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 e82873eb1b Fix the u8 shift overflow in the buffer_bit_ops benchmark
(#11175)
e82873eb1b is described below
commit e82873eb1bccb3619f3879dae4100681f1b81944
Author: Amit Vijapur <[email protected]>
AuthorDate: Wed Sep 23 02:47:01 2026 +0100
Fix the u8 shift overflow in the buffer_bit_ops benchmark (#11175)
# Which issue does this PR close?
Part of #10952.
# Rationale for this change
Running the benches once under `cargo test --benches` (the check #10952
asks for) turns up one bench that cannot even start: `buffer_bit_ops`
panics at line 31 with `attempt to shift left with overflow`.
`0b01010101 << i` is assigned into a `u8` slot, so the literal infers
`u8` and the shift overflows for every `i >= 8`. `cargo bench` builds in
release, where the shift amount is masked and the bench runs, which is
why it has never shown up.
# What changes are included in this PR?
The two shifts become `wrapping_shl`, which masks the shift amount the
same way release mode does, so the fixture bytes are identical to what
the benchmark has been measuring all along (checked for `i` in `0..64`).
rustfmt split the expression over three lines.
# Are these changes tested?
`cargo test -p arrow --features test_utils --bench buffer_bit_ops`
panics on `main` and passes here. The byte-identity check was a small
standalone program comparing the new expression against the masked
semantics of the old one.
# Are there any user-facing changes?
No.
---
arrow/benches/buffer_bit_ops.rs | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/arrow/benches/buffer_bit_ops.rs b/arrow/benches/buffer_bit_ops.rs
index 2304779c38..dbbd166d60 100644
--- a/arrow/benches/buffer_bit_ops.rs
+++ b/arrow/benches/buffer_bit_ops.rs
@@ -28,7 +28,9 @@ fn create_buffer(size: usize) -> Buffer {
let mut result = MutableBuffer::new(size).with_bitset(size, false);
for i in 0..size {
- result.as_slice_mut()[i] = 0b01010101 << i << (i % 4);
+ result.as_slice_mut()[i] = 0b01010101u8
+ .wrapping_shl(i as u32)
+ .wrapping_shl((i % 4) as u32);
}
result.into()
