g++ Rcpp Undefined symbols for architecture x86_64 on osx - macos

Question

I am trying to create a dylib file in OSX El Capitain from a C++ source file to read into R via Rcpp. Here is a grossly simplified example of what I've created:
//test.cpp
#include <Rcpp.h>
RcppExport SEXP dosum(SEXP _a, SEXP _b){
double a = Rcpp::as<double>(_a);
double b = Rcpp::as<double>(_b);
return Rcpp::wrap(a+b);
}
When I compile/link using the following Makefile on a Linux server that is available to me at my university,
CC=g++
# note the use of c++14, used to use c++11
all: test
test :
${CC} -DNDEBUG \
-I. \
-I/usr/share/R/include \
-I/server/linux/lib/R/3.0/x86_64/site-library/Rcpp/include \
-I/usr/share/Rcpp_0.12.3/include \
-fpic -O3 -pipe \
-std=c++1y \
-c test.cpp
${CC} -shared -o test.so test.o
clean:
#find . \( -name "*.o" -o -name "*.so" \) -exec rm {} \;
I get the desired test.so file, which I can then read in during an R session via dyn.load('test.so'). Once it has been loaded, I can use the function dosum() via dosum(x,y). Great.
But I would like to do the exact same thing in OSX El Capitain on my personal Mac. I have tried many variations of a Makefile, and am currently using,
CC=g++
all: temp
temp:
${CC} \
-I. \
-I/usr/share/R/include \
-I/usr/local/include/Rcpp/Rcpp_0.12.5/inst/include \
-fPIC \
-c temp.cpp
${CC} -dynamiclib *.o -o temp.dylib
clean:
#find . \( -name "*.o" -o -name "*.dylib" \) -exec rm {} \;
When I run this Makefile, I get the following,
g++ \
-I. \
-I/usr/share/R/include \
-I/usr/local/include/Rcpp/Rcpp_0.12.5/inst/include \
-fPIC \
-c temp.cpp
g++ -dynamiclib *.o -o temp.dylib
Undefined symbols for architecture x86_64:
"_REprintf", referenced from:
Rcpp::Rstreambuf<false>::xsputn(char const*, long) in temp.o
Rcpp::Rstreambuf<false>::overflow(int) in temp.o
"_R_FlushConsole", referenced from:
Rcpp::Rstreambuf<true>::sync() in temp.o
Rcpp::Rstreambuf<false>::sync() in temp.o
"_R_GetCCallable", referenced from:
dataptr(SEXPREC*) in temp.o
"_R_NilValue", referenced from:
Rcpp::Rcpp_protect(SEXPREC*) in temp.o
Rcpp::Shield<SEXPREC*>::~Shield() in temp.o
"_Rf_allocVector", referenced from:
SEXPREC* Rcpp::internal::primitive_wrap__impl__cast<double>(double const&, Rcpp::traits::integral_constant<bool, false>) in temp.o
"_Rf_coerceVector", referenced from:
SEXPREC* Rcpp::internal::basic_cast<14>(SEXPREC*) in temp.o
"_Rf_length", referenced from:
double Rcpp::internal::primitive_as<double>(SEXPREC*) in temp.o
"_Rf_protect", referenced from:
Rcpp::Rcpp_protect(SEXPREC*) in temp.o
"_Rf_unprotect", referenced from:
Rcpp::Shield<SEXPREC*>::~Shield() in temp.o
"_Rprintf", referenced from:
Rcpp::Rstreambuf<true>::xsputn(char const*, long) in temp.o
Rcpp::Rstreambuf<true>::overflow(int) in temp.o
"_TYPEOF", referenced from:
SEXPREC* Rcpp::r_cast<14>(SEXPREC*) in temp.o
SEXPREC* Rcpp::internal::basic_cast<14>(SEXPREC*) in temp.o
ld: symbol(s) not found for architecture x86_64
clang: error: linker command failed with exit code 1 (use -v to see invocation)
make: *** [temp] Error 1
That is, it appears that things are getting garbled and an underscore is being added at the front of the variables in question. What am I doing wrong? I have tried adding/dropping various flags like -fpic, -std=XXX and so on.
I have found related questions, but they appear to be cases where the original poster is confusing how to use Rcpp in the first place. Here, I just want to do on the Mac what I can do in Linux. I have no reason to believe that there is a fundamental difference between the header files on my Mac versus those on the Linux box, but I have not yet checked in detail.
FWIW, I can create dylibs using the same approach on my Mac as long as they don't use Rcpp (for example, a simple cout << "hello world" <<endl; function), so maybe my Rcpp installation is somehow messed up.

Answer 1

Why are you using a Makefile?
It all works if you just let R do its business. Here is one-to-four-liner (depending on how you count):
R> cppFunction("SEXP dosum(SEXP a_, SEXP b_){
double a = Rcpp::as<double>(a_);
double b = Rcpp::as<double>(b_);
return Rcpp::wrap(a+b);
}")
R> dosum(4, 7)
[1] 11
R>
Note, however, that I had to revert your invalid identifier _a, _b to valid a_, b_.
All this can of course also be had as a one-liner given the templated conversions we get for free:
R> cppFunction("double dosum2(double a, double b) { return a+b; }")
R> dosum2(5, 8)
[1] 13
R>