eq_op false positive with floats

[lukebitts]

Consider the following function:

fn nan_test() -> bool {
	let v = std::f32::NAN;

	v != v //warning: equal expressions as operands to `!=`
}

The eq_op warning is wrong here since floats only implement PartialOrd and doing v != v allows me to detect them in a generic context:

fn max<T: PartialOrd>(v1: T, v2: T) -> T {
	if v2 != v2 { v2 }
	else if v2 > v1 { v2 } else { v1 }
}

Maybe a solution is to check whether the result of the expression returns something that implements Ord or not?

[llogiq]

I'd rather specialize reporting for floats and suggest v.is_nan() instead.

[lukebitts]

I can't use is_nan in a generic context. I could bind the parameter to num::Float though.

[llogiq]

True, but that's a fairly niche use case. Maybe it's best to simply #[allow(eq_op)] in your max function?

[lukebitts]

Actually num::Float and is_nan turned out to be a better abstraction for my project. So I guess the warning was right

[BusyJay]

#6167 makes eq_op lint on assert families, but when users choose to implement PartialEq for a custom type, they may use assert_eq to test if it's correctly implemented, which leads to the lint error.