Trying to understand multidimensional arrays in D

[Profile Anaysis via Digitalmars-d-learn]

I'm a bit confused by how D does arrays.

I would like to create a array of matrices but I do not seem to 
get the correct behavior:

     int[][4][4] matrix_history;

What I would like is to have a 4x4 matrix and have a history of 
it. Just n 4x4 matrices but each matrix is a fixed size but there 
can be an arbitrary(dynamic) number.

I would like, for example,

matrix_history[0] to be the first 4x4 matrix,
matrix_history[1] to be the second 4x4 matrix,
...

and I would, in fact, like to be able to append a matrix like

matrix_history ~= some_4x4matrix.

I try to assign like

matrix_history[total-1] = new int[][](8,8);

or append

matrix_history ~= new int[][](4,4);

but the append fails with

Error: cannot append type int[][] to type int[][4][4]

which is confusing because the type per entry in the matrix 
history is of type int[][].

e.g., I could wrap the int[][] in a struct and then just have a 
singular array of these matrices and, to me, the logic should be 
the same. e.g.,

struct matrix
{
     int[4][4] data;
}

then

matrix[] matrix_history.

and

matrix_history ~= new matrix;

so, the logic should be the same between two. This method works 
but the more direct method doesn't seem to.

If I do

auto x = matrix_history[0];

x is not a int[4][4] but of type int[4](as reported by the 
debugger), which is very confusing.

it seems that the way D indexes multidimensional arrays is not 
logical nor consistent from my perspective.

auto x = matrix_history[0] returns an array of size 4.
auto x = matrix_history[0][0] returns an 2d array of size 4x4.
auto x = matrix_history[0][0][0] returns an int(as expected).


does this mean that have

     int[][4][4] matrix_history;

backwards?

     int[4][4][] matrix_history;

this creates even a more set of problems.

I guess I will have to revert to wrapping the matrix in a struct 
to get the natural extension of single arrays unless someone can 
clue me in on what is going on.