Mir Slice.shape is not consistent with the actual array shape

[9il]

On Sunday, 24 May 2020 at 14:17:33 UTC, Pavel Shkadzko wrote:
> I am confused by the return value of Mir shape.
> Consider the following example.
>
> ///////////////////////////////////////////////////////////////////////////
> import std.stdio;
> import std.conv;
> import std.array: array;
> import std.range: chunks;
> import mir.ndslice;
>
> int[] getShape(T : int)(T obj, int[] dims = null)
> {
>     return dims;
> }
>
> // return arr shape
> int[] getShape(T)(T obj, int[] dims = null)
> {
>     dims ~= obj.length.to!int;
>     return getShape!(typeof(obj[0]))(obj[0], dims);
> }
>
> void main() {
>     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
> 15, 16];
>     int[][][] a = arr.chunks(4).array.chunks(2).array;
>
>     writeln(arr);
>     writeln(arr.shape);
>
>     auto arrSlice = arr.sliced;
>     writeln(arrSlice);
>     writeln(arrSlice.shape);
>
> }
> ///////////////////////////////////////////////////////////////////////////
>
> [[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
> 16]]]
> [2, 2, 4] <-- correct shape
> [[[1, 2, 3, 4], [5, 6, 7, 8]], [[9, 10, 11, 12], [13, 14, 15, 
> 16]]]
> [2] <-- which shape is that?
>
> I would expect sliced to create a Slice with the same dims. 
> Well, sliced returns a shell over the array, but why does it 
> return its own shape instead of the shape of the array it 
> provides view into? This makes it even more confusing once you 
> print both representations.
> What's the rationale here?

BTW, the code example above doesn't compiles.

OT:
Instead of

>     int[] arr = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 
> 15, 16];
>     int[][][] a = arr.chunks(4).array.chunks(2).array;

you can generate the same common D array using Mir:

     auto a = [2, 2, 4].iota!int(1).ndarray;