rstest icon indicating copy to clipboard operation
rstest copied to clipboard

[question] Test with different specializations of a generic type?

Open Seanmatthews opened this issue 1 year ago • 4 comments

Is it possible to test with different specializations of a generic type? For example, how can I do something like this:

struct MyStruct<T, const U: usize, V: MyEnumType> { ... }

impl<...> MyStruct<...> {
    const u: usize = U;
    
    fn do_something() -> usize { U + 1 } 
}

#[rstest]
#[case(f64, 5, MyEnumType::ONE)]
fn test_do_something<#[case] T, #[case] const U: usize, #[case] V: MyEnumType>() {
    let a = MyStruct<T, U, V>{};
    assert!(U + 1 == a.do_something());
}

The documentation mentions generics, but I don't see any examples of dealing with generics.

Seanmatthews avatar Aug 23 '23 02:08 Seanmatthews

Yes you can but you you should just use values of different types. In your example you cannot build MyStruct without set all its inner values that constrains the generic types. I'll try to fill your example but ...

  1. MyEnumType should be a trait and not an enum... I remove it as generic type
  2. const generic argument are not supported yet
use rstest::rstest;

struct MyStruct<T, const U: usize> {
    a: [T; U],
}

impl<T, const U: usize> MyStruct<T, U> {
    const u: usize = U;

    fn do_something(&self) -> usize {
        U + 1
    }
}

#[rstest]
#[case(1_f64)]
fn test_do_something<T: Copy>(#[case] t: T) {
    let a = MyStruct { a: [t; 5] };
    assert!(6 == a.do_something());
}

If you can provide to me some more concrete examples I'll happy to help you. I can think to introduce something to define a const argument but I would like some examples were that is really necessary.

la10736 avatar Aug 24 '23 06:08 la10736

I have the following use case of specifying generics explicitly:

I have a trait that defines a constant epsilon value on the type level:

pub trait ConstEps {
    fn eps() -> f64;
}

then I have a function that does some calculation and uses a generic epsilon value for numerical precision, for example:

fn almost_equal<E: ConstEps>(left: f64, right: f64) -> bool {
    (right-left).abs() < E::eps()
}

I can have now many different epsilon types like:

struct Eps1eMinus7;
impl ConstEps for Eps1eMinus7 {
    fn eps() -> f64 {
        1e-7
    }
}

struct Eps1eMinus8;
impl ConstEps for Eps1eMinus8 {
    fn eps() -> f64 {
        1e-8
    }
}

struct Eps1eMinus9;
impl ConstEps for Eps1eMinus9 {
    fn eps() -> f64 {
        1e-9
    }
}
...

Now I want to test this function with various different epsilons (and left and right values). I don't know how I could do this at the moment without writing separate tests for each epsilon.

Any ideas? Note that the given example function almost_equal is of course just a simple example, my real use case has much more complex numerical calculations that use the generic epsilon value.

Edit: I found a solution with a slight hack, if there would be native support it could be much nicer I think:

use rstest::rstest;
use std::marker::PhantomData;

#[rstest]
#[case(PhantomData::<Eps1eMinus7>, 0.50000001f64, 0.5, true)]
#[case(PhantomData::<Eps1eMinus9>, 0.50000001f64, 0.5, false)]
fn test_almost_equal<E: ConstEps>(
    #[case] _e: PhantomData<E>,
    #[case] left: f64,
    #[case] right: f64,
    #[case] expect: bool,
) {
    let result = almost_equal::<E>(left, right);
    assert_eq!(result, expect);
}

benediktsatalia avatar Feb 06 '24 09:02 benediktsatalia

That's exactly what I were pasting right now :smile:

Yesterday I saw something in divan crate documentation that ring me a bell... So maybe I can introduce the following syntax:

#[rstest]
#[case(Eps1eMinus7, 0.50000001f64, 0.5, true)]
#[case(Eps1eMinus9, 0.50000001f64, 0.5, false)]
fn test_almost_equal<#[case] E: ConstEps>(
    #[case] left: f64,
    #[case] right: f64,
    #[case] expect: bool,
) {
    let result = almost_equal::<E>(left, right);
    assert_eq!(result, expect);
}

The sad news is that I've no time to implement it .... :cry:

la10736 avatar Feb 06 '24 11:02 la10736

Thank you for the quick answer! That would be nice to have this syntax, although I understand it is not high priority (the PhantomData version works well enough).

Btw. there is an interesting edge case if one uses _ as variable name the macro fails:

If I use

#[rstest]
#[case(PhantomData::<Eps1eMinus7>, 0.50000001f64, 0.5, true)]
#[case(PhantomData::<Eps1eMinus9>, 0.50000001f64, 0.5, false)]
fn test_almost_equal<E: ConstEps>(
    #[case] _: PhantomData<E>,
    #[case] left: f64,
    #[case] right: f64,
    #[case] expect: bool,
) {
    let result = almost_equal::<E>(left, right);
    assert_eq!(result, expect);
}

compilation fails with:

error: Wrong case signature: should match the given parameters list.
   --> src/investigate_rstest.rs:51:16
    |
 51 |         #[case(PhantomData::<Eps1eMinus7>, 0.50000001f64, 0.5, true)]
    |                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

error: Wrong case signature: should match the given parameters list.
   --> src/investigate_rstest.rs:52:16
    |
 52 |         #[case(PhantomData::<Eps1eMinus9>, 0.50000001f64, 0.5, false)]
    |                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

benediktsatalia avatar Feb 06 '24 11:02 benediktsatalia