This is an automated email from the ASF dual-hosted git repository.
alamb pushed a commit to branch main
in repository https://gitbox.apache.org/repos/asf/arrow-rs.git
The following commit(s) were added to refs/heads/main by this push:
new aa61e07246 [arrow-select] Replace `ArrayData` with direct `Array`
construction in filter kernels (#9986)
aa61e07246 is described below
commit aa61e0724612e8eb9e576de2903827e9d6c23a8d
Author: Liam Bao <[email protected]>
AuthorDate: Mon May 25 07:18:05 2026 -0400
[arrow-select] Replace `ArrayData` with direct `Array` construction in
filter kernels (#9986)
# 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.
-->
- Part of #9298.
# 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.
-->
# 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.
-->
- Replaces several `ArrayDataBuilder` paths in
`arrow-select/src/filter.rs` with direct typed array constructors.
- Adds a small helper for filtered null buffers that reuses the
already-computed null count.
# 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)?
If this PR claims a performance improvement, please include evidence
such as benchmark results.
-->
Covered by exsiting tests
# 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 call them out.
-->
No
---------
Co-authored-by: Andrew Lamb <[email protected]>
---
arrow-select/src/filter.rs | 231 ++++++++++++++++-----------------------------
1 file changed, 81 insertions(+), 150 deletions(-)
diff --git a/arrow-select/src/filter.rs b/arrow-select/src/filter.rs
index e95d01f2b5..aaab9d2020 100644
--- a/arrow-select/src/filter.rs
+++ b/arrow-select/src/filter.rs
@@ -26,9 +26,10 @@ use arrow_array::types::{
ArrowDictionaryKeyType, ArrowPrimitiveType, ByteArrayType, ByteViewType,
RunEndIndexType,
};
use arrow_array::*;
-use arrow_buffer::{ArrowNativeType, BooleanBuffer, NullBuffer, RunEndBuffer,
bit_util};
+use arrow_buffer::{
+ ArrowNativeType, BooleanBuffer, NullBuffer, OffsetBuffer, RunEndBuffer,
ScalarBuffer, bit_util,
+};
use arrow_buffer::{Buffer, MutableBuffer};
-use arrow_data::ArrayDataBuilder;
use arrow_data::bit_iterator::{BitIndexIterator, BitSliceIterator};
use arrow_data::transform::MutableArrayData;
use arrow_schema::*;
@@ -408,6 +409,22 @@ impl FilterPredicate {
pub fn count(&self) -> usize {
self.count
}
+
+ /// Filters the given `nulls` buffer using this predicate.
+ ///
+ /// Returns `None` when there is nothing to track in the output, either
+ /// because the input `nulls` was `None`, the input had no nulls, or the
+ /// filtered result has no nulls. Otherwise returns the filtered
+ /// [`NullBuffer`] with its precomputed null count.
+ pub fn filter_nulls(&self, nulls: Option<&NullBuffer>) ->
Option<NullBuffer> {
+ let (null_count, nulls) = filter_null_mask(nulls, self)?;
+ let buffer = BooleanBuffer::new(nulls, 0, self.count);
+
+ debug_assert_eq!(null_count, buffer.len() - buffer.count_set_bits());
+ // SAFETY: `filter_null_mask` derived `null_count` from `buffer`, so it
+ // matches the number of unset bits as required by `new_unchecked`.
+ Some(unsafe { NullBuffer::new_unchecked(buffer, null_count) })
+ }
}
fn filter_array(values: &dyn Array, predicate: &FilterPredicate) ->
Result<ArrayRef, ArrowError> {
@@ -624,18 +641,11 @@ fn filter_bits(buffer: &BooleanBuffer, predicate:
&FilterPredicate) -> Buffer {
/// `filter` implementation for boolean buffers
fn filter_boolean(array: &BooleanArray, predicate: &FilterPredicate) ->
BooleanArray {
- let values = filter_bits(array.values(), predicate);
-
- let mut builder = ArrayDataBuilder::new(DataType::Boolean)
- .len(predicate.count)
- .add_buffer(values);
+ let buffer = filter_bits(array.values(), predicate);
+ let values = BooleanBuffer::new(buffer, 0, predicate.count);
+ let nulls = predicate.filter_nulls(array.nulls());
- if let Some((null_count, nulls)) = filter_null_mask(array.nulls(),
predicate) {
- builder = builder.null_count(null_count).null_bit_buffer(Some(nulls));
- }
-
- let data = unsafe { builder.build_unchecked() };
- BooleanArray::from(data)
+ BooleanArray::new(values, nulls)
}
#[inline(never)]
@@ -681,18 +691,17 @@ fn filter_primitive<T>(array: &PrimitiveArray<T>,
predicate: &FilterPredicate) -
where
T: ArrowPrimitiveType,
{
- let values = array.values();
- let buffer = filter_native(values, predicate);
- let mut builder = ArrayDataBuilder::new(array.data_type().clone())
- .len(predicate.count)
- .add_buffer(buffer);
-
- if let Some((null_count, nulls)) = filter_null_mask(array.nulls(),
predicate) {
- builder = builder.null_count(null_count).null_bit_buffer(Some(nulls));
+ let buffer = filter_native(array.values(), predicate);
+ let values = ScalarBuffer::new(buffer, 0, predicate.count);
+ let nulls = predicate.filter_nulls(array.nulls());
+ let filtered = PrimitiveArray::new(values, nulls);
+
+ // Avoid the compatibility check when the physical type already matches.
+ if array.data_type() == &T::DATA_TYPE {
+ filtered
+ } else {
+ filtered.with_data_type(array.data_type().clone())
}
-
- let data = unsafe { builder.build_unchecked() };
- PrimitiveArray::from(data)
}
/// [`FilterBytes`] is created from a source [`GenericByteArray`] and can be
@@ -824,17 +833,15 @@ where
IterationStrategy::All | IterationStrategy::None => unreachable!(),
}
- let mut builder = ArrayDataBuilder::new(T::DATA_TYPE)
- .len(predicate.count)
- .add_buffer(filter.dst_offsets.into())
- .add_buffer(filter.dst_values.into());
-
- if let Some((null_count, nulls)) = filter_null_mask(array.nulls(),
predicate) {
- builder = builder.null_count(null_count).null_bit_buffer(Some(nulls));
- }
+ // SAFETY: `dst_offsets` starts at `[0]` and only grows by the running
+ // `cur_offset`, so it is monotonically non-decreasing.
+ let offsets = unsafe {
OffsetBuffer::new_unchecked(filter.dst_offsets.into()) };
+ let nulls = predicate.filter_nulls(array.nulls());
- let data = unsafe { builder.build_unchecked() };
- GenericByteArray::from(data)
+ // SAFETY: `offsets` index into `dst_values` by construction, and each slot
+ // is a byte-for-byte copy from `array`, so UTF-8 validity (if any) is
preserved.
+ // Length invariant: `offsets.len() - 1 == predicate.count == nulls.len()`.
+ unsafe { GenericByteArray::new_unchecked(offsets,
filter.dst_values.into(), nulls) }
}
/// `filter` implementation for byte view arrays.
@@ -843,17 +850,14 @@ fn filter_byte_view<T: ByteViewType>(
predicate: &FilterPredicate,
) -> GenericByteViewArray<T> {
let new_view_buffer = filter_native(array.views(), predicate);
-
- let mut builder = ArrayDataBuilder::new(T::DATA_TYPE)
- .len(predicate.count)
- .add_buffer(new_view_buffer)
- .add_buffers(array.data_buffers().to_vec());
-
- if let Some((null_count, nulls)) = filter_null_mask(array.nulls(),
predicate) {
- builder = builder.null_count(null_count).null_bit_buffer(Some(nulls));
- }
-
- GenericByteViewArray::from(unsafe { builder.build_unchecked() })
+ let views = ScalarBuffer::new(new_view_buffer, 0, predicate.count);
+ let buffers = array.data_buffers().to_vec();
+ let nulls = predicate.filter_nulls(array.nulls());
+
+ // SAFETY: each view is copied unchanged from `array.views()` and `buffers`
+ // is the same buffer list, so every view still points to an in-bounds
+ // (and, for strings, UTF-8 valid) range.
+ unsafe { GenericByteViewArray::new_unchecked(views, buffers, nulls) }
}
fn filter_fixed_size_binary(
@@ -902,16 +906,10 @@ fn filter_fixed_size_binary(
}
IterationStrategy::All | IterationStrategy::None => unreachable!(),
};
- let mut builder = ArrayDataBuilder::new(array.data_type().clone())
- .len(predicate.count)
- .add_buffer(buffer.into());
- if let Some((null_count, nulls)) = filter_null_mask(array.nulls(),
predicate) {
- builder = builder.null_count(null_count).null_bit_buffer(Some(nulls));
- }
+ let nulls = predicate.filter_nulls(array.nulls());
- let data = unsafe { builder.build_unchecked() };
- FixedSizeBinaryArray::from(data)
+ FixedSizeBinaryArray::new(array.value_length(), buffer.into(), nulls)
}
/// `filter` implementation for dictionaries
@@ -992,24 +990,19 @@ fn filter_list_view<OffsetType: OffsetSizeTrait>(
let filtered_offsets = filter_native::<OffsetType>(array.offsets(),
predicate);
let filtered_sizes = filter_native::<OffsetType>(array.sizes(), predicate);
- // Filter the nulls
- let nulls = if let Some((null_count, nulls)) =
filter_null_mask(array.nulls(), predicate) {
- let buffer = BooleanBuffer::new(nulls, 0, predicate.count);
-
- Some(unsafe { NullBuffer::new_unchecked(buffer, null_count) })
- } else {
- None
+ let field = match array.data_type() {
+ DataType::ListView(field) | DataType::LargeListView(field) =>
field.clone(),
+ _ => unreachable!(),
};
-
- let list_data = ArrayDataBuilder::new(array.data_type().clone())
- .nulls(nulls)
- .buffers(vec![filtered_offsets, filtered_sizes])
- .child_data(vec![array.values().to_data()])
- .len(predicate.count);
-
- let list_data = unsafe { list_data.build_unchecked() };
-
- GenericListViewArray::from(list_data)
+ let offsets = ScalarBuffer::new(filtered_offsets, 0, predicate.count);
+ let sizes = ScalarBuffer::new(filtered_sizes, 0, predicate.count);
+ let values = array.values().clone();
+ let nulls = predicate.filter_nulls(array.nulls());
+
+ // SAFETY: each `(offset, size)` pair is copied unchanged from `array` and
+ // indexes into the same `values` child, so every range stays in-bounds.
+ // `field` and `values`' data type are unchanged from `array`.
+ unsafe { GenericListViewArray::new_unchecked(field, offsets, sizes,
values, nulls) }
}
#[cfg(test)]
@@ -1018,7 +1011,6 @@ mod tests {
use arrow_array::builder::*;
use arrow_array::cast::as_run_array;
use arrow_array::types::*;
- use arrow_data::ArrayData;
use rand::distr::uniform::{UniformSampler, UniformUsize};
use rand::distr::{Alphanumeric, StandardUniform};
use rand::prelude::*;
@@ -1494,49 +1486,22 @@ mod tests {
#[test]
fn test_filter_list_array() {
- let value_data = ArrayData::builder(DataType::Int32)
- .len(8)
- .add_buffer(Buffer::from_slice_ref([0, 1, 2, 3, 4, 5, 6, 7]))
- .build()
- .unwrap();
-
- let value_offsets = Buffer::from_slice_ref([0i64, 3, 6, 8, 8]);
-
- let list_data_type =
-
DataType::LargeList(Arc::new(Field::new_list_field(DataType::Int32, false)));
- let list_data = ArrayData::builder(list_data_type)
- .len(4)
- .add_buffer(value_offsets)
- .add_child_data(value_data)
- .null_bit_buffer(Some(Buffer::from([0b00000111])))
- .build()
- .unwrap();
-
+ let field = Arc::new(Field::new_list_field(DataType::Int32, false));
+ let offsets = OffsetBuffer::new(vec![0i64, 3, 6, 8, 8].into());
+ let value_array = Arc::new(Int32Array::from_iter_values(0..8));
+ let nulls = Some(NullBuffer::from(vec![true, true, true, false]));
// a = [[0, 1, 2], [3, 4, 5], [6, 7], null]
- let a = LargeListArray::from(list_data);
+ let a = LargeListArray::new(field.clone(), offsets, value_array,
nulls);
let b = BooleanArray::from(vec![false, true, false, true]);
let result = filter(&a, &b).unwrap();
// expected: [[3, 4, 5], null]
- let value_data = ArrayData::builder(DataType::Int32)
- .len(3)
- .add_buffer(Buffer::from_slice_ref([3, 4, 5]))
- .build()
- .unwrap();
-
- let value_offsets = Buffer::from_slice_ref([0i64, 3, 3]);
-
- let list_data_type =
-
DataType::LargeList(Arc::new(Field::new_list_field(DataType::Int32, false)));
- let expected = ArrayData::builder(list_data_type)
- .len(2)
- .add_buffer(value_offsets)
- .add_child_data(value_data)
- .null_bit_buffer(Some(Buffer::from([0b00000001])))
- .build()
- .unwrap();
+ let offsets = OffsetBuffer::new(vec![0i64, 3, 3].into());
+ let value_array = Arc::new(Int32Array::from_iter_values([3, 4, 5]));
+ let nulls = Some(NullBuffer::from(vec![true, false]));
+ let expected: ArrayRef = Arc::new(LargeListArray::new(field, offsets,
value_array, nulls));
- assert_eq!(&make_array(expected), &result);
+ assert_eq!(&expected, &result);
}
fn test_case_filter_list_view<T: OffsetSizeTrait>() {
@@ -1719,14 +1684,7 @@ mod tests {
let truncated_length = mask_len - offset - truncate;
- let data = ArrayDataBuilder::new(DataType::Boolean)
- .len(truncated_length)
- .offset(offset)
- .add_buffer(buffer)
- .build()
- .unwrap();
-
- let filter = BooleanArray::from(data);
+ let filter = BooleanArray::new(BooleanBuffer::new(buffer, offset,
truncated_length), None);
let slice_bits: Vec<_> = SlicesIterator::new(&filter)
.flat_map(|(start, end)| start..end)
@@ -1949,18 +1907,9 @@ mod tests {
#[test]
fn test_filter_fixed_size_list_arrays() {
- let value_data = ArrayData::builder(DataType::Int32)
- .len(9)
- .add_buffer(Buffer::from_slice_ref([0, 1, 2, 3, 4, 5, 6, 7, 8]))
- .build()
- .unwrap();
- let list_data_type = DataType::new_fixed_size_list(DataType::Int32, 3,
false);
- let list_data = ArrayData::builder(list_data_type)
- .len(3)
- .add_child_data(value_data)
- .build()
- .unwrap();
- let array = FixedSizeListArray::from(list_data);
+ let field = Arc::new(Field::new_list_field(DataType::Int32, false));
+ let value_array = Arc::new(Int32Array::from_iter_values(0..9));
+ let array = FixedSizeListArray::new(field, 3, value_array, None);
let filter_array = BooleanArray::from(vec![true, false, false]);
@@ -1996,28 +1945,10 @@ mod tests {
#[test]
fn test_filter_fixed_size_list_arrays_with_null() {
- let value_data = ArrayData::builder(DataType::Int32)
- .len(10)
- .add_buffer(Buffer::from_slice_ref([0, 1, 2, 3, 4, 5, 6, 7, 8, 9]))
- .build()
- .unwrap();
-
- // Set null buts for the nested array:
- // [[0, 1], null, null, [6, 7], [8, 9]]
- // 01011001 00000001
- let mut null_bits: [u8; 1] = [0; 1];
- bit_util::set_bit(&mut null_bits, 0);
- bit_util::set_bit(&mut null_bits, 3);
- bit_util::set_bit(&mut null_bits, 4);
-
- let list_data_type = DataType::new_fixed_size_list(DataType::Int32, 2,
false);
- let list_data = ArrayData::builder(list_data_type)
- .len(5)
- .add_child_data(value_data)
- .null_bit_buffer(Some(Buffer::from(null_bits)))
- .build()
- .unwrap();
- let array = FixedSizeListArray::from(list_data);
+ let field = Arc::new(Field::new_list_field(DataType::Int32, false));
+ let value_array = Arc::new(Int32Array::from_iter_values(0..10));
+ let nulls = Some(NullBuffer::from(vec![true, false, false, true,
true]));
+ let array = FixedSizeListArray::new(field, 2, value_array, nulls);
let filter_array = BooleanArray::from(vec![true, true, false, true,
false]);
