Error when trying to compile with sourceCpp in R

Question

I am a newbie in C++ and the package Rccp but I found some code in rcpp's gallery that lets you generate from a multivariate normal distribution. The code is

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

using namespace Rcpp;

// [[Rcpp::export]]
arma::mat mvrnormArma(int n, arma::vec mu, arma::mat sigma) {
   int ncols = sigma.n_cols;
   arma::mat Y = arma::randn(n, ncols);
   return arma::repmat(mu, 1, n).t() + Y * arma::chol(sigma);
}

Saving it in the file multivgaussian.cpp, when I try to compile it in R with

sourceCpp("multivgaussian.cpp")

I get this error message

Error in dyn.load("/tmp/RtmpGoFAwi/sourcecpp_6a751b6a3bee/sourceCpp_67198.so") : 
  unable to load shared object '/tmp/RtmpGoFAwi/sourcecpp_6a751b6a3bee/sourceCpp_67198.so':
  /tmp/RtmpGoFAwi/sourcecpp_6a751b6a3bee/sourceCpp_67198.so: undefined symbol: _ZN4Rcpp8internal14r_vector_startILi13EEEPNS_6traits12storage_typeIXT_EE4typeEP7SEXPREC

I also tried to see what will happen (in the terminal) if I try to compile it.

R CMD SHLIB multivgaussian.cpp
g++ -I/maths/R/lib64/R/include -DNDEBUG  -I/usr/local/include    -fpic  -g -O2  -c multivgaussian.cpp -o multivgaussian.o
multivgaussian.cpp:1:27: error: RcppArmadillo.h: No such file or directory
multivgaussian.cpp:4: error: ‘Rcpp’ is not a namespace-name
multivgaussian.cpp:4: error: expected namespace-name before ‘;’ token
multivgaussian.cpp:7: error: ‘arma’ has not been declared
multivgaussian.cpp:7: error: expected constructor, destructor, or type conversion before ‘mvrnormArma’
make: *** [multivgaussian.o] Error 1

It's probably something simple but I couldn't find anything online. Thanks a lot, Charis

Answer 1

Rcpp went through a release this week which requires its user packages to be rebuilt. Make sure your RcppArmadillo is rebuilt as well.

In your R CMD SHLIB example you are not telling R about the RcppArmadillo dependency so it cannot work. In the first example the line

// [[Rcpp::depends(RcppArmadillo)]]

takes care of that, but you still have your linker issue -- possibly due to version mismatch.

Answer 2

I recently run into the same error when compiling a simple Rcpp function that has several for loops in it. While the code was run successfully in Rstudio, it run into the same error of dyn.load and unable to load shared object when run on an lsf grid.

What I did was specifying a cache directory to compile the function using the cacheDir option in sourceCpp and cppFunction. Worked out very well for me too.