This is an automated email from the ASF dual-hosted git repository.
github-merge-queue[bot] pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/datafusion.git
The following commit(s) were added to refs/heads/main by this push:
new 3e058f09c9 feat: Implement state conversion for remaining group
accumulators (#23275)
3e058f09c9 is described below
commit 3e058f09c91a18c8f36b71a656231603b33888c2
Author: linfeng <[email protected]>
AuthorDate: Sat Jul 11 08:26:41 2026 +0800
feat: Implement state conversion for remaining group accumulators (#23275)
## Which issue does this PR close?
<!--
We generally require a GitHub issue to be filed for all bug fixes and
enhancements and this helps us generate change logs for our releases.
You can link an issue to this PR using the GitHub syntax. For example
`Closes #123` indicates that this PR will close issue #123.
-->
- Part of #23081.
## Rationale for this change
<!--
Why are you proposing this change? If this is already explained clearly
in the issue then this section is not needed.
Explaining clearly why changes are proposed helps reviewers understand
your changes and offer better suggestions for fixes.
-->
#23081 proposes making `GroupsAccumulator::convert_to_state` mandatory
and removing `supports_convert_to_state`.
Before that API cleanup can happen, all existing `GroupsAccumulator`
implementations need a correct `convert_to_state` implementation.
This PR is the first step toward that cleanup: it fills in missing
`convert_to_state` implementations for the remaining group accumulators
that currently participate in grouped aggregation.
## What changes are included in this PR?
<!--
There is no need to duplicate the description in the issue here but it
is sometimes worth providing a summary of the individual changes in this
PR.
-->
This PR adds `convert_to_state` support for:
- `PrimitiveDistinctCountGroupsAccumulator`
- `HllGroupsAccumulator` for `approx_distinct`
- `VarianceGroupsAccumulator`
- `StddevGroupsAccumulator`
- `CorrelationGroupsAccumulator`
- `GeometricMeanGroupsAccumulator` in the advanced UDAF example
## Are these changes tested?
<!--
We typically require tests for all PRs in order to:
1. Prevent the code from being accidentally broken by subsequent changes
2. Serve as another way to document the expected behavior of the code
If tests are not included in your PR, please explain why (for example,
are they covered by existing tests)?
-->
Yes. This PR adds unit tests for the new `convert_to_state`
implementations.
## Are there any user-facing changes?
<!--
If there are user-facing changes then we may require documentation to be
updated before approving the PR.
-->
<!--
If there are any breaking changes to public APIs, please add the `api
change` label.
-->
No. This PR does not remove `supports_convert_to_state` yet and does not
change the public `GroupsAccumulator` API.
---
datafusion-examples/examples/udf/advanced_udaf.rs | 43 ++++++-
.../tests/user_defined/user_defined_aggregates.rs | 16 +++
.../src/aggregate/count_distinct/groups.rs | 121 +++++++++++++++++-
.../functions-aggregate/src/approx_distinct.rs | 129 +++++++++++++++++++
datafusion/functions-aggregate/src/correlation.rs | 141 +++++++++++++++++++++
datafusion/functions-aggregate/src/stddev.rs | 16 ++-
datafusion/functions-aggregate/src/variance.rs | 123 ++++++++++++++++++
.../test_files/aggregate_skip_partial.slt | 38 ++++++
8 files changed, 622 insertions(+), 5 deletions(-)
diff --git a/datafusion-examples/examples/udf/advanced_udaf.rs
b/datafusion-examples/examples/udf/advanced_udaf.rs
index b990740159..096753d2b5 100644
--- a/datafusion-examples/examples/udf/advanced_udaf.rs
+++ b/datafusion-examples/examples/udf/advanced_udaf.rs
@@ -23,8 +23,10 @@ use datafusion::{arrow::datatypes::DataType,
logical_expr::Volatility};
use std::sync::Arc;
use arrow::array::{
- ArrayRef, AsArray, Float32Array, PrimitiveArray, PrimitiveBuilder,
UInt32Array,
+ Array, ArrayRef, AsArray, BooleanArray, Float32Array, PrimitiveArray,
+ PrimitiveBuilder, UInt32Array,
};
+use arrow::buffer::NullBuffer;
use arrow::datatypes::{ArrowNativeTypeOp, ArrowPrimitiveType, Float64Type,
UInt32Type};
use arrow::record_batch::RecordBatch;
use arrow_schema::FieldRef;
@@ -237,7 +239,7 @@ impl GroupsAccumulator for GeometricMeanGroupsAccumulator {
&mut self,
values: &[ArrayRef],
group_indices: &[usize],
- opt_filter: Option<&arrow::array::BooleanArray>,
+ opt_filter: Option<&BooleanArray>,
total_num_groups: usize,
) -> Result<()> {
assert_eq!(values.len(), 1, "single argument to update_batch");
@@ -359,6 +361,43 @@ impl GroupsAccumulator for GeometricMeanGroupsAccumulator {
])
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ assert_eq!(values.len(), 1, "single argument to convert_to_state");
+
+ let prods = values[0]
+ .as_primitive::<Float64Type>()
+ .clone()
+ .with_data_type(self.prod_data_type.clone());
+ let counts = UInt32Array::from_value(1, prods.len());
+
+ let filter_nulls = opt_filter.map(|filter| {
+ let validity = match filter.nulls() {
+ Some(nulls) => filter.values() & nulls.inner(),
+ None => filter.values().clone(),
+ };
+ NullBuffer::new(validity)
+ });
+ let nulls = NullBuffer::union(filter_nulls.as_ref(), prods.nulls());
+
+ let prods =
+ PrimitiveArray::<Float64Type>::new(prods.values().clone(),
nulls.clone())
+ .with_data_type(self.prod_data_type.clone());
+ let counts = UInt32Array::new(counts.values().clone(), nulls);
+
+ Ok(vec![
+ Arc::new(prods) as ArrayRef,
+ Arc::new(counts) as ArrayRef,
+ ])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
self.counts.capacity() * size_of::<u32>()
+ self.prods.capacity() * size_of::<Float64Type>()
diff --git a/datafusion/core/tests/user_defined/user_defined_aggregates.rs
b/datafusion/core/tests/user_defined/user_defined_aggregates.rs
index b895cb9c7c..1d4b222301 100644
--- a/datafusion/core/tests/user_defined/user_defined_aggregates.rs
+++ b/datafusion/core/tests/user_defined/user_defined_aggregates.rs
@@ -877,6 +877,22 @@ impl GroupsAccumulator for TestGroupsAccumulator {
Ok(())
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ _opt_filter: Option<&arrow::array::BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ let len = values.first().map_or(0, |value| value.len());
+ Ok(vec![
+ Arc::new(PrimitiveArray::<UInt64Type>::from_value(self.result,
len))
+ as ArrayRef,
+ ])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
size_of::<u64>()
}
diff --git
a/datafusion/functions-aggregate-common/src/aggregate/count_distinct/groups.rs
b/datafusion/functions-aggregate-common/src/aggregate/count_distinct/groups.rs
index 60fe0388c4..986d4ec0d7 100644
---
a/datafusion/functions-aggregate-common/src/aggregate/count_distinct/groups.rs
+++
b/datafusion/functions-aggregate-common/src/aggregate/count_distinct/groups.rs
@@ -16,7 +16,8 @@
// under the License.
use arrow::array::{
- ArrayRef, AsArray, BooleanArray, Int64Array, ListArray, PrimitiveArray,
+ Array, ArrayRef, AsArray, BooleanArray, Int64Array, ListArray, ListBuilder,
+ PrimitiveArray, PrimitiveBuilder,
};
use arrow::buffer::{OffsetBuffer, ScalarBuffer};
use arrow::datatypes::{ArrowPrimitiveType, Field};
@@ -182,9 +183,127 @@ where
Ok(())
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> datafusion_common::Result<Vec<ArrayRef>> {
+ debug_assert_eq!(values.len(), 1);
+ let arr = values[0].as_primitive::<T>();
+
+ let values_builder = PrimitiveBuilder::<T>::with_capacity(arr.len());
+ let mut builder = ListBuilder::new(values_builder)
+ .with_field(Arc::new(Field::new_list_field(T::DATA_TYPE, true)));
+
+ for row in 0..arr.len() {
+ let included = arr.is_valid(row)
+ && opt_filter
+ .is_none_or(|filter| filter.is_valid(row) &&
filter.value(row));
+ if included {
+ builder.values().append_value(arr.value(row));
+ }
+ builder.append(true);
+ }
+
+ Ok(vec![Arc::new(builder.finish())])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
size_of::<Self>()
+ self.seen.capacity() * (size_of::<(usize, T::Native)>() +
size_of::<u64>())
+ self.counts.capacity() * size_of::<i64>()
}
}
+
+#[cfg(test)]
+mod tests {
+ use super::*;
+ use arrow::array::Int32Array;
+ use arrow::datatypes::Int32Type;
+ use datafusion_common::Result;
+
+ #[test]
+ fn convert_to_state_roundtrips_through_merge() -> Result<()> {
+ let values = Arc::new(Int32Array::from(vec![
+ Some(1),
+ Some(2),
+ Some(2),
+ None,
+ Some(3),
+ Some(4),
+ Some(5),
+ Some(5),
+ ])) as ArrayRef;
+ let filter = BooleanArray::from(vec![
+ Some(true),
+ Some(true),
+ Some(true),
+ Some(true),
+ None,
+ Some(true),
+ Some(true),
+ Some(true),
+ ]);
+ let group_indices = vec![0usize, 1, 0, 1, 0, 0, 0, 0];
+
+ let mut direct =
PrimitiveDistinctCountGroupsAccumulator::<Int32Type>::new();
+ direct.update_batch(
+ std::slice::from_ref(&values),
+ &group_indices,
+ Some(&filter),
+ 2,
+ )?;
+ let direct = direct.evaluate(EmitTo::All)?;
+
+ let converter =
PrimitiveDistinctCountGroupsAccumulator::<Int32Type>::new();
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&values),
Some(&filter))?;
+ assert_eq!(state[0].null_count(), 0);
+ let mut merged =
PrimitiveDistinctCountGroupsAccumulator::<Int32Type>::new();
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let merged = merged.evaluate(EmitTo::All)?;
+
+ assert_eq!(
+ direct.as_any().downcast_ref::<Int64Array>().unwrap(),
+ merged.as_any().downcast_ref::<Int64Array>().unwrap()
+ );
+ Ok(())
+ }
+
+ #[test]
+ fn convert_to_state_preserves_empty_and_filtered_rows() -> Result<()> {
+ let converter =
PrimitiveDistinctCountGroupsAccumulator::<Int32Type>::new();
+ let empty_values =
+ Arc::new(Int32Array::from(Vec::<Option<i32>>::new())) as ArrayRef;
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&empty_values),
None)?;
+ assert_eq!(state[0].len(), 0);
+ assert_eq!(state[0].null_count(), 0);
+
+ let values = Arc::new(Int32Array::from(vec![Some(1), Some(2), None]))
as ArrayRef;
+ let filter = BooleanArray::from(vec![Some(false), None, Some(false)]);
+ let group_indices = vec![0usize, 1, 0];
+
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&values),
Some(&filter))?;
+ assert_eq!(state[0].len(), values.len());
+ assert_eq!(state[0].null_count(), 0);
+ let list_state = state[0].as_list::<i32>();
+ for row in 0..list_state.len() {
+ assert_eq!(list_state.value_length(row), 0);
+ }
+
+ let mut merged =
PrimitiveDistinctCountGroupsAccumulator::<Int32Type>::new();
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let result = merged.evaluate(EmitTo::All)?;
+ assert_eq!(
+ result.as_any().downcast_ref::<Int64Array>().unwrap(),
+ &Int64Array::from(vec![0, 0])
+ );
+ Ok(())
+ }
+}
diff --git a/datafusion/functions-aggregate/src/approx_distinct.rs
b/datafusion/functions-aggregate/src/approx_distinct.rs
index 0e35b47d64..74bc9ad6cb 100644
--- a/datafusion/functions-aggregate/src/approx_distinct.rs
+++ b/datafusion/functions-aggregate/src/approx_distinct.rs
@@ -583,6 +583,43 @@ impl GroupsAccumulator for HllGroupsAccumulator {
Ok(vec![Arc::new(builder.finish())])
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ assert_eq!(values.len(), 1, "single argument to convert_to_state");
+ let array = values[0].as_ref();
+ let mut hashes = vec![0; array.len()];
+ create_hashes([array], &HLL_HASH_STATE, &mut hashes)?;
+
+ let filter_nulls = opt_filter.map(filter_to_nulls);
+ let value_nulls = array.logical_nulls();
+ let combined_nulls =
+ NullBuffer::union(filter_nulls.as_ref(), value_nulls.as_ref());
+
+ let mut builder = BinaryBuilder::new();
+ let mut scratch = Vec::new();
+ for (row, hash) in hashes.into_iter().enumerate() {
+ if combined_nulls
+ .as_ref()
+ .is_none_or(|nulls| nulls.is_valid(row))
+ {
+ scratch.clear();
+ scratch.extend_from_slice(&hash.to_le_bytes());
+ builder.append_value(&scratch);
+ } else {
+ builder.append_value([]);
+ }
+ }
+
+ Ok(vec![Arc::new(builder.finish())])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
self.groups.capacity() * size_of::<GroupHll>()
+ self.allocated_bytes
@@ -1081,6 +1118,98 @@ mod tests {
assert_eq!(counts.value(0), expected);
}
+ #[test]
+ fn groups_convert_to_state_roundtrips_through_merge() {
+ let values: ArrayRef = Arc::new(Int64Array::from(vec![
+ Some(1),
+ Some(2),
+ Some(2),
+ None,
+ Some(3),
+ ]));
+ let filter = BooleanArray::from(vec![
+ Some(true),
+ Some(true),
+ Some(true),
+ Some(true),
+ None,
+ ]);
+ let group_indices = vec![0usize, 1, 0, 1, 0];
+
+ let mut direct = HllGroupsAccumulator::new();
+ direct
+ .update_batch(
+ std::slice::from_ref(&values),
+ &group_indices,
+ Some(&filter),
+ 2,
+ )
+ .unwrap();
+ let direct = direct
+ .evaluate(EmitTo::All)
+ .unwrap()
+ .as_any()
+ .downcast_ref::<UInt64Array>()
+ .unwrap()
+ .clone();
+
+ let converter = HllGroupsAccumulator::new();
+ let state = converter
+ .convert_to_state(std::slice::from_ref(&values), Some(&filter))
+ .unwrap();
+ assert_eq!(state[0].null_count(), 0);
+ let mut merged = HllGroupsAccumulator::new();
+ merged.merge_batch(&state, &group_indices, 2).unwrap();
+ let merged = merged
+ .evaluate(EmitTo::All)
+ .unwrap()
+ .as_any()
+ .downcast_ref::<UInt64Array>()
+ .unwrap()
+ .clone();
+
+ assert_eq!(direct, merged);
+ }
+
+ #[test]
+ fn groups_convert_to_state_preserves_empty_and_filtered_rows() {
+ let converter = HllGroupsAccumulator::new();
+ let empty_values: ArrayRef =
+ Arc::new(Int64Array::from(Vec::<Option<i64>>::new()));
+ let state = converter
+ .convert_to_state(std::slice::from_ref(&empty_values), None)
+ .unwrap();
+ assert_eq!(state[0].len(), 0);
+ assert_eq!(state[0].null_count(), 0);
+
+ let values: ArrayRef =
+ Arc::new(Int64Array::from(vec![Some(1), Some(2), None]));
+ let filter = BooleanArray::from(vec![Some(false), None,
Some(false)]);
+ let group_indices = vec![0usize, 1, 0];
+ let state = converter
+ .convert_to_state(std::slice::from_ref(&values), Some(&filter))
+ .unwrap();
+ assert_eq!(state[0].len(), values.len());
+ assert_eq!(state[0].null_count(), 0);
+ let state =
state[0].as_any().downcast_ref::<BinaryArray>().unwrap();
+ for row in 0..state.len() {
+ assert_eq!(state.value(row), b"");
+ }
+
+ let mut merged = HllGroupsAccumulator::new();
+ merged
+ .merge_batch(&[Arc::new(state.clone())], &group_indices, 2)
+ .unwrap();
+ let result = merged
+ .evaluate(EmitTo::All)
+ .unwrap()
+ .as_any()
+ .downcast_ref::<UInt64Array>()
+ .unwrap()
+ .clone();
+ assert_eq!(result, UInt64Array::from(vec![0, 0]));
+ }
+
/// Regression: a short (≤ 12-byte) Utf8View string must hash
identically
/// in an all-inline batch and in a mixed batch that also contains a
long
/// string (which forces a data buffer).
diff --git a/datafusion/functions-aggregate/src/correlation.rs
b/datafusion/functions-aggregate/src/correlation.rs
index 7fcf4bb61f..2e90cac6d9 100644
--- a/datafusion/functions-aggregate/src/correlation.rs
+++ b/datafusion/functions-aggregate/src/correlation.rs
@@ -489,6 +489,61 @@ impl GroupsAccumulator for CorrelationGroupsAccumulator {
])
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ assert_eq!(values.len(), 2, "two arguments to convert_to_state");
+ let array_x = downcast_array::<Float64Array>(&values[0]);
+ let array_y = downcast_array::<Float64Array>(&values[1]);
+
+ let len = array_x.len();
+ let mut counts = Vec::with_capacity(len);
+ let mut sum_x = Vec::with_capacity(len);
+ let mut sum_y = Vec::with_capacity(len);
+ let mut sum_xy = Vec::with_capacity(len);
+ let mut sum_xx = Vec::with_capacity(len);
+ let mut sum_yy = Vec::with_capacity(len);
+
+ for row in 0..len {
+ let included = array_x.is_valid(row)
+ && array_y.is_valid(row)
+ && opt_filter
+ .is_none_or(|filter| filter.is_valid(row) &&
filter.value(row));
+ if included {
+ let x = array_x.value(row);
+ let y = array_y.value(row);
+ counts.push(1);
+ sum_x.push(x);
+ sum_y.push(y);
+ sum_xy.push(x * y);
+ sum_xx.push(x * x);
+ sum_yy.push(y * y);
+ } else {
+ counts.push(0);
+ sum_x.push(0.0);
+ sum_y.push(0.0);
+ sum_xy.push(0.0);
+ sum_xx.push(0.0);
+ sum_yy.push(0.0);
+ }
+ }
+
+ Ok(vec![
+ Arc::new(UInt64Array::from(counts)),
+ Arc::new(Float64Array::from(sum_x)),
+ Arc::new(Float64Array::from(sum_y)),
+ Arc::new(Float64Array::from(sum_xy)),
+ Arc::new(Float64Array::from(sum_xx)),
+ Arc::new(Float64Array::from(sum_yy)),
+ ])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn merge_batch(
&mut self,
values: &[ArrayRef],
@@ -593,4 +648,90 @@ mod tests {
});
assert!(result.is_err());
}
+
+ #[test]
+ fn convert_to_state_roundtrips_through_merge() -> Result<()> {
+ let x = Arc::new(Float64Array::from(vec![
+ Some(1.0),
+ Some(2.0),
+ None,
+ Some(4.0),
+ Some(8.0),
+ Some(16.0),
+ Some(32.0),
+ ])) as ArrayRef;
+ let y = Arc::new(Float64Array::from(vec![
+ Some(2.0),
+ Some(4.0),
+ Some(6.0),
+ None,
+ Some(16.0),
+ Some(32.0),
+ Some(64.0),
+ ])) as ArrayRef;
+ let filter = BooleanArray::from(vec![
+ Some(true),
+ Some(false),
+ Some(true),
+ Some(true),
+ None,
+ Some(true),
+ Some(true),
+ ]);
+ let values = vec![x, y];
+ let group_indices = vec![0, 1, 0, 1, 0, 0, 0];
+
+ let mut direct = CorrelationGroupsAccumulator::new();
+ direct.update_batch(&values, &group_indices, Some(&filter), 2)?;
+ let direct = direct.evaluate(EmitTo::All)?;
+
+ let converter = CorrelationGroupsAccumulator::new();
+ let state = converter.convert_to_state(&values, Some(&filter))?;
+ let mut merged = CorrelationGroupsAccumulator::new();
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let merged = merged.evaluate(EmitTo::All)?;
+
+ assert_eq!(
+ direct.as_any().downcast_ref::<Float64Array>().unwrap(),
+ merged.as_any().downcast_ref::<Float64Array>().unwrap()
+ );
+ Ok(())
+ }
+
+ #[test]
+ fn convert_to_state_preserves_empty_and_filtered_rows() -> Result<()> {
+ let converter = CorrelationGroupsAccumulator::new();
+ let empty_values = vec![
+ Arc::new(Float64Array::from(Vec::<Option<f64>>::new())) as
ArrayRef,
+ Arc::new(Float64Array::from(Vec::<Option<f64>>::new())) as
ArrayRef,
+ ];
+ let state = converter.convert_to_state(&empty_values, None)?;
+ for state_array in &state {
+ assert_eq!(state_array.len(), 0);
+ assert_eq!(state_array.null_count(), 0);
+ }
+
+ let values = vec![
+ Arc::new(Float64Array::from(vec![Some(1.0), Some(2.0), None])) as
ArrayRef,
+ Arc::new(Float64Array::from(vec![Some(2.0), None, Some(4.0)])) as
ArrayRef,
+ ];
+ let filter = BooleanArray::from(vec![Some(false), None, Some(false)]);
+ let group_indices = vec![0, 1, 0];
+ let state = converter.convert_to_state(&values, Some(&filter))?;
+ for state_array in &state {
+ assert_eq!(state_array.len(), values[0].len());
+ assert_eq!(state_array.null_count(), 0);
+ }
+
+ let counts = state[0].as_any().downcast_ref::<UInt64Array>().unwrap();
+ assert_eq!(counts, &UInt64Array::from(vec![0, 0, 0]));
+
+ let mut merged = CorrelationGroupsAccumulator::new();
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let result = merged.evaluate(EmitTo::All)?;
+ let result = result.as_any().downcast_ref::<Float64Array>().unwrap();
+ assert_eq!(result.len(), 2);
+ assert_eq!(result.null_count(), 2);
+ Ok(())
+ }
}
diff --git a/datafusion/functions-aggregate/src/stddev.rs
b/datafusion/functions-aggregate/src/stddev.rs
index f0482b23d1..a31517b93e 100644
--- a/datafusion/functions-aggregate/src/stddev.rs
+++ b/datafusion/functions-aggregate/src/stddev.rs
@@ -22,7 +22,7 @@ use std::hash::Hash;
use std::mem::align_of_val;
use std::sync::Arc;
-use arrow::array::Float64Array;
+use arrow::array::{BooleanArray, Float64Array};
use arrow::datatypes::FieldRef;
use arrow::{array::ArrayRef, datatypes::DataType, datatypes::Field};
use datafusion_common::ScalarValue;
@@ -318,7 +318,7 @@ impl GroupsAccumulator for StddevGroupsAccumulator {
&mut self,
values: &[ArrayRef],
group_indices: &[usize],
- opt_filter: Option<&arrow::array::BooleanArray>,
+ opt_filter: Option<&BooleanArray>,
total_num_groups: usize,
) -> Result<()> {
self.variance
@@ -345,6 +345,18 @@ impl GroupsAccumulator for StddevGroupsAccumulator {
self.variance.state(emit_to)
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ self.variance.convert_to_state(values, opt_filter)
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
self.variance.size()
}
diff --git a/datafusion/functions-aggregate/src/variance.rs
b/datafusion/functions-aggregate/src/variance.rs
index 551fcfe120..0278ce2c23 100644
--- a/datafusion/functions-aggregate/src/variance.rs
+++ b/datafusion/functions-aggregate/src/variance.rs
@@ -580,6 +580,44 @@ impl GroupsAccumulator for VarianceGroupsAccumulator {
])
}
+ fn convert_to_state(
+ &self,
+ values: &[ArrayRef],
+ opt_filter: Option<&BooleanArray>,
+ ) -> Result<Vec<ArrayRef>> {
+ assert_eq!(values.len(), 1, "single argument to convert_to_state");
+ let values = as_float64_array(&values[0])?;
+
+ let len = values.len();
+ let mut counts = Vec::with_capacity(len);
+ let mut means = Vec::with_capacity(len);
+ let mut m2s = Vec::with_capacity(len);
+
+ for row in 0..len {
+ if values.is_valid(row)
+ && opt_filter
+ .is_none_or(|filter| filter.is_valid(row) &&
filter.value(row))
+ {
+ counts.push(1);
+ means.push(values.value(row));
+ } else {
+ counts.push(0);
+ means.push(0.0);
+ }
+ m2s.push(0.0);
+ }
+
+ Ok(vec![
+ Arc::new(UInt64Array::new(counts.into(), None)),
+ Arc::new(Float64Array::new(means.into(), None)),
+ Arc::new(Float64Array::new(m2s.into(), None)),
+ ])
+ }
+
+ fn supports_convert_to_state(&self) -> bool {
+ true
+ }
+
fn size(&self) -> usize {
self.m2s.capacity() * size_of::<f64>()
+ self.means.capacity() * size_of::<f64>()
@@ -679,4 +717,89 @@ mod tests {
assert_eq!(result.value(0), 1.0);
Ok(())
}
+
+ #[test]
+ fn convert_to_state_roundtrips_through_merge() -> Result<()> {
+ let values = Arc::new(Float64Array::from(vec![
+ Some(1.0),
+ Some(2.0),
+ None,
+ Some(4.0),
+ Some(8.0),
+ Some(16.0),
+ Some(32.0),
+ ])) as ArrayRef;
+ let filter = BooleanArray::from(vec![
+ Some(true),
+ Some(false),
+ Some(true),
+ None,
+ Some(true),
+ Some(true),
+ Some(true),
+ ]);
+ let group_indices = vec![0, 1, 0, 1, 0, 0, 0];
+
+ let mut direct = VarianceGroupsAccumulator::new(StatsType::Sample);
+ direct.update_batch(
+ std::slice::from_ref(&values),
+ &group_indices,
+ Some(&filter),
+ 2,
+ )?;
+ let direct = direct.evaluate(EmitTo::All)?;
+
+ let converter = VarianceGroupsAccumulator::new(StatsType::Sample);
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&values),
Some(&filter))?;
+ let mut merged = VarianceGroupsAccumulator::new(StatsType::Sample);
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let merged = merged.evaluate(EmitTo::All)?;
+
+ let direct = direct.as_any().downcast_ref::<Float64Array>().unwrap();
+ let merged = merged.as_any().downcast_ref::<Float64Array>().unwrap();
+ assert_eq!(direct.len(), merged.len());
+ for row in 0..direct.len() {
+ assert_eq!(direct.is_null(row), merged.is_null(row));
+ if direct.is_valid(row) {
+ assert!((direct.value(row) - merged.value(row)).abs() < 1e-12);
+ }
+ }
+ Ok(())
+ }
+
+ #[test]
+ fn convert_to_state_preserves_empty_and_filtered_rows() -> Result<()> {
+ let converter = VarianceGroupsAccumulator::new(StatsType::Sample);
+ let empty_values =
+ Arc::new(Float64Array::from(Vec::<Option<f64>>::new())) as
ArrayRef;
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&empty_values),
None)?;
+ for state_array in &state {
+ assert_eq!(state_array.len(), 0);
+ assert_eq!(state_array.null_count(), 0);
+ }
+
+ let values =
+ Arc::new(Float64Array::from(vec![Some(1.0), Some(2.0), None])) as
ArrayRef;
+ let filter = BooleanArray::from(vec![Some(false), None, Some(false)]);
+ let group_indices = vec![0, 1, 0];
+ let state =
+ converter.convert_to_state(std::slice::from_ref(&values),
Some(&filter))?;
+ for state_array in &state {
+ assert_eq!(state_array.len(), values.len());
+ assert_eq!(state_array.null_count(), 0);
+ }
+
+ let counts = state[0].as_any().downcast_ref::<UInt64Array>().unwrap();
+ assert_eq!(counts, &UInt64Array::from(vec![0, 0, 0]));
+
+ let mut merged = VarianceGroupsAccumulator::new(StatsType::Sample);
+ merged.merge_batch(&state, &group_indices, 2)?;
+ let result = merged.evaluate(EmitTo::All)?;
+ let result = result.as_any().downcast_ref::<Float64Array>().unwrap();
+ assert_eq!(result.len(), 2);
+ assert_eq!(result.null_count(), 2);
+ Ok(())
+ }
}
diff --git a/datafusion/sqllogictest/test_files/aggregate_skip_partial.slt
b/datafusion/sqllogictest/test_files/aggregate_skip_partial.slt
index 195441a119..2ed4c9921f 100644
--- a/datafusion/sqllogictest/test_files/aggregate_skip_partial.slt
+++ b/datafusion/sqllogictest/test_files/aggregate_skip_partial.slt
@@ -636,6 +636,44 @@ FROM aggregate_test_100_null GROUP BY c2 ORDER BY c2;
4 3
5 6
+# Test variance and stddev with nullable fields and filters
+query IRRR
+SELECT c2,
+ var_samp(c11) FILTER (WHERE c3 > 0),
+ stddev_samp(c11) FILTER (WHERE c3 > 0),
+ stddev_pop(c11) FILTER (WHERE c3 > 0)
+FROM aggregate_test_100_null GROUP BY c2 ORDER BY c2;
+----
+1 0.085786074994 0.29289259976 0.276141791266
+2 0.070769227104 0.266024861816 0.24884347232
+3 0.087515365779 0.295829960922 0.270054571305
+4 0.038697229167 0.196716113134 0.182123731489
+5 0.081817232141 0.286037116719 0.23354832782
+
+# Test corr with nullable fields and filters
+query IR
+SELECT c2,
+ corr(c3, c11) FILTER (WHERE c5 > 0)
+FROM aggregate_test_100_null GROUP BY c2 ORDER BY c2;
+----
+1 -0.38515658251
+2 0.414489249329
+3 -0.796447131429
+4 -0.938568248748
+5 -0.051058146743
+
+# Test count distinct with nullable fields and filters
+query II
+SELECT c2,
+ count(distinct c3) FILTER (WHERE c11 > 0.5)
+FROM aggregate_test_100_null GROUP BY c2 ORDER BY c2;
+----
+1 10
+2 6
+3 3
+4 3
+5 6
+
# Test median with nullable fields and filter
query IRR
SELECT c2,
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
