`manual_is_power_of_two` is too broad for a default lint

[newpavlov]

Description

In my project I have a bunch of code like this:

use static_assertions as sa;

const MY_FLAG: u32 = const {
    // const computation
};

// Enforce that the flag mask contains only one bit
sa::const_assert_eq!(MY_FLAG.count_ones(), 1);

The new manual_is_power_of_two lint gets triggered by the assert. Changing it to is_power_of_two() as suggested by the lint would make the intent less clear.

Version

rustc 1.84.0-nightly (27861c4 2024-10-13)
binary: rustc
commit-hash: 27861c4
commit-date: 2024-10-13
host: x86_64-unknown-linux-gnu
release: 1.84.0-nightly
LLVM version: 19.1.1

Additional Labels

No response

[SpriteOvO]

@rustbot claim

The category pedantic might be more appropriate.

[daira]

The stated rationale applies for x.count_ones() == 1 but not for x & (x - 1) == 0. But note that the latter is not equivalent to x.count_ones() == 1 or x.is_power_of_two(): it gives a different result for x == 0, and the lint message doesn't point that out. So I agree with the outcome of moving it to pedantic.

Even better would be to split it into two lints that trigger separately on x.count_ones() == 1 and x & (x - 1) == 0. The former can be pedantic. The latter should have a message that says that the behaviour for x == 0 might be unintended, as well as pointing out the lack of clarity. Filed #14309 for this.