This is an automated email from the ASF dual-hosted git repository.
Jefffrey 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 368dec1274 perf(ipc): use Vec instead of BufferBuilder for run-ends
re-encoding (#10245) (#11005)
368dec1274 is described below
commit 368dec12746f32ee50ff81d3707a42d01713f109
Author: Seowoo Jang <[email protected]>
AuthorDate: Mon Sep 7 09:34:23 2026 +0900
perf(ipc): use Vec instead of BufferBuilder for run-ends re-encoding
(#10245) (#11005)
## What does this PR do?
Replaces `BufferBuilder::<R::Native>` with `Vec::<R::Native>` when
re-encoding sliced run-ends in `into_zero_offset_run_array`
(`arrow-ipc/src/writer.rs`).
This is part of #10245 — Rust's `Vec` has a highly optimized
implementation, and switching from `BufferBuilder`/`OffsetBufferBuilder`
to `Vec` typically yields a speedup.
No behavior change: the re-encoded run-ends buffer is byte-identical.
Existing run-array roundtrip tests (`test_run_array_unslice`,
`test_roundtrip_stream_run_array_sliced`) cover the re-encoding path for
all slice lengths and both slice offsets.
## Checklist
- [x] Ran `cargo test -p arrow-ipc` (131 passed)
- [x] Ran `cargo clippy -p arrow-ipc --all-targets`
- [x] Ran `cargo fmt -p arrow-ipc -- --check`
---
arrow-ipc/src/writer.rs | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)
diff --git a/arrow-ipc/src/writer.rs b/arrow-ipc/src/writer.rs
index 57a5ef7448..30555a738a 100644
--- a/arrow-ipc/src/writer.rs
+++ b/arrow-ipc/src/writer.rs
@@ -32,7 +32,6 @@ use std::sync::Arc;
use flatbuffers::FlatBufferBuilder;
-use arrow_array::builder::BufferBuilder;
use arrow_array::cast::*;
use arrow_array::types::{Int16Type, Int32Type, Int64Type, RunEndIndexType};
use arrow_array::*;
@@ -1295,17 +1294,18 @@ fn into_zero_offset_run_array<R: RunEndIndexType>(
// build new run_ends array by subtracting offset from run ends.
let offset = R::Native::usize_as(run_ends.offset());
- let mut builder = BufferBuilder::<R::Native>::new(physical_length);
+ let mut run_ends_values = Vec::<R::Native>::with_capacity(physical_length);
for run_end_value in
&run_ends.values()[start_physical_index..end_physical_index] {
- builder.append(run_end_value.sub_wrapping(offset));
+ run_ends_values.push(run_end_value.sub_wrapping(offset));
}
- builder.append(R::Native::from_usize(run_array.len()).unwrap());
+ run_ends_values.push(R::Native::from_usize(run_array.len()).unwrap());
+ let offset_buffer = Buffer::from_vec(run_ends_values);
let new_run_ends = unsafe {
// Safety:
// The function builds a valid run_ends array and hence need not be
validated.
ArrayDataBuilder::new(R::DATA_TYPE)
.len(physical_length)
- .add_buffer(builder.finish())
+ .add_buffer(offset_buffer)
.build_unchecked()
};
