Applying manual_clamp lint results in slower code

[Shnatsel]

Description

The manual_clamp lint nags me to replace .max(0).min(255) with .clamp(0,255) in image-webp crate, but doing that results in significantly slower code.

You can find the full details and the assembly comparison in the issue I opened on the rustc repo: rust-lang/rust#141915

The root cause of the slowdown is still being investigated.

We've put an #[allow] on the affected piece of code in image-web, but it would be nice to disable this lint by default so it wouldn't cause any further performance regressions across the ecosystem.

Version

I've reproduced the regression on rustc 1.75, 1.82 and 1.87

Additional Labels

No response

[samueltardieu]

I suggest we wait a bit until rust-lang/rust#141915 concludes that this cannot be fixed before allowing the lint by default. Allowing it would not solve the problem retroactively, and it might be fixed soon in the compiler now that the issue is known.

@rustbot label +S-blocked

[Shnatsel]

Turns out the issue has been known for a year already: rust-lang/rust#125738
Despite that, a fix has not materialized.

The same argument was made on the Clippy side a year ago when this issue was initially discovered: #12826

Given that it's been a year already and there is no fix in sight, but Clippy has been actively making people's code slower through that entire year. Sadly disabling the lint by default will not undo the damage already done, but perhaps it's time to at least stop slowing down new code?