Mola-maker opened a new pull request, #12327: URL: https://github.com/apache/arrow-rs/pull/12327
Closes #12326 # Rationale for this change `DecimalType::is_valid_decimal_precision` is called once per value, and the four implementations in `arrow-array` are not `#[inline]`. They are not generic, so a caller in another crate depends on LTO to inline them. The issue reports that in DataFusion Comet the call stays out of line even in release builds with thin LTO, with the grouped decimal SUM/AVG accumulators 10-37% slower when the call is forced out of line. # What changes are included in this PR? Adds `#[inline]` to the four `is_valid_decimal_precision` implementations (`Decimal32Type`, `Decimal64Type`, `Decimal128Type`, `Decimal256Type`) in `arrow-array/src/types.rs`. # Are these changes tested? No new tests: the change is an inlining hint only and does not change behavior, so there is no observable behavior a regression test could assert. The existing decimal test suites pass unchanged (arrow-array 29 passed, arrow-cast 20 passed, arrow-data 2 passed; `cargo fmt --check` and `cargo clippy -p arrow-array --all-targets --all-features` clean). The inlining effect was verified directly with a small out-of-tree release binary calling `<Decimal128Type as DecimalType>::is_valid_decimal_precision` per value (no LTO): before this change the caller contained an out-of-line call to the trait method; after, the caller contains no call at all and the bounds check (precision <= 38 plus the MIN/MAX table lookups) is inlined. Before: the caller contains an out-of-line call to the trait method <img width="781" height="175" alt="arrow-rs-12326-verify-before" src="https://github.com/user-attachments/assets/aa2ba1af-6a88-4857-adb1-f9836e105d1f" /> After: the caller contains no call; the bounds check is inlined, and the decimal test suites pass <img width="628" height="280" alt="arrow-rs-12326-verify-after" src="https://github.com/user-attachments/assets/80ed88dc-41cf-45df-9ce5-d822afd4264d" /> # Are there any significant user-facing changes? No. # AI usage disclosure AI assistance was used for verification and drafting; the change itself is the four `#[inline]` attributes requested in the issue. -- 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]
