Dimension swap from cube to matrix

Start by reproducing the R examples with RcppArmadillo, especially the arma::cube::row() case where one remaining dimension is 1. Compare the returned dimensions for Rarray1 and Rarray2, then inspect the cube-to-matrix conversion used by arma::mat. Done means row extraction preserves the intended dimensions without regressing the column and slice examples.

I encountered a bug in RcppArmadillo when saving a row of a cube to a matrix. If one of the remaining dimensions is just of dimension 1, then dimensions are swapped. I'll illustrate the bug with an example

cppFunction("arma::mat test_row(arma::cube Rarray) {
arma::mat test = Rarray.row(0);
return test;
}", depends="RcppArmadillo")
Rarray1 <- array(NA, dim=c(100,1,2))
dim(test_row(Rarray1))
> 1 2
Rarray2 <- array(NA, dim=c(100,2,1))
dim(test_row(Rarray2))
> 1 2

When pulling out the first row and saving it to a matrix, dimensions are not kept in the second example, but swapped. I'll tested also whether this happens also when pulling out the first column or slice, but there everything seems fine. I'll just attach the code

cppFunction("arma::mat test_col(arma::cube Rarray) {
arma::mat test = Rarray.col(0);
return test;
}", depends="RcppArmadillo")
Rarray3 <- array(NA, dim=c(1,100,2))
dim(test_col(Rarray3))
> 1 2
Rarray4 <- array(NA, dim=c(2,100,1))
dim(test_col(Rarray4))
> 2 1

cppFunction("arma::mat test_slice(arma::cube Rarray) {
arma::mat test = Rarray.slice(0);
return test;
}", depends="RcppArmadillo")

Rarray5 <- array(NA, dim=c(1,2,100))
dim(test_slice(Rarray5))
> 1 2
Rarray6 <- array(NA, dim=c(2,1,100))
dim(test_slice(Rarray6))
> 2 1