Implement RangeBounds for &Range*

[qsantos]

I have encountered a situation very similar to a Stack Overflow question. Essentially, I want to write a method that accepts either an owned or borrowed range as an argument, to avoid users from having to write f(&(1..32)). Naturally, we can use Borrow to do this:

fn f<T: Debug, U: Debug + RangeBounds<T>, V: Borrow<U>>(range: V) {
    let range = range.borrow();
    println!("{range:?}");
}

However, the compiler cannot infer the type of U since it could be anything, not just the dereferenced argument. Thus, when calling it with a borrowed range, we need to provide the type:

    let range = 1..2;
    f::<_, Range<u32>, _>(&range);

If &Range<T>, and others were to implement RangeBound, the problem would become moot, since one could just write:

fn f<T: Debug, U: Debug + RangeBounds<T>>(range: U) {
    println!("{range:?}");
}

Are there any gotchas that would make this a bad idea?

[jdahlstrom]

I'd like to have this for another reason (and was slightly surprised it doesn't exist): if you want to reuse a range (eg. to index two slices in lockstep^[1]) you (infamously) have to either resort to cloning or use two separate range literals:

fn foo(a: &[i32], b: &[i32]) {
    let r = compute_start()..compute_end();
    bar(a[r.clone()]);
    bar(b[r]);
    // Or
    let (start, end) = (compute_start(), compute_end());
    bar(a[start..end)]);
    bar(b[start..end]);
}

---

1. My real-world use case: the color and depth buffers in a 3D rasterizer ↩︎

[benschulz]

@qsantos

However, the compiler cannot infer the type of U since it could be anything, not just the dereferenced argument. Thus, when calling it with a borrowed range, we need to provide the type:

You can restrict the options that inference can consider:

Rust Playground

[qsantos]

Oh, that's clever, that would work in my case. Thanks!