Is it okay to modify a mapped matrix in RcppEigen?

Question

I'm writing some code in RcppEigen, and I came across this ominous warning in the documentation:

One must, of course, be careful not to modify the contents of the R object in the C++ code. A recommended practice is always to declare mapped objects as const.

This refers to "mapped" matrices, i.e. matrices in Eigen that use the same memory as the corresponding R object.

Is this warning just about good functional programming practice, or are there other things that can go wrong if I use mapped objects to do modify-in-place?

---

This is a simplified version of what I'm doing in Rcpp:

#include <RcppEigen.h>
// [[Rcpp::depends(RcppEigen)]]

// [[Rcpp::export]]
void modify(Eigen::Map<Eigen::MatrixXd> X) {

    X(0,0) = 0;
    return;
}

Which does things like this in R:

X <- matrix(1:4, 2) + 0
X
#      [,1] [,2]
# [1,]    1    3
# [2,]    2    4
modify(X)
X
#      [,1] [,2]
# [1,]    0    3
# [2,]    2    4

For the particular problem I'm working on I'm getting a 5x speed increase by modifying my matrix in place, so I'm prepared for that small sacrifice in terms of readability and maintainability. As far as I can tell it's working as intended, but I'm worried that I'm inviting some sort of subtle bug some time in the future.

Answer 1

It is perfectly ok to modify provided you remain aware of the side-effect of the variable in R.

In a purely functional world, you would want no side-effects: inputs are processed and not altered, a result is returned.

Here we have a more hybrid approach. Objects are passed down from R as SEXP, and the P stands for pointer---so changes persist. That confuses some people, and we sometimes have to explain here how to avoid it. :)

But in short you are safe, particularly if just set elements and do not do crazy stuff like altering dimensions etc pp.