keep them Dictionary encode
+fn bool_and_or_aggregate_data_type(input_type: DataType) -> DataType {
+ if let DataType::Dictionary(_, value_type) = input_type {
+ *value_type
+ } else {
+ input_type
+ }
+}
+
+// returns the new value after bool_and/bool_or with the new values, taking nullability into account
+macro_rules! typed_bool_and_or_batch {
+ ($VALUES:expr, $ARRAYTYPE:ident, $SCALAR:ident, $OP:ident) => {{
+ let array = downcast_value!($VALUES, $ARRAYTYPE);
+ let delta = $OP(array);
+ Ok(ScalarValue::$SCALAR(delta))
+ }};
+}
+
+// bool_and/bool_or the array and returns a ScalarValue of its corresponding type.
+macro_rules! bool_and_or_batch {
+ ($VALUES:expr, $OP:ident) => {{
+ match $VALUES.data_type() {
+ DataType::Boolean => {
+ typed_bool_and_or_batch!($VALUES, BooleanArray, Boolean, $OP)
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "Bool and/Bool or is not expected to receive the type {e:?}"
+ )));
+ }
+ }
+ }};
+}
+
+/// dynamically-typed bool_and(array) -> ScalarValue
+fn bool_and_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_and)
+}
+
+/// dynamically-typed bool_or(array) -> ScalarValue
+fn bool_or_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_or)
+}
+
+// bool_and/bool_or of two scalar values.
+macro_rules! typed_bool_and_or {
+ ($VALUE:expr, $DELTA:expr, $SCALAR:ident, $OP:ident) => {{
+ ScalarValue::$SCALAR(match ($VALUE, $DELTA) {
+ (None, None) => None,
+ (Some(a), None) => Some(*a),
+ (None, Some(b)) => Some(*b),
+ (Some(a), Some(b)) => Some((*a).$OP(*b)),
+ })
+ }};
+}
+
+// bool_and/bool_or of two scalar values.
+macro_rules! typed_bool_and_or_v2 {
+ ($INDEX:ident, $ACC:ident, $SCALAR:expr, $TYPE:ident, $OP:ident) => {{
+ paste::item! {
+ match $SCALAR {
+ None => {}
+ Some(v) => $ACC.[<$OP _ $TYPE>]($INDEX, *v as $TYPE)
+ }
+ }
+ }};
+}
+
+// bool_and/bool_or of two scalar values of the same type
+macro_rules! bool_and_or {
+ ($VALUE:expr, $DELTA:expr, $OP:ident) => {{
+ Ok(match ($VALUE, $DELTA) {
+ (ScalarValue::Boolean(lhs), ScalarValue::Boolean(rhs)) => {
+ typed_bool_and_or!(lhs, rhs, Boolean, $OP)
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "BOOL AND/BOOL OR is not expected to receive scalars of incompatible types {:?}",
+ e
+ )))
+ }
+ })
+ }};
+}
+
+macro_rules! bool_and_or_v2 {
+ ($INDEX:ident, $ACC:ident, $SCALAR:expr, $OP:ident) => {{
+ Ok(match $SCALAR {
+ ScalarValue::Boolean(rhs) => {
+ typed_bool_and_or_v2!($INDEX, $ACC, rhs, bool, $OP)
+ }
+ ScalarValue::Null => {
+ // do nothing
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "BOOL AND/BOOL OR is not expected to receive scalars of incompatible types {:?}",
+ e
+ )))
+ }
+ })
+ }};
+}
+
+/// the bool_and of two scalar values
+pub fn booland(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue> {
+ bool_and_or!(lhs, rhs, bitand)
+}
+
+pub fn bool_and_row(
+ index: usize,
+ accessor: &mut RowAccessor,
+ s: &ScalarValue,
+) -> Result<()> {
+ bool_and_or_v2!(index, accessor, s, bitand)
+}
+
+/// the bool_or of two scalar values
+pub fn boolor(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue> {
+ bool_and_or!(lhs, rhs, bitor)
+}
+
+pub fn bool_or_row(
+ index: usize,
+ accessor: &mut RowAccessor,
+ s: &ScalarValue,
+) -> Result<()> {
+ bool_and_or_v2!(index, accessor, s, bitor)
+}
+
+/// BOOL_AND aggregate expression
+#[derive(Debug, Clone)]
+pub struct BoolAnd {
+ name: String,
+ pub data_type: DataType,
+ expr: Arc<dyn PhysicalExpr>,
+ nullable: bool,
+}
+
+impl BoolAnd {
+ /// Create a new BOOL_AND aggregate function
+ pub fn new(
+ expr: Arc<dyn PhysicalExpr>,
+ name: impl Into<String>,
+ data_type: DataType,
+ ) -> Self {
+ Self {
+ name: name.into(),
+ expr,
+ data_type: bool_and_or_aggregate_data_type(data_type),
+ nullable: true,
+ }
+ }
+}
+
+impl AggregateExpr for BoolAnd {
+ /// 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 create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
+ Ok(Box::new(BoolAndAccumulator::try_new(&self.data_type)?))
+ }
+
+ fn state_fields(&self) -> Result<Vec<Field>> {
+ Ok(vec![Field::new(
+ format_state_name(&self.name, "bool_and"),
+ self.data_type.clone(),
+ self.nullable,
+ )])
+ }
+
+ fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
+ vec![self.expr.clone()]
+ }
+
+ fn name(&self) -> &str {
+ &self.name
+ }
+
+ fn row_accumulator_supported(&self) -> bool {
+ is_row_accumulator_support_dtype(&self.data_type)
+ }
+
+ fn supports_bounded_execution(&self) -> bool {
+ true
Review Comment:
As I mention below I am not sure this aggregate supports bounded execution
##########
datafusion/physical-expr/src/aggregate/bool_and_or.rs:
##########
@@ -0,0 +1,643 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{ArrayRef, BooleanArray},
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+
+fn bool_and(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.false_count() == 0)
+}
+
+fn bool_or(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.true_count() != 0)
+}
+
+// Bool and/Bool or aggregation can take Dictionary encode input but always produces unpacked
Review Comment:
The idea of dictionary packing booleans seems bad to me (the encoded form would perform worse). I think we should not support this case. If we want to keep this code, can we please add a test ?
You can make such an array like
```
select arrow_cast(bool_col, 'Dictionary(Boolean, Int8)`)
```
##########
datafusion/physical-expr/src/aggregate/bool_and_or.rs:
##########
@@ -0,0 +1,643 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{ArrayRef, BooleanArray},
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+
+fn bool_and(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.false_count() == 0)
+}
+
+fn bool_or(array: &BooleanArray) -> Option<bool> {
+ if array.null_count() == array.len() {
+ return None;
+ }
+ Some(array.true_count() != 0)
+}
+
+// Bool and/Bool or aggregation can take Dictionary encode input but always produces unpacked
+// (aka non Dictionary) output. We need to adjust the output data type to reflect this.
+// The reason bool and/bool or aggregate produces unpacked output because there is only one
+// bool and/bool or value per group; there is no needs to keep them Dictionary encode
+fn bool_and_or_aggregate_data_type(input_type: DataType) -> DataType {
+ if let DataType::Dictionary(_, value_type) = input_type {
+ *value_type
+ } else {
+ input_type
+ }
+}
+
+// returns the new value after bool_and/bool_or with the new values, taking nullability into account
+macro_rules! typed_bool_and_or_batch {
+ ($VALUES:expr, $ARRAYTYPE:ident, $SCALAR:ident, $OP:ident) => {{
+ let array = downcast_value!($VALUES, $ARRAYTYPE);
+ let delta = $OP(array);
+ Ok(ScalarValue::$SCALAR(delta))
+ }};
+}
+
+// bool_and/bool_or the array and returns a ScalarValue of its corresponding type.
+macro_rules! bool_and_or_batch {
+ ($VALUES:expr, $OP:ident) => {{
+ match $VALUES.data_type() {
+ DataType::Boolean => {
+ typed_bool_and_or_batch!($VALUES, BooleanArray, Boolean, $OP)
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "Bool and/Bool or is not expected to receive the type {e:?}"
+ )));
+ }
+ }
+ }};
+}
+
+/// dynamically-typed bool_and(array) -> ScalarValue
+fn bool_and_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_and)
+}
+
+/// dynamically-typed bool_or(array) -> ScalarValue
+fn bool_or_batch(values: &ArrayRef) -> Result<ScalarValue> {
+ bool_and_or_batch!(values, bool_or)
+}
+
+// bool_and/bool_or of two scalar values.
+macro_rules! typed_bool_and_or {
+ ($VALUE:expr, $DELTA:expr, $SCALAR:ident, $OP:ident) => {{
+ ScalarValue::$SCALAR(match ($VALUE, $DELTA) {
+ (None, None) => None,
+ (Some(a), None) => Some(*a),
+ (None, Some(b)) => Some(*b),
+ (Some(a), Some(b)) => Some((*a).$OP(*b)),
+ })
+ }};
+}
+
+// bool_and/bool_or of two scalar values.
+macro_rules! typed_bool_and_or_v2 {
+ ($INDEX:ident, $ACC:ident, $SCALAR:expr, $TYPE:ident, $OP:ident) => {{
+ paste::item! {
+ match $SCALAR {
+ None => {}
+ Some(v) => $ACC.[<$OP _ $TYPE>]($INDEX, *v as $TYPE)
+ }
+ }
+ }};
+}
+
+// bool_and/bool_or of two scalar values of the same type
+macro_rules! bool_and_or {
+ ($VALUE:expr, $DELTA:expr, $OP:ident) => {{
+ Ok(match ($VALUE, $DELTA) {
+ (ScalarValue::Boolean(lhs), ScalarValue::Boolean(rhs)) => {
+ typed_bool_and_or!(lhs, rhs, Boolean, $OP)
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "BOOL AND/BOOL OR is not expected to receive scalars of incompatible types {:?}",
+ e
+ )))
+ }
+ })
+ }};
+}
+
+macro_rules! bool_and_or_v2 {
+ ($INDEX:ident, $ACC:ident, $SCALAR:expr, $OP:ident) => {{
+ Ok(match $SCALAR {
+ ScalarValue::Boolean(rhs) => {
+ typed_bool_and_or_v2!($INDEX, $ACC, rhs, bool, $OP)
+ }
+ ScalarValue::Null => {
+ // do nothing
+ }
+ e => {
+ return Err(DataFusionError::Internal(format!(
+ "BOOL AND/BOOL OR is not expected to receive scalars of incompatible types {:?}",
+ e
+ )))
+ }
+ })
+ }};
+}
+
+/// the bool_and of two scalar values
+pub fn booland(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue> {
+ bool_and_or!(lhs, rhs, bitand)
+}
+
+pub fn bool_and_row(
+ index: usize,
+ accessor: &mut RowAccessor,
+ s: &ScalarValue,
+) -> Result<()> {
+ bool_and_or_v2!(index, accessor, s, bitand)
+}
+
+/// the bool_or of two scalar values
+pub fn boolor(lhs: &ScalarValue, rhs: &ScalarValue) -> Result<ScalarValue> {
+ bool_and_or!(lhs, rhs, bitor)
+}
+
+pub fn bool_or_row(
+ index: usize,
+ accessor: &mut RowAccessor,
+ s: &ScalarValue,
+) -> Result<()> {
+ bool_and_or_v2!(index, accessor, s, bitor)
+}
+
+/// BOOL_AND aggregate expression
+#[derive(Debug, Clone)]
+pub struct BoolAnd {
+ name: String,
+ pub data_type: DataType,
+ expr: Arc<dyn PhysicalExpr>,
+ nullable: bool,
+}
+
+impl BoolAnd {
+ /// Create a new BOOL_AND aggregate function
+ pub fn new(
+ expr: Arc<dyn PhysicalExpr>,
+ name: impl Into<String>,
+ data_type: DataType,
+ ) -> Self {
+ Self {
+ name: name.into(),
+ expr,
+ data_type: bool_and_or_aggregate_data_type(data_type),
+ nullable: true,
+ }
+ }
+}
+
+impl AggregateExpr for BoolAnd {
+ /// 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 create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
+ Ok(Box::new(BoolAndAccumulator::try_new(&self.data_type)?))
+ }
+
+ fn state_fields(&self) -> Result<Vec<Field>> {
+ Ok(vec![Field::new(
+ format_state_name(&self.name, "bool_and"),
+ self.data_type.clone(),
+ self.nullable,
+ )])
+ }
+
+ fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
+ vec![self.expr.clone()]
+ }
+
+ fn name(&self) -> &str {
+ &self.name
+ }
+
+ fn row_accumulator_supported(&self) -> bool {
+ is_row_accumulator_support_dtype(&self.data_type)
+ }
+
+ fn supports_bounded_execution(&self) -> bool {
+ true
+ }
+
+ fn create_row_accumulator(
+ &self,
+ start_index: usize,
+ ) -> Result<Box<dyn RowAccumulator>> {
+ Ok(Box::new(BoolAndRowAccumulator::new(
+ start_index,
+ self.data_type.clone(),
+ )))
+ }
+
+ fn reverse_expr(&self) -> Option<Arc<dyn AggregateExpr>> {
+ Some(Arc::new(self.clone()))
+ }
+
+ fn create_sliding_accumulator(&self) -> Result<Box<dyn Accumulator>> {
+ Ok(Box::new(BoolAndAccumulator::try_new(&self.data_type)?))
+ }
+}
+
+impl PartialEq<dyn Any> for BoolAnd {
+ fn eq(&self, other: &dyn Any) -> bool {
+ down_cast_any_ref(other)
+ .downcast_ref::<Self>()
+ .map(|x| {
+ self.name == x.name
+ && self.data_type == x.data_type
+ && self.nullable == x.nullable
+ && self.expr.eq(&x.expr)
+ })
+ .unwrap_or(false)
+ }
+}
+
+#[derive(Debug)]
+struct BoolAndAccumulator {
+ bool_and: ScalarValue,
+}
+
+impl BoolAndAccumulator {
+ /// new bool_and accumulator
+ pub fn try_new(data_type: &DataType) -> Result<Self> {
+ Ok(Self {
+ bool_and: ScalarValue::try_from(data_type)?,
+ })
+ }
+}
+
+impl Accumulator for BoolAndAccumulator {
+ fn update_batch(&mut self, values: &[ArrayRef]) -> Result<()> {
+ let values = &values[0];
+ let delta = &bool_and_batch(values)?;
+ self.bool_and = booland(&self.bool_and, delta)?;
+ Ok(())
+ }
+
+ fn merge_batch(&mut self, states: &[ArrayRef]) -> Result<()> {
+ self.update_batch(states)
+ }
+
+ fn state(&self) -> Result<Vec<ScalarValue>> {
+ Ok(vec![self.bool_and.clone()])
+ }
+
+ fn evaluate(&self) -> Result<ScalarValue> {
+ Ok(self.bool_and.clone())
+ }
+
+ fn size(&self) -> usize {
+ std::mem::size_of_val(self) - std::mem::size_of_val(&self.bool_and)
+ + self.bool_and.size()
+ }
+}
+
+#[derive(Debug)]
+struct BoolAndRowAccumulator {
+ index: usize,
+ datatype: DataType,
+}
+
+impl BoolAndRowAccumulator {
+ pub fn new(index: usize, datatype: DataType) -> Self {
+ Self { index, datatype }
+ }
+}
+
+impl RowAccumulator for BoolAndRowAccumulator {
+ fn update_batch(
+ &mut self,
+ values: &[ArrayRef],
+ accessor: &mut RowAccessor,
+ ) -> Result<()> {
+ let values = &values[0];
+ let delta = &bool_and_batch(values)?;
+ bool_and_row(self.index, accessor, delta)
+ }
+
+ fn update_scalar_values(
+ &mut self,
+ values: &[ScalarValue],
+ accessor: &mut RowAccessor,
+ ) -> Result<()> {
+ let value = &values[0];
+ bool_and_row(self.index, accessor, value)
+ }
+
+ fn update_scalar(
+ &mut self,
+ value: &ScalarValue,
+ accessor: &mut RowAccessor,
+ ) -> Result<()> {
+ bool_and_row(self.index, accessor, value)
+ }
+
+ fn merge_batch(
+ &mut self,
+ states: &[ArrayRef],
+ accessor: &mut RowAccessor,
+ ) -> Result<()> {
+ self.update_batch(states, accessor)
+ }
+
+ fn evaluate(&self, accessor: &RowAccessor) -> Result<ScalarValue> {
+ Ok(accessor.get_as_scalar(&self.datatype, self.index))
+ }
+
+ #[inline(always)]
+ fn state_index(&self) -> usize {
+ self.index
+ }
+}
+
+/// BOOL_OR aggregate expression
+#[derive(Debug, Clone)]
+pub struct BoolOr {
+ name: String,
+ pub data_type: DataType,
+ expr: Arc<dyn PhysicalExpr>,
+ nullable: bool,
+}
+
+impl BoolOr {
+ /// Create a new BOOL_OR aggregate function
+ pub fn new(
+ expr: Arc<dyn PhysicalExpr>,
+ name: impl Into<String>,
+ data_type: DataType,
+ ) -> Self {
+ Self {
+ name: name.into(),
+ expr,
+ data_type: bool_and_or_aggregate_data_type(data_type),
+ nullable: true,
+ }
+ }
+}
+
+impl AggregateExpr for BoolOr {
+ /// 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 create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
+ Ok(Box::new(BoolOrAccumulator::try_new(&self.data_type)?))
+ }
+
+ fn state_fields(&self) -> Result<Vec<Field>> {
+ Ok(vec![Field::new(
+ format_state_name(&self.name, "bool_or"),
+ self.data_type.clone(),
+ self.nullable,
+ )])
+ }
+
+ fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
+ vec![self.expr.clone()]
+ }
+
+ fn name(&self) -> &str {
+ &self.name
+ }
+
+ fn row_accumulator_supported(&self) -> bool {
+ is_row_accumulator_support_dtype(&self.data_type)
+ }
+
+ fn supports_bounded_execution(&self) -> bool {
+ true
Review Comment:
I am not sure this aggregator actually supports bounded execution (that are used in sliding window functions) -- to do so I think it would need to remember how many tries it has seen perhaps.
@mustafasrepo or @ozankabak can you double check this?
##########
datafusion/physical-expr/src/aggregate/bit_and_or_xor.rs:
##########
@@ -0,0 +1,1103 @@
+// 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::{AggregateExpr, PhysicalExpr};
+use arrow::datatypes::DataType;
+use arrow::{
+ array::{
+ ArrayRef, Int16Array, Int32Array, Int64Array, Int8Array, UInt16Array,
+ UInt32Array, UInt64Array, UInt8Array,
+ },
+ datatypes::Field,
+};
+use datafusion_common::{downcast_value, DataFusionError, Result, ScalarValue};
+use datafusion_expr::Accumulator;
+
+use crate::aggregate::row_accumulator::{
+ is_row_accumulator_support_dtype, RowAccumulator,
+};
+use crate::aggregate::utils::down_cast_any_ref;
+use crate::expressions::format_state_name;
+use arrow::array::Array;
+use arrow::array::PrimitiveArray;
+use arrow::datatypes::ArrowNativeTypeOp;
+use arrow::datatypes::ArrowNumericType;
+use datafusion_row::accessor::RowAccessor;
+use std::ops::BitAnd as BitAndImplementation;
+use std::ops::BitOr as BitOrImplementation;
+use std::ops::BitXor as BitXorImplementation;
+
+fn bit_and<T>(array: &PrimitiveArray<T>) -> Option<T::Native>
Review Comment:
I think we should strive to move these kernels into arrow-rs eventually.
Can you please file a ticket in arrow-rs to do so (or I can do to so if you prefer) and leave a comment here with a reference to that ticket?
--
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: github-unsubscribe@arrow.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org