Should arrays-of-arrays support promotion across nestings?

[bradcray]

While working on some array-of-array code recently, I found myself wondering which sorts of promotions we think ought to work in Chapel for arrays-of-arrays:

// Given:
var A: [1..3] [1..3] [1..3] real;

// This works because A and D have the same type/shape
var D: [1..3] [1..3] [1..3] real;
A = D;

// this works because C.type == A.eltType, so it's a normal promotion
var C: [1..3] [1..3] real;
A = C;


// this seems like perhaps it should work, but it doesn't.  We get "cannot iterate over values of type real(64)
A = 1.2;                        // error: cannot iterate over values of type real(64)

// this also seems like it could arguably work as a promotion over the second level of A, but it similarly doesn't
var B: [1..3] real;
A = B;                          // error: cannot iterate over values of type real(64)

Similarly, should one be able to call sin(A) or A + D and get an array-of-array-of-array-of-real result?

In essence, this issue asks "Should promotion be recursively considered in array-of-array situations, or restricted only to the outermost array, as it currently seems to be?"

[damianmoz]

I am assuming that the third case of A = C is just

A[1] = C, A[2] = C, A[3] = C

I am still trying to understand the precise difference between a multi-dimensional array and an array of arrays [of arrays]. Too much historical baggage in my head and a lack of understanding of what must work for the case of distributed memory.

[bradcray]

I am assuming that the third case of A = C is just…

That's right. And if the A = B; case was supported, you could imagine it being:

A[1][1] = B;  A[1][2] = B;  A[1][3] = B;
A[2][1] = B;  A[2][2] = B;  A[2][3] = B;
A[3][1] = B;  A[3][2] = B;  A[3][3] = B;

I am still trying to understand the precise difference between a multi-dimensional array and an array of arrays [of arrays].

Let us know if we can help clarify anything here (though ideally in a forum other than this issue).