Inconsistency with Array(T)#[]?(range : Range) when range is beginless or starts with 0.

[stephannv]

Bug Report

When using #[]? with a range, the behavior is different when using range.begin == 0 than when using range.begin != 0

Playground: link

array = [] of String

array[0..5]? # => []
array[..5]? # => []
array[1..5]? # => nil
array[-2..5]? # => nil
array[5..]? # => nil

The #[]?(range : Range) doc says:

Like #[](Range), but returns nil if range.begin is out of range.

But is 0 out of range? Yes.

array = [] of String
array[0] # => Unhandled exception: Index out of bounds (IndexError)

Maybe this is the cause: https://github.com/crystal-lang/crystal/blob/14bfa992e/src/array.cr#L661

System info:

Crystal 1.8.0 (2023-04-14)

LLVM: 15.0.7
Default target: aarch64-apple-darwin21.6.0

[stephannv]

I think that the correct way following the documentation would be always returning nil in these cases.

But I think would be nice if []? returns [] in these cases. eg.: master...stephannv:crystal:patch-1 . But I'm not sure if this will break the Crystal philosophy.

[straight-shoota]

This behaviour is not specific to the index 0. If you increase the size of the array as well as the accessor indices, you get the same result:

ary = [""]
ary[1..6]? # => [] of String
ary[2..6]? # => nil

If the start index is equivalent to the collection size, normalize_start_and_count accepts that and calculates the count of elements to be retrieved as 0.
I agree that this looks quite unexpected. I don't think think there's any reason for the first unset index to have a special meaning for this method.

[HertzDevil]

I think the intent here is to allow one to refer to the empty subrange of elements at the end of an array:

arr = ["a", "b", "c"]
arr[3, 0] = "d"
arr # => ["a", "b", "c", "d"]

This is why the one-past-the-end start index is allowed for start + count overloads, even though it cannot be used to refer to a single element.