Implement Index for &[T]

[kennytm]

As titled. Although slices support the a[b] operator, it does not implement Index<uint, T>, so one cannot pass in a slice in the following example:

fn do_something<V: Clone, T: Collection + Index<uint, V>>(t: &T) -> Vec<V> {
    let mut v = Vec::with_capacity(t.len());
    for i in range(0, t.len()) {
        v.push((*t)[i].clone());
    }
    v
}

fn main() {
    println!("{}", do_something(&vec![1u,2,3]));  // works
    println!("{}", do_something(&&[1u,2,3])); // does not work
}

[huonw]

I'm somewhat inclined to think we could just remove the built-in indexing from &[T] entirely and thus remove/reduce the need for the special lang-items, although there's at least three considerations:

• the failure message for out-of-bounds indexing would be less informative (no file/line information), but debugging any other fail! (e.g. Option.unwrap) already requires a debugger to get this info, so maybe it's not so bad.
• [T, .. n] can't yet implement Index properly, so this would still need built-in support with [] (and thus the out-of-bounds failure lang items can't be entirely removed just yet), unless implicit coercions will work.
• Improve method lookup auto-deref behavior (subissue of trait reform) #12825 needs to be fixed before IndexMut will work properly (and I guess DST would make it work more nicely).

Users would theoretically not see a change, since this would all happen inside std (and core).

[thestinger]

@huonw: It's not possible due to lack of CTFE.

[aturon]

cc me

[japaric]

Should I send a PR implementing Index/IndexMut for [T] now, and we can evaluate later removing the built-in indexing?

I already tested that PR, and the built-in indexing is preferred in the operation slice[index], but one go through the Index trait using slice.index(&index).

[aturon]

@japaric Yes, please go ahead and do so. Thanks!