Float::abs_sub isn't working properly

[mdinger]

I'm not sure what Float::abs_sub() is supposed to do but it's not working correctly.

#![feature(std_misc)]

fn main() {
    use std::num::Float;

    // Rules:    
    // - If `self < other`: `0`
    // - Otherwise `self - other`

    assert_eq!(3.0f64.abs_sub(5.0), 0.0); // okay
    assert_eq!(3.0f64.abs_sub(-5.0), 8.0);// okay

    assert_eq!(3.0f64.abs_sub(2.0), 1.0); // okay
    assert_eq!(3.0f64.abs_sub(-2.0), 5.0);// okay

    assert_eq!(-3.0f64.abs_sub(2.0), -1.0); // Wrong. What should this do?
    assert_eq!(-3.0f64.abs_sub(-2.0), -5.0);// Wrong. What should this do?

    assert_eq!(-1.0f64.abs_sub(2.0), 0.0);  // okay
    assert_eq!(-1.0f64.abs_sub(-2.0), -3.0);// Wrong. What should this do?

    assert_eq!(0.0f64.abs_sub(1.0), 0.0);   // okay
    assert_eq!(0.0f64.abs_sub(-1.0), 1.0);  // okay
}

[huonw]

These call the C function fdim (and variants) directly, which is described by its man page as

double fdim(double x, double y);

These functions return the positive difference, max(x-y,0), between their arguments.

so this behaviour seems rather confusing, and may be a bug in libm??

[huonw]

Oh, no, it's just an operator precedence thing: -3.0f64.abs_sub(-2.0) is parsed as -(3.0f64.abs_sub(-2.0)), not (-3.0f64).abs_sub(-2.0) (which is presumably the intention?).

Adding the parens makes it behave as expected:

#![feature(std_misc)]

fn main() {
    use std::num::Float;

    // Rules:    
    // - If `self < other`: `0`
    // - Otherwise `self - other`

    assert_eq!((-3.0f64).abs_sub(2.0), 0.0);
    assert_eq!((-3.0f64).abs_sub(-2.0), 0.0);
    assert_eq!((-1.0f64).abs_sub(-2.0), 1.0);
}

Closing as not-a-bug, but thanks for checking things so carefully!

[mdinger]

Hmm...that's a downside to using method syntax here. I can see that tripping people up often but don't see a way around it except recommending save to variables before calling the methods.

@huonw thanks!