When to use Rcpp instead of standard C/C++ code in R packages?
Rcpp is an R package for interfacing C++ code with R. My rule of thumb is: use the simple .C interface if I can. Otherwise use .Call and there are essentially two options
- Learn the standard C functions defined by <Rinternals.h>, or
- Learn the Rcpp package, which implicitly uses Rinternals.h functions.
In either case you have to learn something, so it is your choice. Rcpp provides a user-friendly layer of abstraction for converting R data types to C++ data types. So Rcpp should be good if you don’t care exactly what the code is doing. But if you want more full control then you should learn Rinternals.h (after all, that is how they wrote Rcpp).
For my clusterpath project, I wrote the clusterpath package that uses Rinternals.h, and the clusterpathRcpp package that uses Rcpp.
clusterpath/src/interface.cpp has #include <Rinternals.h> and uses
standard functions such as PROTECT and SET_VECTOR_ELT.
clusterpathRcpp/src/interface.cpp has #include <Rcpp.h> and uses
Rcpp types such as NumericMatrix, IntegerVector, and DataFrame.