February 1, 2023

Rustler - Using Rust crates in Elixir

Bartlomiej Dudzik

You couldn't find an Elixir package? Let's see if Rust can help with that.

The Rust and Elixir logos on a gray backround picture

anchorAre you on a Rust journey?

We can help with mentoring, training, Team reinforcement, and custom development.
Talk to us!

Rust and Elixir work great combined! Rust and Elixir can help you increase performance significantly. Discord shared some details on their blog (and also wrote about why they switched from Go to Rust).

In this blog post, I will show you how easily you can build and use small programs inside Elixir using Rustler.

I recently needed a function to create and edit PDF files with Elixir for a small private project. I searched for some packages and the most notable one for Elixir is elixir-pdf, but it only offers to create PDFs without additional manipulation. That was fine by me since I wanted an excuse to do some more things in Rust anyway, and I found lopdf.

In the beginning, I had written my program as a regular project in Rust i.e. cargo new my_pdf and then it was happy coding 🙂. I wanted the program to take some configuration and create/edit a PDF file based on it.

After I played around with it, I started wondering “How much work would it be to make it an Elixir NIF?”.

Turns out - there’s not a whole lot of additional stuff we need to do in order to comfortably use crates inside Elixir projects.

Rustler is a library for writing Erlang NIFs in a very easy and straightforward way.

There are about 2 things we need to keep in mind and take care of:

  • Error handling: there needs to be something that maps Rust errors into Elixir atoms so that we could easily handle them.
  • Rustler traits: typically your functions and structs will need to derive the implementation from appropriate traits so they can be usable inside Elixir (only the public parts of it of course).

anchorSetup

Setting up a project with Rustler is fairly easy. You can find a link to instructions here: https://github.com/rusterlium/rustler#getting-started

I’ll assume you have both Elixir and Rust installed in your development environment.

  1. Create a new Elixir project mix new rustler_pdf

  2. Add the :rustler dependency in mix.exs

     # mix.exs
  defp deps do
    [
      {:rustler, "~> 0.26.0"}
    ]
  end

  3. Download packages: mix deps.get

  4. Setup Rustler

    1. Run mix rustler.new
    2. On “Module name” prompt type in RustlerPdf (Name of your Elixir module that Rustler registers NIFs to)
    3. On the “Library name” prompt type in rustlerpdf (Name of your cargo crate)

  5. Configure Rustler

    1. Add the Rustler behaviour in rustler_pdf.ex
    defmodule RustlerPdf do
  use Rustler,
      otp_app: :rustler_pdf, # must match the name of the project in `mix.exs`
      crate: :rustlerpdf # must match the name of the crate in `native/rustlerpdf/Cargo.toml`

  def add(_arg1, _arg2), do: :erlang.nif_error(:nif_not_loaded)
end
    1. Make sure Rust’s boilerplate is this:
    // native/src/lib.rs

#[rustler::nif]
fn add(a: i64, b: i64) -> i64 {
    a + b
}

rustler::init!("Elixir.RustlerPdf", [add]);

  6. Test that the setup works

    1. Run iex -S mix
    2. Execute RustlerPdf.add(1,2)
    3. See if it successfully outputs 3

anchorImplementing the PDF program

I’ll mostly showcase contracts and interfaces of the Rust part of the program. If you're interested in the full implementation, you can find it here.

First off let’s take a look at pure Rust structs and functions.

// Enum that determinates value formatting
pub enum FieldType {
    Money,
    Text,
    Slotted,
}

// Struct describing text
pub struct PdfWriterOperation {
    page_number: i32,
    font: (String, i32),
    dimensions: (f64, f64),
    field_type: FieldType,
    value: Option<String>,
}

// Struct which is the program's input
pub struct PdfWriterConfiguration {
    input_file_path: Option<String>,
    output_file_path: String,
    operations: Vec<PdfWriterOperation>,
}

pub fn read_config() -> PdfWriterConfiguration {
    PdfWriterConfiguration {
        input_file_path: None,
        output_file_path: "PIT-8C-modified.pdf".to_string(),
        operations: vec![
            PdfWriterOperation {
                page_number: 0,
                font: ("F1".to_string(), 10),
                dimensions: (462.82, 55.92),
                value: Some("120.99".to_string()),
                field_type: FieldType::Money,
            },
            PdfWriterOperation {
                page_number: 0,
                font: ("F1".to_string(), 10),
                dimensions: (43.32, 347.81),
                value: Some("41.0".to_string()),
                field_type: FieldType::Money,
            },
        ],
    }
}

pub fn create_pdf(config: PdfWriterConfiguration) -> Result<(), std::io::Error> {}

This is the overview of pretty much the entirety of the Rust implementation “minus” the lopdf interaction. In general, the idea is that given PdfWriterConfiguration a create_pdf function will create a PDF file. Then I want to use those methods directly in Elixir.

The only relevant part of the implementation here are the structs themselves, as you can see they have i32, f64, string, tuple, struct and enum Rust types. Later we’ll see how they map to Elixir data structures.

anchorRustler-ize-it

Now we need to add traits to our data and functions so Rustler knows how to marshall the data between Rust and Elixir environments.

// import Rustler traits
use rustler::{NifStruct, NifUnitEnum};

#[derive(NifUnitEnum)]
pub enum FieldType {
    Money,
    Text,
    Slotted,
}

#[derive(NifStruct)]
#[module = "RustlerPdf.PdfWriterOperation"]
pub struct PdfWriterOperation {
    page_number: i32,
    font: (String, i32),
    dimensions: (f64, f64),
    field_type: FieldType,
    value: Option<String>,
}

#[derive(NifStruct)]
#[module = "RustlerPdf.PdfWriterConfiguration"]
pub struct PdfWriterConfiguration {
    input_file_path: Option<String>,
    output_file_path: String,
    operations: Vec<PdfWriterOperation>,
}

The NifStruct NifUnitEnum and module = "Elixir.ModuleName attributes provide implementations and metadata for the languages to communicate.

anchorEnum variant with value - NifTaggedEnum

An additional note here: I’m using NifUnitEnum which is a simple Enum variant. If you’d like to use a Rust Enum variant that also carries data, you might use NifTaggedEnum in order to use such Enums:

use rustler::NifTaggedEnum;

#[derive(NifTaggedEnum)]
pub enum FieldType {
    Money(String),
    Text(String),
    Slotted(String),
}`

Then in Elixir they are represented as a tuple like this:

{:slotted, "some value"}

anchorBack to Rustlerizing - Functions

#[rustler::nif]
pub fn read_config() -> PdfWriterConfiguration {
  priv_read_config()
}

fn priv_read_config() -> PdfWriterConfiguration {
  PdfWriterConfiguration {
        input_file_path: None,
        output_file_path: "PIT-8C-modified.pdf".to_string(),
        operations: vec![
            PdfWriterOperation {
                page_number: 0,
                font: ("F1".to_string(), 10),
                dimensions: (462.82, 55.92),
                value: Some("120.99".to_string()),
                field_type: FieldType::Money,
            },
            PdfWriterOperation {
                page_number: 0,
                font: ("F1".to_string(), 10),
                dimensions: (43.32, 347.81),
                value: Some("41.0".to_string()),
                field_type: FieldType::Money,
            },
        ],
    }
}

The read_config function is simple enough that it doesn’t require any additional treatment. Since it’s only delivering hardcoded data and is not dealing with any kind of IO, it just needs the #[rustler::nif] attribute.

I’m also splitting the functions into the plain Rust functions and the public ones that Rustler exposes to Elixir. This most definitely should be built as a regular Rust crate and another Rustler ‘bridge’ module. I decided to just add prefixes here - for the purpose of this post, I believe this simplifies things a bit.

use rustler::{Atom, Env, Error as RustlerError, NifStruct, NifUnitEnum, Term};
use std::io::Error as IoError;
use std::io::ErrorKind as IoErrorKind;

// Creates an atoms module using the rustler macro
mod atoms {
    rustler::atoms! {
        ok,
        error,
        eof,

        // Posix
        enoent, // File does not exist
        eacces, // Permission denied
        epipe, // Broken pipe
        eexist, // File exists

        unknown // Other error
    }
}

// Translates std library errors into Rustler atoms
fn io_error_to_term(err: &IoError) -> Atom {
    match err.kind() {
        IoErrorKind::NotFound => atoms::enoent(),
        IoErrorKind::PermissionDenied => atoms::eacces(),
        IoErrorKind::BrokenPipe => atoms::epipe(),
        IoErrorKind::AlreadyExists => atoms::eexist(),
        _ => atoms::unknown(),
    }
}

#[rustler::nif]
pub fn create_pdf(env: Env, config: PdfWriterConfiguration) -> Result<Term, RustlerError> {
    match priv_create_pdf(config) {
        Ok(()) => Ok(atoms::ok().to_term(env)),
        Err(ref error) => return Err(RustlerError::Term(Box::new(io_error_to_term(error)))),
    }
}

fn priv_create_pdf(config: PdfWriterConfiguration) -> Result<(), std::io::Error> {}

create_pdf is slightly more complicated as it involves some error handling.

priv_create_pdf is dealing with creating an actual file on the filesystem so it might result in an error.

In order to handle that situation, we’ve defined a module called atoms which uses the rustler::atoms! macro to create methods to simplify creating and decoding atoms.

io_error_to_term takes an std::io::error, matches on the kind of error and translates that into an Atom.

As you’ve probably noticed, create_pdf expects an env argument which is always present with all NIFs. Env is used by Rustler for communication and encoding/decoding data between the BEAM and Rust.

read_config and add in the previous examples are not using it so it’s possible to omit its declaration.

anchorExporting NIFs

rustler::init!("Elixir.RustlerPdf", [read_config, create_pdf]);

At the end of the file, change the rustler::init! macro to the above: this lets Rustler know to bind read_config function to the Elixir.RustlerPdf module.

anchorThe Elixir part

Once the Rust crate exports the functions, we need to let the Elixir side know what functions are expected to be bound to the module. Note that knowing the functions' arities is important here, otherwise you’ll get errors saying that a function couldn’t be loaded.

defmodule RustlerPdf do
  use Rustler,
    otp_app: :rustler_pdf,
    crate: :rustlerpdf

  def read_config(), do: :erlang.nif_error(:nif_not_loaded)
  def create_pdf(_pdf_writer_configuration), do: :erlang.nif_error(:nif_not_loaded)
end

Now we can use the NIFs like regular Elixir functions. Let’s try it out in the interactive console iex -S mix

[nix-shell:~/Projects/elixir-pdf-experiment/rustler_pdf]$ iex -S mix

iex(1)> RustlerPdf.read_config()
%{
  __struct__: RustlerPdf.PdfWriterConfiguration,
  input_file_path: nil,
  operations: [
    %{
      __struct__: RustlerPdf.PdfWriterOperation,
      dimensions: {462.82, 55.92},
      field_type: :money,
      font: {"F1", 10},
      page_number: 0,
      value: "120.99"
    },
    %{
      __struct__: RustlerPdf.PdfWriterOperation,
      dimensions: {43.32, 347.81},
      field_type: :money,
      font: {"F1", 10},
      page_number: 0,
      value: "41.0"
    }
  ],
  output_file_path: "PIT-8C-modified.pdf"
}
iex(2)>

Notice how the Rust struct now maps to a plain Elixir map type with additional __struct__ metadata: this is how Rustler ensures type safety. If the __struct__ would be missing, then we’d get an ArgumentError exception.

Out of curiosity I was also benchmarking this implementation against an Elixir library called elixir-pdf. The Elixir library was roughly 2x slower, but interestingly used 1/3 less memory.

def benchmark() do
    Benchee.run(
      %{
        "create_pdf" => fn -> RustlerPdf.create_pdf(RustlerPdf.read_config()) end,
        "e_create_pdf" => fn -> RustlerPdf.e_create_pdf() end
      },
      time: 10,
      memory_time: 1
    )
  end
Operating System: macOS
CPU Information: Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
Number of Available Cores: 12
Available memory: 16 GB
Elixir 1.14.2
Erlang 23.3.4.17

Benchmark suite executing with the following configuration:
warmup: 2 s
time: 10 s
memory time: 1 s
reduction time: 0 ns
parallel: 1
inputs: none specified
Estimated total run time: 26 s

Benchmarking create_pdf ...
Benchmarking e_create_pdf ...

Name                   ips        average  deviation         median         99th %
create_pdf          2.02 K        0.49 ms    ±56.09%        0.48 ms        0.70 ms
e_create_pdf        0.96 K        1.04 ms    ±42.20%        1.03 ms        1.22 ms

Comparison:
create_pdf          2.02 K
e_create_pdf        0.96 K - 2.11x slower +0.55 ms

Memory usage statistics:

Name            Memory usage
create_pdf           1.65 KB
e_create_pdf         1.17 KB - 0.71x memory usage -0.47656 KB

anchorConclusion

I've shown how you can make use of Rustler in order to add a functionality to your Elixir program which natively might not exist. I hope I also managed to make you consider using Rust for your next high-peformant and type-safe Elixir module.

In my opinion, Rust and Elixir are a great match. Rust offers amazing processing performance while Elixir and the BEAM are excellent at low latency connections and message passing. With Rustler you don't need to choose between one or another, but you could use both instead :) So don't hesitate and experiment, it's both fun and productive for the developers and could be very beneficial for your bussiness.

Consider taking a brief look at the full implementation of the Rust program here: https://github.com/BobrImperator/rustler_pdf

Further reading:

anchorAre you on a Rust journey?

We can help with mentoring, training, Team reinforcement, and custom development.
Talk to us!

Grow your business with us

Our experts are ready to guide you through your next big move. Let us know how we can help.
Get in touch