Reputation: 353
In my understanding, converting vectors between Rcpp and C++ creates new vectors as follows. Is my understanding right?
When converting an Rcpp vector to a C++ vector, we use Rcpp::as<T>()
(e.g. Rcpp::as<std::string>
for Rcpp::CharacterVector
).
std::vector<std::string>
is created, and the original Rcpp elements are copied into the C++ vector as std::string
.
This means that modifying the newly created C++ vector elements does not affect the original Rcpp vector elements.
When converting a C++ vector to an Rcpp vector, we use Rcpp::wrap()
.
Rcpp vector with the corresponding type is created, and the C++ elements are copied into the Rcpp vector as Rcpp objects. This means that modifying the newly created Rcpp vector elements does not affect the original C++ vector elements.
Upvotes: 4
Views: 794
Reputation: 20746
Correct, the following functions perform conversions:
// conversion from R to C++
Rcpp::as<T>();
// conversion from C++ to R
Rcpp::wrap();
In short, moving data from an Rcpp object into a C++ vector and vice versa causes a copy to be made. This has been explained a bit here:
Declare a variable as a reference in Rcpp
For more on the templating, please see the Rcpp Extending vignette.
Details on cost from an Rcpp::*Vector
to std::vector<T>
can be found here:
Should I prefer Rcpp::NumericVector over std::vector?
One way to avoid the copy is to re-use memory if possible. This is hinted at here:
Deciding between NumericVector and arma::vec in Rcpp
Upvotes: 8