Add syntax for FullRange before 1.0

[bluss]

Why should FullRange have dedicated syntax?

Because the other range types have dedicated syntax and are usable as first-class values in custom APIs. Not having FullRange is inconsistent.

Motivation

Slicing syntax is neat, it gives us ranges and slices using one syntax device, ...

We can already use the Range, RangeTo, RangeFrom values to create custom interfaces:

trait AnyRange<Idx> {
    // Imagine all the Range types implement this trait
}
struct Mat {
    data: Vec<f32>
}

impl Mat {
    fn slice<R, S>(&self, xs: R, ys: S) -> Mat  where
         R: AnyRange<uint>, S: AnyRange<uint>
    {
         // Implement 2D slicing here.
         // Maybe the type is even CoW and returns a view transparently?
    }
}

With the slice syntax, we can implement things like:

let m = Mat::new(...)
let submatrix = m.slice(1.., 1..);

But FullRange is missing syntax! We can’t do this today:

let submatrix = m.slice(1.., ..);

Workarounds?

We can use 0.. instead of .., in almost all cases.

[pcwalton]

This doesn’t seem backwards incompatible to me.

[dgrunwald]

I was just looking at the implementation of .. parsing for #20811. I don’t think that this would cause any ambiguities, and it should by trivial to implement.

Of course, this would add yet another way to borrow the contents of a String. &*s, s.as_slice(), &s[], &s[..], in the future possibly &s with deref coercions.

Maybe this could replace the current &s[] syntax?

[bluss]

That’s what I thought dgrunwald.

pcwalton stressing breaking changes only is correct, but if we want to fix this with consistency, it is breaking because it wants &buf[] to become &buf[…]

[nikomatsakis]

There is no way we are going through another round of break-the-world changes just to avoid this relatively minor case of overlap between &foo[..] and &foo[]. Besides, I think that converting to a slice is sufficiently common that it warrants extra sugar.

[bluss]

That’s reasonable. I didn’t fully visualize the extent of changes required, and indeed this is a realization that is a few weeks to late.

[bluss]

The extra sugar argument falls now, with deref coercions! &foo is now the way to take a slice of a String or Vec.