David Rush writes:

> Per Bothner writes:
>> The main problem with (2) is that it adds a new concept - an "index
>> object" data-type.
>
> Yep, but it actually adds value, I think. Sooner or later people are
> going to want slicing; having the index-object present helps pave
> the way for that.

It's easy to pick all kinds of parts of an array even with the current
mechanisms. Take the third slice from front of a 4*4*4 cube as a 4*4
matrix (assume zero based, depth first):

  (share-array cube (shape 0 4 0 4) (lambda (r k) (values 2 r k))).

Equally easy to pick second slice from top, or rightmost slice.

And I wrote a very general (every-nth arr d n) that shares every nth
slice of arr along dimension d. For the identity matrix i_4,

(every-nth i_4 0 3)   (every-nth i_4 1 3)
==>                   ==>
1 0 0 0               1 0
0 0 0 1               0 0
                      0 0
                      0 1

It's a dozen lines, including a non-destructive auxiliary for changing
the dth element of a list (the arguments to the affine map).
--
Jussi