Rcpp, making functions available to R from DLL

Question

New to C++, I would like to make functions compiled in a DLL available in R.

So far, I managed to do that for a basic function taking integer as input and returning the same, by following these steps in VisualStudio, then using dyn.load from R to load the created dll.

However, my C++ functions will need to handle R data.frame objects, and I am not sure how to make that possible. I saw from the Rcpp gallery that Rcpp might include some kind of translations between R and c++ data types including data.frame objects, but I don't know if I can generate a dll using Rcpp that I can then include in R using dyn.load.

From this answer by Dirk Eddelbuettel, it seems possible to generate a "dynamic library" using Rcpp, however, I could not find any dll when I tried generating a package with a source .cpp file using rcpp.package.skeleton(). The function I'd like to have a dll for is from the Rcpp gallery

#include <Rcpp.h>
using namespace Rcpp;
// [[Rcpp::export]]
DataFrame modifyDataFrame(DataFrame df) {
  // access the columns
  IntegerVector a = df["a"];
  CharacterVector b = df["b"];

  // make some changes
  a[2] = 42;
  b[1] = "foo";       
  // return a new data frame
  return DataFrame::create(_["a"]= a, _["b"]= b);
}

I tried to just paste that code into VisualStudio to try and generate that DLL, however I have the error "cannot find Rcpp.h" which I quite expected.

I then followed these steps in RStudio:

1. Create new project / Package
2. Include this cpp source file as a source file for this package
3. include Rcpp and enter Rcpp.package.skeleton("mypackage") so far, no DLL in the package folders
4. Tried to build the package in RStudio by going to Build/Install and Restart, but then I get an error message "Building R Packages needs installation of additional build tools, do you want to install them?" However I already have RbuildTools 3.4 installed, and when I click "YES" in RStudio nothing happens.

PS: Happy to hear about other methods but here the DLL format should be used if possible. Any piece of info is greatly appreciated since I have basically no idea of how Rcpp or C++ work

Answer 1

You need to figure out why your setup is hosed. This is meant to be easy and it is easy. Watch:

R> Rcpp::cppFunction('DataFrame modDF(DataFrame df) { IntegerVector a = df["a"]; CharacterVector b = df["b"]; a[2] = 42; b[1] = "foo"; return DataFrame::create(Named("a")=a, Named("b")=b); } ')                  
R> df <- data.frame(a=1:3, b=letters[24:26])                                                                                                                                                                       
R> df                                                                                                                                                                                                              
  a b                                                                                                                                                                                                              
1 1 x                                                                                                                                                                                                              
2 2 y                                                                                                                                                                                                              
3 3 z                                                                                                                                                                                                              
R> modDF(df)                                                                                                                                                                                                       
   a   b                                                                                                                                                                                                           
1  1   x                                                                                                                                                                                                           
2  2 foo                                                                                                                                                                                                           
3 42   z                                                                                                                                                                                                           
R>      

Now, I obviously don't recommend writing this way in a long one-liner. You are already on the right track setting up a package. But you need to sort out what is holding up your tools.

And as a PS the one-liner above with better indentation:

R> Rcpp::cppFunction('DataFrame modDF(DataFrame df) { \
      IntegerVector a = df["a"]; \
      CharacterVector b = df["b"]; \
      a[2] = 42;  \
      b[1] = "foo"; \ 
      return DataFrame::create(Named("a")=a, Named("b")=b); \
    } ')                  

Answer 2

The following seems to work:

1. from R, ran Rcpp.package.skeleton("dfcpp4", cpp_files="modifyDataFrame.cpp"). The second argument is required in order for the modifyDataFrame function to be available from the dll using dyn.load.
2. From the command line ran R CMD build dfcpp4
3. ran R CMD check dfcpp4 --no-manual from the command line.
4. The dll file in now present in the src-x64 folder

I am now able to call this function using

dyn.load("dfcpp4/src-x64/dfcpp4.dll")
df <- data.frame(a = c(1, 2, 3),
                 b = c("x", "y", "z"))
.Call("_dfcpp4_modifyDataFrame", df)
   a   b
1  1   x
2  2 foo
3 42   z

What I don't get is why in this case .Call should be used instead of .C...