integer_decode() for float: example usage in docs doesn't make sense

[AdamSlayer]

Location

float.rs line 1867, in this "example usage" code:

use num_traits::Float;

let num = 2.0f32;

// (8388608, -22, 1)
let (mantissa, exponent, sign) = Float::integer_decode(num);
let sign_f = sign as f32;
let mantissa_f = mantissa as f32;
let exponent_f = num.powf(exponent as f32);

// 1 * 8388608 * 2^(-22) == 2
let abs_difference = (sign_f * mantissa_f * exponent_f - num).abs();

assert!(abs_difference < 1e-10);

Summary

This code only works when num = 2.0. With num having almost any other value, the assert! at the end fails. For example, with let num = 2.5f32;, instead of let num = 2.0f32;, the abs_difference is 2.4815533. I think the problem is not in the function itself, but in the example code provided in the documentation.

[jhorstmann]

Not too familiar with the implementation, but I guess num.powf(exponent as f32) should be 2_f32.powf(exponent as f32), regardless of the input number. Choosing 2.0 also as the input for this example is indeed confusing.

[mtilda]

I believe this issue was posted in the wrong repository, as this relates to the crate num-traits. I recommend opening a new issue at github.com/rust-num/num-traits.

The example referenced in this issue can be found at num_traits::float::Float::integer_decode.

The source code is located at src/float.rs#L1863-L1881.