I have filed a perf-related issue (#23463) for tracking. I intend to create a PR to fix this, today or soon.

I've approved the workflow run and reviewed the issue. I think this looks like a good change.

I am investigating the CI failures. At least some of them are strange to me. Some others passed when I ran them locally. Still looking.

Hah no, I pressed the approve button, sorry for the missing communication

Thank you! What is the process for these repeated CI failures? The Python Lint test passes locally for me, and this PR is limited to a one-line Rust change. The Python Lint test fails with "Exec format error", same as the other three failing tests, and that smells like a CI infrastructure issue to me. I am naively anxious about ignoring test failures when the change is at a pretty fundamental place and this is my first PR, but I'm also hard-pressed to see how at least most of these errors are interesting. And I'd of course much rather see them pass.

Could it be that the runners have an older glibc? I agree "Exec format error" usually means shenanigans but CI is passing for other branches so I at least have some signal that infrastructure is ok (and that failure is repeated after I reran)

(I am also suspicious, but) I don't yet think that should have impact. RE this change, and of course barring anything I'm not yet aware of: the Rust stdlib makes a decision at runtime whether to take the fast or slow path, based on whether the env glibc is sufficiently new. If it's too old, then simply the slow path is taken (as opposed to some failure path).

I think I understand one of the failures and will followup later today

I see your PR 23479; thank you!