One of the limitations of the D programming language is that it doesn’t have the same selection of libraries for data analysis that you have in a language like R. One way to solve that problem would be to port the many thousands of R libraries to D. Doing that would take millions of programmer hours, and new R libraries appear every day. A more realistic solution, which I’ve taken here, is to facilitate communication between D and R. This can take two forms:

Calling D functions from R. The main program is written in R, but bottlenecks and anything for which D is better are written in D, compiled into a shared library, and loaded into R. This is the procedure commonly used to call C, C++, and Fortran code from R. Other R users can call the D functions you’ve written even if they don’t know anything about D.

Calling R functions from D. You can use the excellent RInside project to embed an R interpreter inside your D program. Data is passed efficiently because everything in R is a C struct (SEXPREC). You can allocate these structs from D code and pass pointers to them between the languages. This approach is of particular interest to a current R user wanting to move to D without giving up any of their existing R code and libraries.

Is This Project Active?

This project is largely complete. Lack of recent activity is a sign that the project is stable. I realize that there’s a trend to say projects are dead if no new features have been added in the last 30 days. If you see that the last activity was two years ago, that means it’s been working so well that I haven’t had a reason to make any changes in two years. You shouldn’t expect to see much activity in an interface between two mature languages; regularly adding features would be a sign that something is wrong.

As of this update (January 2020) I am unaware of anything that doesn’t work. If something doesn’t work, file an issue.


Documentation was produced using adrdox. You can view it here.


File an issue to ask a question if you can’t get it to work. I have successfully used this library on Linux, Mac, and Windows, but I probably can’t help much on Mac or Windows, since I don’t have access to development machines running either of those operating systems.

Linux Installation

If you only want to call D functions from R, installation is easy.

1. Install R and the dmd compiler (obvious, I know, but I just want to be sure). I recommend updating to the latest version of R. 2. Install the embedr package using devtools:


If you have a standard installation (i.e., as long as you haven’t done something strange to cause to be hidden in a place the system can’t find it) installation is done.

Windows Installation

Update (January 2020): I’ve decided to officially abandon Windows support. The main reason for this is the fact that Microsoft’s WSL is so convenient to use, that you should be using it if at all possible. See this article on using VS Code with WSL. I’ve used that approach and it works well. There’s no meaningful difference in the editing/compiling/running steps relative to doing that in Windows natively, except that setup is more complicated.

WSL is available only for Windows 10, but Windows 10 has been out for five years, Windows 7 is no longer supported, and Windows 8 is not heavily used. I’m more than happy to accept pull requests if someone wants to take over Windows support.

Docker also works. You can find an example here.

Mac Installation

Installation on Mac is similar to Linux, but as I don’t have access to a Mac, it’s hard for me to add that functionality to embedr.

Docker works well.

Please contact me if you are a Mac user and would like to take over embedr’s Mac support. File an issue if you have questions about getting it to work.

Embedding R Inside D

I show first how to compile and run from within R. I also provide an example dub.sdl file below if that’s your preference.

Currently I can only give documentation for Linux because that is all I have used. If you use Windows or Mac and are interested in adding documentation, please let me know.

Linux Installation

Embedding R inside D requires you to install a slightly modified version of the RInside package in addition to everything above.

1. Install R and the dmd compiler (obvious, I know, but I just want to be sure). I recommend updating to the latest version of R.
2. Install RInsideC using devtools. In R:


3. Install the embedr package using devtools:


If you have a standard installation (i.e., as long as you haven’t done something strange to cause to be hidden in a place the system can’t find it) you are done.

Example: Calling D Functions From R

Note that Windows requires an explicit export attribute when defining a function that is to be included as part of a shared library. I present both versions of the simple example to clarify that, but for the other examples you will have to add the export attribute. There are some other differences in the Linux and Windows versions. These are due to the fact that those functions were created at different times, and I have not yet had time to make things consistent. I will eventually get around to doing that.

Simple Example (Linux)

Save this code in a file called librtest.d:

extern(C) {
  Robj add(Robj rx, Robj ry) {
    double result = rx.scalar + ry.scalar;
    return result.robj;

Then in R, from the same directory as librtest.d, create and load the shared library using the dmd function:

compileFile("librtest.d", "rtest")

Test it out:

.Call("add", 2.5, 3.65)

Example: Calling R Functions From D

Let’s start with an example that tells R print “Hello, World!” to the screen. Put the following code in a file named hello.d:

import embedr.r;

void main() {
    evalRQ(`print("Hello, World!")`);

In the directory containing hello.d, run the following in R:


This tells dmd to compile your file, handling includes and linking for you, and then run it for you. You should see “Hello, World!” printed somewhere. The other examples are the same: save the code in a .d file, then call the dmd function to compile and run it.


I do not normally use Dub. However, many folks do, and if you want to add dependencies on other libraries like Mir, you don’t have much choice but to use Dub.

Put hello.d in a subdirectory called src. In your project’s root directory, i.e., the parent of src, put a file called dub.sdl with the following information:

name "myproject"
description "embedr hello world"
authors "Lance Bachmeier"
copyright "Copyright 2020, Lance Bachmeier"
license "GPLv2"
versions "standalone"
targetType "executable"
lflags "/usr/lib/" "/usr/local/lib/R/site-library/RInsideC/lib/"

The lflags paths may be different on your machine. The first argument points to, which depends on where R is installed. The second argument depends on where the RInsideC package is installed. To find these, I use this in Bash:

locate -l 1

and this in R:

paste0(find.package("RInsideC")[1], "/lib/")

Alternatively, open R in your project directory and run the following:


and it will create a dub.sdl file including the correct paths, create a src/ directory if it doesn’t exist, and add r.d to the src/ directory if it’s not already there.

Drop your source files into the src/ subdirectory, open the terminal in the project root directory, and compile/run with

dub run

Pulling R Data Into D

If you are embedding R inside a D program, and you want to pull data from R into D, please read this first.

More Examples

The examples above were too basic to be practical. Here are some examples that demonstrate more useful functionality.

Passing a Matrix From D to R

Let’s write a program that tells R to allocate a (2x2) matrix, fills the elements in D, and prints it out in both D and R.

import embedr.r;

void main() {
    auto m = RMatrix(2,2);
    m[0,0] = 1.5;
    m[0,1] = 2.5;
    m[1,0] = 3.5;
    m[1,1] = 4.5;
    m.print("Matrix allocated by R, but filled in D");

RMatrix is a struct that holds a pointer to an R object plus the dimensions of the matrix. When the constructor is called with two integer arguments, it has R allocate a matrix with those dimensions.

The library includes some basic functionality for working with m, including getting and setting elements, and printing. Alternatively, we could have passed m to R and told R to print it:

import embedr.r;

void main() {
    auto m = RMatrix(2,2);
    m[0,0] = 1.5;
    m[0,1] = 2.5;
    m[1,0] = 3.5;
    m[1,1] = 4.5;
  m.toR("mm"); // Now there is an object inside R called mm

Passing a Matrix From R to D

We can also pass a matrix in the opposite direction. Let’s allocate and fill a matrix in R and then work with it in D.

import embedr.r;

void main() {
  // Generate a (20x5) random matrix in R
  evalRQ(`m <- matrix(rnorm(100), ncol=5)`);
  // Create an RMatrix struct in D that holds a pointer to m
  auto dm = RMatrix("m");
  dm.print("This is a matrix that was created in R");
  // Change one element and verify that it has changed in R
  dm[0,0] = dm[0,0]*4.5;

A comment on the last line: printR uses the R API printing function to print an R object. If you pass a string as the argument to printR, it will print the object with that name in R. It will not print the string that you pass to it as an argument. D does not know anything about m. It only knows about dm, which holds a pointer to m.


A vector can be represented as a matrix with one column. In R, vectors and matrices are entirely different objects. That doesn’t matter much in D because vectors are represented as matrices in D. I have added an RVector struct to allow the use of foreach. Here is an example:

import embedr.r;
import std.stdio;

void main() {
  // Have R allocate a vector with 5 elements and copy the elements of the double[] into it
    auto v = RVector([1.1, 2.2, 3.3, 4.4, 5.5]);
  // Pass v to R, creating variable rv inside the R interpreter
  // Use foreach to print the elements
    foreach(val; v) {

RVector Slices

You can slice an RVector, as shown in this example.

import embedr.r;
import std.stdio;

void main() {
    evalRQ(`v <- rnorm(15)`);
    auto rv = RVector("v");
    foreach(val; rv) {
    rv[1..5].print("This is a slice of the vector");

Working With R Lists

Lists are very important in R, as they are the most common way to construct a heterogeneous vector. Although you could work directly with an R list (there’s an RList struct to do that) you lose most of the nice features if you do. For that reason I created the NamedList struct. You can refer to elements by number (elements are ordered as in any array) or by name. You can add elements by name (but only that way, because every element in a NamedList needs a name).

import embedr.r;
import std.stdio;

void main() {
  // Create a list in R
  evalRQ(`rl <- list(a=rnorm(15), b=matrix(rnorm(100), ncol=4))`);
  // Create a NamedList struct to work with it in D
  auto dl = NamedList("rl");
  // Pull out a matrix
  auto dlm = RMatrix(dl["b"]);
  dlm.print("This is the matrix from that list");
  // Pull out a vector
  auto dlv = RVector(dl["a"]);
  dlv.print("This is the vector from that list");

  // Create a NamedList in D and put some R objects into it
  // NamedList holds pointers to R objects, which can be pulled
  // out using .data
  NamedList nl;
  nl["a"] =;
  nl["b"] =;
  // Send to R as rl2
  // Can verify that the elements are reversed

Scalars and Strings

R does not have a scalar type. What appears to be a scalar is a vector with one element. On the D side, however, there are scalars, so you have to specify that you are working with a scalar your D code. On a different note, we can pass strings between R and D.

import embedr.r;
import std.stdio;

void main() {
  // Create some "scalars" in R
  evalRQ(`a <- 4L`);
  evalRQ(`b <- 4.5`);
  evalRQ(`d <- "hello world"`);
  // Print the values of those R variables from D
  // Pull the integer a from R into D
  // Also pulls in an integer, but creates a long rather than int
  // The default type of scalar is double, so it is not necessary to specify the type in that case
  // Pull the string d from R into D
  // Can also work with a string[]
  ["foo", "bar", "baz"].toR("dstring");
  // Create a vector of strings in R and pull it into D as a string[]
  evalRQ(`rstring <- c("under", "the", "bridge")`);

There is more functionality available (the entire R API, in fact) but the single goal of this library is to facilitate the passing of commonly-used data types between the two languages. Other libraries are available for the functions in the R standalone math library, optimization, and so on.


What’s the difference between functions dmd and compileFile?
Why use R for compilation rather than Dub?
Can I use LDC?