Remove pull array index (unsafe), add uncons.

[sjoerdvisscher]

Fixes #471

I removed index, but it was used by the example in Data.Array.Polarized, so I added uncons which seems to be what the example actually needed, and it's clear that it's safe.

[b-mehta]

I agree uncons is safe and is what's actually needed in the example, but in the Nothing case loop should return an empty array, I'm pretty sure.

[sjoerdvisscher]

I agree uncons is safe and is what's actually needed in the example, but in the Nothing case loop should return an empty array, I'm pretty sure.

That's what it is doing. The error is a filler for a function that should never be called. I kept it from the original example, and my first reaction was the same as yours, so perhaps we should fix it. I guess adding an empty function would clarify things.

[aspiwack]

Thought: a natural generalisation of uncons is splitAt, with

splitAt :: Int -> Array a %1 -> Maybe (Array a, a, Array a)
splitAt i (Array f n)
  | n <= i = Nothing
  | otherwise = Just (fromFunction f i, f i, fromFunction (\x -> f (x+i+1)) (n-i-1)

With uncons is (up to isomorphism) splitAt 0.

Do you think it's worth exposing this generalisation?

[sjoerdvisscher]

When implementing uncons with splitAt 0, is there a way to discard the initial array in a total way? It should be empty, but there's no proof.

[aspiwack]

We don't appear to have made such a function available. But we should. Of course, it'd be partial and be an error if the length is non-0.

[aspiwack]

Maybe splitAt isn't a great name. At least we've used split for another similar function which doesn't have the singled-out element. (this means we can also define splitA, whatever we name it, as a composition of split and uncons).

---

Ok, so, now that I think about. There is a stupid way to disappear the null array: let !(a1, x, a2) in (x, append a1 a2). Is it a good idea? Probably not. But it's possible.

[sjoerdvisscher]

Good point about splitAt being the composition of split and uncons. It feels like that splitAt doesn't add much then?

[aspiwack]

Maybe not. Let's merge this PR at any rate.

You know what's funny though? splitAt = traverse uncons split may be the first time I see the Traversable (a,) instance show up in practice.