Using a generic in a struct and implementing via a trait

Question

I have struct defined like this

struct Shape<T> {
    line_one: T,
    line_two: T,
}

I am trying to create a simple trait and implementation that takes this struct to calculate some simple math.

My trait and impls look like this

trait Calculate<T: Mul<Output = T>> {
    fn calc(&self, Shape: T) -> T;
}

impl<T> Calculate<T> for Shape<T> {
    fn calc(&self, Shape: T) -> T {
        self.line_one * 2 + self.line_two * 2
    }
}

impl Calculate<i32> {
    fn calc(&self) -> i32 {
        self.line_one * 2 + self.line_two * 2
    }
}

fn calc_fn<T: Calculate<i32>>(calculate: T) {
    calculate.calc();
}

When I put this into the Rust playground, the compile fails as Mul is not implemented in the impl Calculate<T>. However, if I change <T> to <T: Mul<Output = T>>, I get the error that

<anon>:14:21: 14:22 error: expected one of `!`, `(`, `+`, `,`, `::`, `<`, or `>`, found `:`
<anon>:14 impl <T> Calculate<T: Mul<Output = T>> for Shape<T> 

I'm at a loss as how to implement Mul for T in impl Calculate<T>.

What I'm trying to achieve is code that I can send either a f32 or i32 in without needing to create two different impl definitions so can just pass in T.

Answer 1

This syntax is incorrect:

impl<T> Calculate<T: Mul<Output = T>> for Shape<T>

you want

impl<T: Mul<Output = T>> Calculate<T> for Shape<T>

I'd almost always recommend using where clause instead; I think it reads better and might have helped prevent this case:

impl<T> Calculate<T> for Shape<T>
    where T: Mul<Output = T>,

This just unlocks more errors. You are trying to multiply by 2, an integral variable that hasn't been fully nailed down yet — is it a u8? is it an i32? The restrictions don't say that T can be multiplied by a 2. Additionally, you try to add values together, but there's no guarantee that you can add the type T.

The impl Calculate<i32> and calc_fn blocks don't really make sense; it's unlikely you want to implement functions for the trait and you don't supply a Shape in the latter. The trait also doesn't use the Shape: T parameter (and Rust uses snake_case variable names anyway).

What I'm trying to achieve is code that I can send either a f32 or i32 in without needing to create two different impl definitions so can just pass in T.

In this case, it'd probably be way easier:

trait Calculate<T> {
    fn calc(&self) -> T;
}

impl Calculate<i32> for Shape<i32> {
    fn calc(&self) -> i32 {
        self.line_one * 2 + self.line_two * 2
    }
}

impl Calculate<f32> for Shape<f32> {
    fn calc(&self) -> f32 {
        self.line_one * 2.0 + self.line_two * 2.0
    }
}

Note that these are not equivalent! The f64 has floating point literals (2.0) not integral (2).

If you must have a generic, you are going to need to convert the literal 2 to the type T or add a restriction that T can be multiplied by some known size. Unfortunately, f64 cannot by multiplied by any integral value without casting.

The FromPrimitive trait from the num crate is key here. Making the types Copy also makes the implementation easier.

extern crate num;

use std::ops::{Add, Mul};
use num::FromPrimitive;

struct Shape<T> {
    line_one: T,
    line_two: T,
}

trait Calculate<T> {
    fn calc(&self) -> T;
}

impl<T> Calculate<T> for Shape<T>
    where T: Copy + FromPrimitive + Add<Output = T> + Mul<Output = T>,
{
    fn calc(&self) -> T {
        let two = T::from_u8(2).expect("Unable to create a value of two");
        self.line_one * two + self.line_two * two
    }
}

fn main() {
    let s1 = Shape { line_one: 2, line_two: 3 };
    let s2 = Shape { line_one: 2.0, line_two: 3.0 };

    println!("{}", s1.calc());
    println!("{}", s2.calc());
}