cmp_nan only compares against NaN constant

[killercup]

IIUC, the cmp_nan lint just checks whether an expression is a constant whose last path segement is NAN.

That means that

0 == std::f32::NAN; //~ERROR doomed comparison with NAN

works as expected but more complicated (but still rather trivial) stuff like

let z = 42_f64;

let am_i_nan = std::f64::NAN;
z == am_i_nan; //~ERROR doomed comparison with NAN

let am_i_nan = std::f64::NEG_INFINITY / std::f64::INFINITY;
z == am_i_nan; //~ERROR doomed comparison with NAN

does not produce any of the expected errors.

Since the lint is only supposed to work with those standard number representations that have a NaN value (i.e. f32, f64, f128), the number of cases that evaluate to NaN is hopefully not that large and can probably be hard coded.

[killercup]

Oh, or am I misunderstanding the lint? Is this so people write …::is_nan() to be more explicit?

[mcarton]

Is this so people write …::is_nan() to be more explicit?

It's not to be more explicit, it's just that foo == NAN does not work, NaN equals nothing, not even itself.

[killercup]

NaN equals nothing, not even itself.

Doh! I totally knew that… I'll grab a non-alcoholic drink next.

[Lokathor]

So... should we close this?

[flip1995]

This lint could still be improved, by also evaluating consts: Playground

const MY_NAN: f64 = std::f64::NAN;

fn main() {
    let x = 3.0f64;
    
    if x == MY_NAN { // This currently doesn't get linted
        println!("YaY");
    } else if x == std::f64::NAN {
        println!("NaY")
    }
}