regression: literal out of range

[Mark-Simulacrum]

https://crater-reports.s3.amazonaws.com/beta-1.86-3/beta-2025-02-28/reg/checked_expr-0.3.0/log.txt

cc @oli-obk since I think you were refactoring something here recently?

[INFO] [stdout] error: literal out of range for `i8`
[INFO] [stdout]   --> src/lib.rs:25:29
[INFO] [stdout]    |
[INFO] [stdout] 10 | assert_eq!(checked_expr!(-(-128_i8)), None);
[INFO] [stdout]    |                             ^^^^^^
[INFO] [stdout]    |
[INFO] [stdout]    = note: the literal `128_i8` does not fit into the type `i8` whose range is `-128..=127`
[INFO] [stdout]    = help: consider using the type `u8` instead
[INFO] [stdout]    = note: `#[deny(overflowing_literals)]` on by default

[ehuss]

This bisects to #136393

What's interesting is that the expanded code:

(|| Some((128_i8).checked_neg()?.checked_neg()?))();

fires the lint on stable, but the proc-macro does not.

[oli-obk]

yea I was just reading the proc macro source and was wondering how this ever worked. Turns out 128_i8 overflows into -128_i8 💀

[oli-obk]

So I guess I made another accidental bugfix with these refactorings, yay

[apiraino]

WG-prioritization assigning priority (Zulip discussion).

@rustbot label -I-prioritize +P-high

[wesleywiser]

Discussed briefly in the compiler team triage meeting. It looks like this is an expected fix to lint behavior.

Should we mention this in release notes for 1.86?