RanaPriyansh opened a new pull request, #11138: URL: https://github.com/apache/arrow-rs/pull/11138
# Which issue does this PR close? Related to #11129. This benchmark contribution does not resolve the decoder initialization question. # Rationale for this change The IPC reader benchmark covers uncompressed and ZSTD streams but lacks LZ4 coverage. # What changes are included in this PR? Add a mixed-type LZ4 reader case and fixed-size binary cases with matching uncompressed controls. Each iteration reads ten batches with validation enabled. The controls use 1,024 rows and 262,144 value bytes per batch. One column encodes two buffers, including a 128-byte validity bitmap. Sixteen columns encode 32 buffers, including 2,048 bitmap bytes. Total decoded buffer bytes are 262,272 and 264,192, respectively, excluding IPC prefixes and padding. Criterion reports time per ten batches. These workloads can measure sensitivity to column and buffer counts, but cannot isolate decoder initialization or zero-fill costs. Fixture setup verifies decoded batch equality outside the timed loop. Fixed-size binary fixtures require compression of every nonempty buffer. The mixed fixture permits uncompressed fallback and requires some compressed buffers. # Are these changes tested? All five new cases passed Criterion's functional test mode with `lz4,zstd`. The existing uncompressed case and both new controls also passed with no default features. Formatting and targeted Clippy passed: ```text cargo clippy -p arrow-ipc --features lz4,zstd --bench ipc_reader --locked --offline --no-deps -- -D warnings ``` Validation used Rust 1.95.0 on aarch64 macOS. Dependency-inclusive Clippy reports four existing warnings in unchanged `arrow-data/src/data.rs`. No performance improvement is claimed. # Are there any user-facing changes? No library or API changes. AI assistance generated the benchmark additions and helped review the changes. -- 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: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
