F64::midpoint

libs
1

Hi,

I was looking at the f64::midpoint implementation and was wondering why it does not use the paper “How do you compute the midpoint of an interval?“ (HAL) which, IMHO, behaves better.

Best,
Christophe

2

IIUC, the implementation that paper proposes is:

pub const fn midpoint(self, other: f64) -> f64 {
    let result = (self + other) / 2;
    if !result.is_finite() && self.is_finite() && other.is_finite() {
        (self / 2) + (other / 2)
    } else {
        result
    }
}
3

Other than the underflow checks in the std implementation (which seem suspicious, as noted in Tracking Issue for num_midpoint · Issue #110840 · rust-lang/rust · GitHub), I don't think it’s possible to get different results from the two?

2 Likes
4

Actually, I think those checks don't make a difference. I am pretty sure the implementations give identical results

Edit: Simplify `{f16, f32, f64, f128}::midpoint()` by Jules-Bertholet · Pull Request #146225 · rust-lang/rust · GitHub

2 Likes