andygrove commented on PR #5457: URL: https://github.com/apache/datafusion-comet/pull/5457#issuecomment-5876454790
This is a light fully automated review since there are so many PRs open. The new `Performance (tuned 2026-08-27 ...)` entry at `docs/source/contributor-guide/expression-audits/conversion_funcs.md:41` reports the 12-92% gains measured against `8b072abb`, which is an earlier head of this PR that added the unconditional `take`. `cast_array` on `main` has no preparation step at all, so once this merges the audit log will credit nested casts with a tuning win that `main` never needed. Against `main`, the whitelisted shapes simply keep the path they already had. For example `struct_int_to_long/width=1/nulls=10%` at 2.60 us is the same no-copy `cast_struct_to_struct` call that `main` makes today. The shapes that still compact cost more than they do on `main`. `list_int_to_long/width=4/nulls=50%` measures 134.95 us, while the no-null row, which casts the same 32,768 child values through the path `main` takes at every null ratio, measures 26.61 us. `optimizing_expressions.md` asks for the baseline to be captured on `main`, and these entries exist so contributor s can tell what has already been optimized. Could this line be dropped, or rewritten against `main` so it describes what the PR actually adds for nested casts? -- 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] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
