It looks good to me, but it would be nice to have a way to check if it introduces type inference ambiguities that might cause a project using ndarray to fail compilation.
I think it would desirable, generally speaking, for us to have a cohort of projects to keep an eye on in terms of usage patterns on which to assess the impact of changes (similarly to what crater does for Rust, although on a much bigger scale).

I think it would desirable, generally speaking, for us to have a cohort of projects to keep an eye on in terms of usage patterns on which to assess the impact of changes (similarly to what crater does for Rust, although on a much bigger scale).

Yeah, that would be nice. It might be possible to do this using crates on crates.io that depend on the latest version of ndarray.

It looks good to me, but it would be nice to have a way to check if it introduces type inference ambiguities that might cause a project using ndarray to fail compilation.

I have a (non-public) project with ~1.3e4 lines of code that uses ndarray pretty heavily. I just ran cargo check with this PR. There are some cases that work in ndarray 0.12.1 where type inference is no longer sufficient with this PR.

This is one (simplified) example:

extern crate ndarray;

use ndarray::prelude::*;

fn main() {
    let n = 5;
    let a = Array2::<f64>::zeros((n, n));
    let b = a + Array2::eye(n);
    println!("{}", b);
}

(The compiler can no longer infer the element type for Array2::eye.)

Here's another similar example:

extern crate ndarray;

use ndarray::prelude::*;

fn main() {
    let n = 5;
    let a = Array1::<f64>::zeros(n);
    let b = Array2::zeros((n, n)) + a;
    println!("{}", b);
}

(The compiler can no longer infer the element type for Array2::zeros.)

I've added the "breaking-change" label. I don't see any way around this since Rust doesn't let us specify a default type for impls. The nice thing is that any breakage is easy to identify and fix. I really think allowing different element types is the right way to go.

If everything looks good, let's go ahead and merge this. (We've already started merged some breaking changes for the 0.13 release.)

I had somehow lost track of this conversation.
Shall we go ahead and merge it?