Rcpp broken on Opensuse 15.2

[sboehringer]

When trying to implement a Rcpp module to expose a C++ class I get an error about I get the following error:

  /usr/include/c++/9/bits/stl_vector.h:509:30:   required from ‘std::vector<_Tp, _Alloc>::vector(std::vector<_Tp, _Alloc>::size_type, const allocator_type&) [with _Tp = double; _Alloc = std::allocator<double>; std::vector<_Tp, _Alloc>::size_type = long unsigned int; std::vector<_Tp, _Alloc>::allocator_type = std::allocator<double>]’/home/pingu/lib64/R-4.0.3/library/Rcpp/include/Rcpp/sugar/functions/sample.h:153:28:   required from here/usr/include/c++/9/bits/stl_uninitialized.h:597:18: error: ‘value’ is not a member of ‘std::is_copy_assignable<double>’       const bool __assignable = is_copy_assignable<_ValueType>::value;                  ^~~~~~~~~~~~make: *** [/home/pingu/lib64/R-4.0.3/etc/Makeconf:181: file388d7a0874cb.o] Error 1

It turns out that this error message is independent on the particular module code and also happens with all module examples as provided with the package (this is the minimal example):

library('inline')
library('Rcpp')
unifModCode = '
using namespaceRcpp;
/// create external pointer to a Uniform object
RcppExport SEXP Uniform__new(SEXPmin_,SEXPmax_) {
// convert inputs to appropriate C++ types
double min = as<double>(min_),max = as<double>(max_);
// create pointer to an Uniform object and
// wrap it as an external pointer
Rcpp::XPtr<Uniform>ptr(newUniform( min, max ),true);
// return the external pointer to the R side
return ptr;}
/// invoke the draw method
RcppExport SEXP Uniform__draw(SEXP xp, SEXPn_) {
// grab the object as a XPtr (smart pointer)
// to UniformRcpp::XPtr<Uniform> ptr(xp);
// convert the parameter to int
int n = as<int>(n_);
// invoke the function
NumericVector res = ptr->draw( n );
// return the result to Rreturnres;}
}';

    f1<-cxxfunction( ,"",includes =unifModCode,plugin ="Rcpp");
    ## will display info about'f1
    getDynLib(f1)

My sessionInfo is:

R version 4.0.3 (2020-10-10)
Platform: x86_64-pc-linux-gnu (64-bit)
Running under: openSUSE Leap 15.2

Matrix products: default
BLAS:   /home/user/lib64/R-4.0.3/lib/libRblas.so
LAPACK: /home/user/lib64/R-4.0.3/lib/libRlapack.so

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
 [3] LC_TIME=de_DE.UTF-8        LC_COLLATE=en_US.UTF-8    
 [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
 [9] LC_ADDRESS=C               LC_TELEPHONE=C            
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] Rcpp_1.0.6    inline_0.3.17

loaded via a namespace (and not attached):
[1] compiler_4.0.3   tools_4.0.3      codetools_0.2-16

The compiler version is:

g++ --version
g++ (SUSE Linux) 7.5.0
Copyright (C) 2017 Free Software Foundation, Inc.

The following complier command was generated:

g++ -std=gnu++11 -I"/home/pingu/lib64/R-4.0.3/include" -DNDEBUG  -I"/home/pingu/lib64/R-4.0.3/library/Rcpp/include" -I/usr/local/include   -L"/home/pingu/Documents/src/mmap/build" -lmmap_rcpp -I/home/pingu/lib64/R-4.0.3/library/Rcpp/include -I/usr/include/c++/9 -fpic  -g -O2  -c file388d7a0874cb.cpp -o file388d7a0874cb.o

Thank you very much in advance. Any help appreciated.

[eddelbuettel]

Can you may try to embed your Rcpp Modules in a package as we generally recommend?

I may well be forgetting something but did we ever suggest to try it with package inline? Did you follow a particular piece of doucmentation or tutorial here?

There are existing packages out there using Modules, both on CRAN where they get tested regularly as well as in the "wild" and I just updated one locally for my use. Plus we do have use of Rcpp Modules in our unit tests.

[eddelbuettel]

Sorry I got confused by your lingo. You used "module" but you didn't mean Rcpp Modules.

May I suggest you look at the Rcpp Attributes vignette along with the Rcpp Introduction vignette? Creating a simple function is much easier. (Dealing with external pointers is possible too and we have other examples.) It may help if you first clarified what you really wanted to do.

[eddelbuettel]

And there are various copy and paste errors here missing whitespace, plus the class Uniform you are templating on is not defining. I am not able to repair this for you, sorry.

Might be best to just close it.

[eddelbuettel]

Also, for general use questions, maybe post on rcpp-devel? If your code tries to extend the Uniform example in Modules, I generally found it easiest to keep other extensions in the same file.

[sboehringer]

Thank you very much for the quick reply. I have done Rcpp modules before and have followed the same recipe in my current effort. On seeing the non-obvious error I went back to the examples from the vignette Rcpp-modules.pdf. The example above was copy-pasted from the document. I have also checked other examples with a class definition from the same pdf all resulting in the same error message.

At the moment, incompatible compiler flags/compiler bug/STL incompatibility seem to be the most likely culprits. I have tried to fiddle around with Sys.setenv to change the compiler command/compiler flags without success. Is there some documentation to which environment variables cxxfunction is sensitive to? Or would such changes require to implement a plugin?

I would appreciate it this bug report can stay open for a while to see whether others see this problem too. If I cannot find a solution here, I will turn to the mailing list.

[eddelbuettel]

I just went on a quick walk to get some groceries and mulled over this. I can probably help by injecting the Uniform class.

But you still haven't explained what your issue is. I doubt it is a general inability to compile under OpenSuSE. Detlef Steuer over at HSU has been looking at R and OpenSuSE for almost as long as I have for Debian and things tend to just work. This will be something "simple". We'll get to it.

But I need a _minimally complete and verifiable_ example which your first post wasn't.

[Enchufa2]

@sboehringer I see:

[...] -I/usr/include/c++/9` [...]

so standard library from GCC 9, but

g++ (SUSE Linux) 7.5.0

so GCC 7 from the official repos. If you compiled your custom R version with a custom GCC version, use the same compiler for the packages.

[eddelbuettel]

So here is a cleaned up (lots of whitespace got lost in your version) and simplified (using Rcpp Attributes gets us automatic conversion of types such as C++ scalars making the interface simpler). I also added the Uniform class.

If you save the file as, say, /tmp/rcpp1140.cpp the a simple Rcpp::sourceCpp("/tmp/rcpp1140.cpp") will

• read the file
• compile and link and load it
• export the two tagged functions to R
• run the illustration at the bottom to
  
  
  
  • set the RNG at a seed
  
  
  • instantiate an Uniform drawer object
  
  
  • call it for 5 draws
  
  

which on my machine now gets me

> Rcpp::sourceCpp("/tmp/rcpp1140.cpp")

> set.seed(123) # reproducible draws

> xp <- getUniform(10,20)

> vec <- drawUniform(xp, 5)

> vec
[1] 12.8758 17.8831 14.0898 18.8302 19.4047
> 

Code is below. It should work on any machine with proper compile and headers.

#include <Rcpp.h>

using namespace Rcpp;

class Uniform {
public:
    Uniform(double min_, double max_) : 
       min(min_), max(max_) {}

    NumericVector draw(int n) {
        RNGScope scope;
        return runif(n, min, max);
    }

private:
    double min, max;
};

// create external pointer to a Uniform object
// [[Rcpp::export]]
XPtr<Uniform> getUniform(double min, double max) {
  // create pointer to an Uniform object and
  // wrap it as an external pointer
  Rcpp::XPtr<Uniform> ptr(new Uniform( min, max ), true);
  // return the external pointer to the R side
  return ptr;
}

// invoke the draw method
// [[Rcpp::export]]
NumericVector drawUniform(XPtr<Uniform> xp, int n) {
  // grab the object as a XPtr (smart pointer)
  // to UniformRcpp::XPtr<Uniform> ptr(xp);
  // convert the parameter to int
  // invoke the function
  NumericVector res = xp->draw( n );
  // return the result to R
  return res;
}

/*** R
set.seed(123) # reproducible draws
xp <- getUniform(10,20)
vec <- drawUniform(xp, 5)
vec
*/

[sboehringer]

My aplogies, the examples work after all. I somehow mixed up the error output with those from my own code.

I have now narrowed down the problem to a c++ standard issue. As cxxfunction outputs the compilation commands, I could switch to the temporary directory to tweak the commands. Strangely, I do get different errors even for the same command. Still, by using --std c++17 and clang++ as the compiler, I could generate the shared library manually. I still haven't managed to load the Module from this shared library using the 'Module' function, however, I believe that this should be possible. Using dyn.load on the shared library works as a first step.

The remaining question is whether I can make cxxfunction to use a different compiler and change the c++ standard. I found the //[[Rcpp::plugins(cpp17)]] directive. However, this does not seem to have any effect.

@eddelbuettel: thanks for your help and patience so far.

[Enchufa2]

My suggestion is to use the ~/.R/Makevars file to define CXX and posibly other variables and flags.

[eddelbuettel]

There are a few StackOverflow posts as well as mailing list posts I made over the last decade+ which show the ~/.R/Makevars use. With a little creative googling you should find it. I think it's also in the Rcpp FAQ (or should be ...).

All good then so I close this.