Dot product is only a valid call for ndarray::Array2

[JulianDicken]

Dot product is only a valid call for ndarray::Array2 -- even if the ArrayD is of the same dimensions.

use ndarray::prelude::*;
fn main() {
    let mat1 = Array2::from_shape_vec((3, 2), vec![3.0; 6]).unwrap();
    let mat2 = Array2::from_shape_vec((2, 3), vec![1.0; 6]).unwrap();
    let dot = mat1.dot(&mat2);
}

works, while

use ndarray::prelude::*;
fn main() {
    let mat1 = ArrayD::from_shape_vec(vec![3, 2], vec![3.0; 6]).unwrap();
    let mat2 = ArrayD::from_shape_vec(vec![2, 3], vec![1.0; 6]).unwrap();
    let dot = mat1.dot(&mat2);
}

does not
This results in a very opaque compiler error:

overflow evaluating the requirement `&ArrayBase<_, _>: Not`
consider increasing the recursion limit by adding a `#![recursion_limit = "256"]` attribute to your crate (`ndarray-issue`)
required for `&ArrayBase<_, _>` to implement `Not`
127 redundant requirements hidden
required for `&ArrayBase<OwnedRepr<{float}>, Dim<IxDynImpl>>` to implement `Not`

In my opinion this should either give a clearer error message or be a valid call. I am not experienced enough with rust at the moment to suggest a fix myself.

[JamboChen]

It seems the reason is that the Dot trait hasn’t yet been implemented for IxDyn.

I’m interested in contributing to this repository, but considering my limited experience with developing crates, I’m hesitant to jump in right away. Additionally, there seem to be various conventions for multiplying vectors/matrices of different shapes, so I believe this requires more discussion. Perhaps we could also look into how numpy.matmul handles these cases for reference.