The rextendr package enables two new chunk types for knitr, extendr and extendrsrc. The former can be used to evaluate individual Rust statements while the latter can be used to define Rust functions and export them to R. To use either chunk type, you need to first load the rextendr package in a regular R chunk. You would normally do this in a hidden setup chunk at the top of your R Markdown document.

library(rextendr)

## Evaluating Rust statements

Chunks of type extendr evaluate a block of Rust statements. The value of the last statement can be returned to R and will be printed as chunk output. For example:

{extendr}
rprintln!("Hello from Rust!");

let x = 5;
let y = 7;
let z = x*y;

z


This chunk will look in the knitted document as follows:

rprintln!("Hello from Rust!");

let x = 5;
let y = 7;
let z = x*y;

z
## Hello from Rust!
## [1] 35

In the background, the extendr knit engine casts the Rust integer variable into type Result<Robj> and returns its value to R. Notice the lack of a semicolon (;) at the end of this line, to indicate that you want to return a value. You can also write code that doesn’t return any result to R. In this case, the last line would end in a semicolon, as in this example:

let x = 5;
let y = 7;
let z = x*y;

rprintln!("{}*{} = {}", x, y, z);
## 5*7 = 35

## Accessing R values from Rust

It is possible to access variables defined in R from Rust. Let’s define two R variables, a numeric and a string:

x <- 5.3
y <- "hello"

We can read these variables from Rust using the R!() macro. Unfortunately, it takes a bit of error checking and type-converting to get these into native Rust values. This is the case because things could go wrong at each step. We’re handling the errors with ?. For example, R!() calls the R interpreter and it could generate an error. And even if it returns a value, we’re not guaranteed that the value can be coerced into a real value or a string. However, as_real() and as_str() don’t return errors, they return options, so we need to turn the None option into an error via ok_or().

let x:f64 = R!(x)?
.as_real()
.ok_or("expected a real")?;

let y:String = R!(y)?
.as_str()
.ok_or("expected a string reference")?
.to_string();

rprintln!("The value of x: {}\nThe value of y: {}", x, y);
## The value of x: 5.3
## The value of y: hello

## Chaining chunks together

We may sometimes want to break a block of Rust code into separate Markdown chunks but treat them as one compile unit. We can do so by including the code from one chunk into the compile unit of another, by specifying a preamble chunk option. This option takes a character vector of chunk names, which get included into the compile unit in the order provided.

Consider this example Markdown code:

Define variable x:

{extendr chunk_x, eval = FALSE}
let x = 1;


Define variable y:

{extendr chunk_y, eval = FALSE}
let y = 2;


Print:

{extendr out, preamble = c("chunk_x", "chunk_y")}
rprintln!("x = {}, y = {}", x, y);


It produces the following output.

Define variable x:

let x = 1;

Define variable y:

let y = 2;

Print:

rprintln!("x = {}, y = {}", x, y);
## x = 1, y = 2

## Exporting Rust functions or objects to R

The chunk type extendrsrc compiles Rust functions and objects and registers them with R to be called later. For example, consider this code chunk. It creates R functions hello() and foo(), as well as a class called Counter.

{extendrsrc}
#[extendr]
fn hello() -> &'static str {
"Hello from Rust!"
}

#[extendr]
fn foo(a: &str, b: i64) {
rprintln!("Data sent to Rust: {}, {}", a, b);
}

struct Counter {
n: i32,
}

#[extendr]
impl Counter {
fn new() -> Self {
Self { n: 0 }
}

fn increment(&mut self) {
self.n += 1;
}

fn get(&self) -> i32 {
self.n
}
}


The only output the chunk itself creates is the Rust source code:

#[extendr]
fn hello() -> &'static str {
"Hello from Rust!"
}

#[extendr]
fn foo(a: &str, b: i64) {
rprintln!("Data sent to Rust: {}, {}", a, b);
}

struct Counter {
n: i32,
}

#[extendr]
impl Counter {
fn new() -> Self {
Self { n: 0 }
}

fn increment(&mut self) {
self.n += 1;
}

fn get(&self) -> i32 {
self.n
}
}

However, now we can call the functions hello() and foo() from R and use the Counter object:

out <- hello()
out
## [1] "Hello from Rust!"
foo(out, nchar(out))
## Data sent to Rust: Hello from Rust!, 16
x <- Counter$new() x$get()
## [1] 0
x$increment() x$increment()
x\$get()
## [1] 2

If your code requires external crates, you can set dependencies via the engine.opts chunk option, like so:

{extendrsrc engine.opts = list(dependencies = list(pulldown-cmark = "0.8"))}
use pulldown_cmark::{Parser, Options, html};

#[extendr]
fn md_to_html(input: &str) -> String {
let mut options = Options::empty();
options.insert(Options::ENABLE_TABLES);
let parser = Parser::new_ext(input, options);
let mut output = String::new();
html::push_html(&mut output, parser);
output
}


As before, the Rust code chunk just outputs the source:

use pulldown_cmark::{Parser, Options, html};

#[extendr]
fn md_to_html(input: &str) -> String {
let mut options = Options::empty();
options.insert(Options::ENABLE_TABLES);
let parser = Parser::new_ext(input, options);
let mut output = String::new();
html::push_html(&mut output, parser);
output
}

And the generated functions can be used from R:

md_text <- "# The story of the fox
The quick brown fox **jumps over** the lazy dog.
The quick *brown fox* jumps over the lazy dog."

md_to_html(md_text)
## [1] "<h1>The story of the fox</h1>\n<p>The quick brown fox <strong>jumps over</strong> the lazy dog.\nThe quick <em>brown fox</em> jumps over the lazy dog.</p>\n"