multidimensional colon operator?

I want to do something like that below: copy an N-dimensional array into a N+1-dimensional array at index t in the last dimension. Can this be done efficiently and cleaner than umpteen if-tests or nasty permutes?

 function A = insert_x_into_A(A,x,t)
if ndims(A) == 2
    A(:,t) = x;
elseif ndims(A) == 3
    A(:,:,t) = x;
elseif ndims(A) == 4
    A(:,:,:,t) = x;
...
end

I think that something like the pseudo-code below would be neat

 function A = insert_x_into_A(A,x,t)
A(colon(ndims(x)),t) = x;

Answer 1

[Jan]

Two approaches: 1. Reshape the matrix at first:

function A = insert_x_into_A(A,x,t)
siz = size(A);
A = reshape(A, [], siz(end));
A(:, t) = x(:);
A = reshape(A, siz);

2. Use a cell as index vector: EDITED: Cleaned up with Bruno's suggestion.

function A = insert_x_into_A(A,x,t)
idx(1:ndims(A) - 1) = {':'};
A(idx{:},t) = x;

[Kenneth Eaton]

I vote for the second one. It's often overlooked that you can index a matrix with the character ':'.

[Bruno Luong]

Just simplify few lines of Jan's code:

function A = insert_x_into_A(A,x,t)

idx(1:ndims(A)-1) = {':'};

A(idx{:},t) = x;

[Jan]

@Bruno: Nicer. I'll insert it in my approach.

[Knut]

Is this documented anywhere? Am I supposed to be able to do stuff like:

ord = 3;
buffN = zeros(5*ones(1,ord));
shiftidx1(1:ord) = {'2:end'};
shiftidx2(1:ord) = {'1:(end-1)'};
buffN(shiftidx1{:}) = buffN(shiftidx2{:});

(circshift seems to be made for this purpose, but it seems slow for my purpose)

Answer 2

[Matt Fig]

If you want to get real evil:

function A = insert_x_into_A(A,x,t)
col = repmat(':,',1,ndims(A)-1);
eval(['A(',col,'t) = x;'])

Though I don't know why you would need such a specific function.

[Jan]

Matt, I can see your EVAL!

[Matt Fig]

I couldn't resist. This seems like a safe application anyway.