Introduction

Welcome! This book will teach you about the Rust Programming Language. Rust is a systems programming language focused on three goals: safety, speed, and concurrency. It maintains these goals without having a garbage collector, making it a useful language for a number of use cases other languages aren’t good at: embedding in other languages, programs with specific space and time requirements, and writing low-level code, like device drivers and operating systems. It improves on current languages targeting this space by having a number of compile-time safety checks that produce no runtime overhead, while eliminating all data races. Rust also aims to achieve ‘zero-cost abstractions’ even though some of these abstractions feel like those of a high-level language. Even then, Rust still allows precise control like a low-level language would.

“The Rust Programming Language” is split into eight sections. This introduction is the first. After this:

After reading this introduction, you’ll want to dive into either ‘Learn Rust’ or ‘Syntax and Semantics’, depending on your preference: ‘Learn Rust’ if you want to dive in with a project, or ‘Syntax and Semantics’ if you prefer to start small, and learn a single concept thoroughly before moving onto the next. Copious cross-linking connects these parts together.

Contributing

The source files from which this book is generated can be found on Github: github.com/rust-lang/rust/tree/master/src/doc/trpl

A brief introduction to Rust

Is Rust a language you might be interested in? Let’s examine a few small code samples to show off a few of its strengths.

The main concept that makes Rust unique is called ‘ownership’. Consider this small example:

fn main() {
      let mut x = vec!["Hello", "world"];
  }

This program makes a variable binding named x. The value of this binding is a Vec<T>, a ‘vector’, that we create through a macro defined in the standard library. This macro is called vec, and we invoke macros with a !. This follows a general principle of Rust: make things explicit. Macros can do significantly more complicated things than function calls, and so they’re visually distinct. The ! also helps with parsing, making tooling easier to write, which is also important.

We used mut to make x mutable: bindings are immutable by default in Rust. We’ll be mutating this vector later in the example.

It’s also worth noting that we didn’t need a type annotation here: while Rust is statically typed, we didn’t need to explicitly annotate the type. Rust has type inference to balance out the power of static typing with the verbosity of annotating types.

Rust prefers stack allocation to heap allocation: x is placed directly on the stack. However, the Vec<T> type allocates space for the elements of the vector on the heap. If you’re not familiar with this distinction, you can ignore it for now, or check out ‘The Stack and the Heap’. As a systems programming language, Rust gives you the ability to control how your memory is allocated, but when we’re getting started, it’s less of a big deal.

Earlier, we mentioned that ‘ownership’ is the key new concept in Rust. In Rust parlance, x is said to ‘own’ the vector. This means that when x goes out of scope, the vector’s memory will be de-allocated. This is done deterministically by the Rust compiler, rather than through a mechanism such as a garbage collector. In other words, in Rust, you don’t call functions like malloc and free yourself: the compiler statically determines when you need to allocate or deallocate memory, and inserts those calls itself. To err is to be human, but compilers never forget.

Let’s add another line to our example:

fn main() {
      let mut x = vec!["Hello", "world"];
  
      let y = &x[0];
  }

We’ve introduced another binding, y. In this case, y is a ‘reference’ to the first element of the vector. Rust’s references are similar to pointers in other languages, but with additional compile-time safety checks. References interact with the ownership system by ‘borrowing’ what they point to, rather than owning it. The difference is, when the reference goes out of scope, it will not deallocate the underlying memory. If it did, we’d de-allocate twice, which is bad!

Let’s add a third line. It looks innocent enough, but causes a compiler error:

fn main() {
      let mut x = vec!["Hello", "world"];
  
      let y = &x[0];
  
      x.push("foo");
  }

push is a method on vectors that appends another element to the end of the vector. When we try to compile this program, we get an error:

error: cannot borrow `x` as mutable because it is also borrowed as immutable
      x.push("foo");
      ^
  note: previous borrow of `x` occurs here; the immutable borrow prevents
  subsequent moves or mutable borrows of `x` until the borrow ends
      let y = &x[0];
               ^
  note: previous borrow ends here
  fn main() {
  
  }
  ^

Whew! The Rust compiler gives quite detailed errors at times, and this is one of those times. As the error explains, while we made our binding mutable, we still cannot call push. This is because we already have a reference to an element of the vector, y. Mutating something while another reference exists is dangerous, because we may invalidate the reference. In this specific case, when we create the vector, we may have only allocated space for two elements. Adding a third would mean allocating a new chunk of memory for all those elements, copying the old values over, and updating the internal pointer to that memory. That all works just fine. The problem is that y wouldn’t get updated, and so we’d have a ‘dangling pointer’. That’s bad. Any use of y would be an error in this case, and so the compiler has caught this for us.

So how do we solve this problem? There are two approaches we can take. The first is making a copy rather than using a reference:

fn main() {
      let mut x = vec!["Hello", "world"];
  
      let y = x[0].clone();
  
      x.push("foo");
  }

Rust has move semantics by default, so if we want to make a copy of some data, we call the clone() method. In this example, y is no longer a reference to the vector stored in x, but a copy of its first element, "Hello". Now that we don’t have a reference, our push() works just fine.

If we truly want a reference, we need the other option: ensure that our reference goes out of scope before we try to do the mutation. That looks like this:

fn main() {
      let mut x = vec!["Hello", "world"];
  
      {
          let y = &x[0];
      }
  
      x.push("foo");
  }

We created an inner scope with an additional set of curly braces. y will go out of scope before we call push(), and so we’re all good.

This concept of ownership isn’t just good for preventing dangling pointers, but an entire set of related problems, like iterator invalidation, concurrency, and more.

Getting Started

This first section of the book will get you going with Rust and its tooling. First, we’ll install Rust. Then, the classic ‘Hello World’ program. Finally, we’ll talk about Cargo, Rust’s build system and package manager.

Installing Rust

The first step to using Rust is to install it! There are a number of ways to install Rust, but the easiest is to use the rustup script. If you’re on Linux or a Mac, all you need to do is this:

Note: you don’t need to type in the $s, they just indicate the start of each command. You’ll see many tutorials and examples around the web that follow this convention: $ for commands run as your regular user, and # for commands you should be running as an administrator.

$ curl -sf -L https://static.rust-lang.org/rustup.sh | sh

If you’re concerned about the potential insecurity of using curl | sh, please keep reading and see our disclaimer below. And feel free to use a two-step version of the installation and examine our installation script:

$ curl -f -L https://static.rust-lang.org/rustup.sh -O
  $ sh rustup.sh

If you’re on Windows, please download the appropriate installer. NOTE: By default, the Windows installer will not add Rust to the %PATH% system variable. If this is the only version of Rust you are installing and you want to be able to run it from the command line, click on “Advanced” on the install dialog and on the “Product Features” page ensure “Add to PATH” is installed on the local hard drive.

Uninstalling

If you decide you don’t want Rust anymore, we’ll be a bit sad, but that’s okay. Not every programming language is great for everyone. Just run the uninstall script:

$ sudo /usr/local/lib/rustlib/uninstall.sh

If you used the Windows installer, just re-run the .msi and it will give you an uninstall option.

That disclaimer we promised

Some people, and somewhat rightfully so, get very upset when we tell you to curl | sh. Basically, when you do this, you are trusting that the good people who maintain Rust aren’t going to hack your computer and do bad things. That’s a good instinct! If you’re one of those people, please check out the documentation on building Rust from Source, or the official binary downloads.

Platform support

Oh, we should also mention the officially supported platforms:

  • Windows (7, 8, Server 2008 R2)
  • Linux (2.6.18 or later, various distributions), x86 and x86-64
  • OSX 10.7 (Lion) or greater, x86 and x86-64

We extensively test Rust on these platforms, and a few others, too, like Android. But these are the ones most likely to work, as they have the most testing.

Finally, a comment about Windows. Rust considers Windows to be a first-class platform upon release, but if we’re honest, the Windows experience isn’t as integrated as the Linux/OS X experience is. We’re working on it! If anything does not work, it is a bug. Please let us know if that happens. Each and every commit is tested against Windows just like any other platform.

After installation

If you’ve got Rust installed, you can open up a shell, and type this:

$ rustc --version

You should see the version number, commit hash, and commit date. If you just installed version 1.2.0, you should see:

rustc 1.2.0 (082e47636 2015-08-03)

If you did, Rust has been installed successfully! Congrats!

If you didn’t and you’re on Windows, check that Rust is in your %PATH% system variable. If it isn’t, run the installer again, select “Change” on the “Change, repair, or remove installation” page and ensure “Add to PATH” is installed on the local hard drive.

This installer also installs a copy of the documentation locally, so you can read it offline. On UNIX systems, /usr/local/share/doc/rust is the location. On Windows, it’s in a share/doc directory, inside wherever you installed Rust to.

If not, there are a number of places where you can get help. The easiest is the #rust IRC channel on irc.mozilla.org, which you can access through Mibbit. Click that link, and you’ll be chatting with other Rustaceans (a silly nickname we call ourselves), and we can help you out. Other great resources include the user’s forum, and Stack Overflow.

Hello, world!

Now that you have Rust installed, let’s write your first Rust program. It’s traditional to make your first program in any new language one that prints the text “Hello, world!” to the screen. The nice thing about starting with such a simple program is that you can verify that your compiler isn’t just installed, but also working properly. And printing information to the screen is a pretty common thing to do.

The first thing that we need to do is make a file to put our code in. I like to make a projects directory in my home directory, and keep all my projects there. Rust does not care where your code lives.

This actually leads to one other concern we should address: this guide will assume that you have basic familiarity with the command line. Rust itself makes no specific demands on your editing tooling, or where your code lives. If you prefer an IDE to the command line, you may want to check out SolidOak, or wherever plugins are for your favorite IDE. There are a number of extensions of varying quality in development by the community. The Rust team also ships plugins for various editors. Configuring your editor or IDE is out of the scope of this tutorial, so check the documentation for your setup, specifically.

With that said, let’s make a directory in our projects directory.

$ mkdir ~/projects
  $ cd ~/projects
  $ mkdir hello_world
  $ cd hello_world

If you’re on Windows and not using PowerShell, the ~ may not work. Consult the documentation for your shell for more details.

Let’s make a new source file next. We’ll call our file main.rs. Rust files always end in a .rs extension. If you’re using more than one word in your filename, use an underscore: hello_world.rs rather than helloworld.rs.

Now that you’ve got your file open, type this in:

fn main() {
      println!("Hello, world!");
  }

Save the file, and then type this into your terminal window:

$ rustc main.rs
  $ ./main # or main.exe on Windows
  Hello, world!

Success! Let’s go over what just happened in detail.

fn main() {
  
  }

These lines define a function in Rust. The main function is special: it’s the beginning of every Rust program. The first line says “I’m declaring a function named main which takes no arguments and returns nothing.” If there were arguments, they would go inside the parentheses (( and )), and because we aren’t returning anything from this function, we can omit the return type entirely. We’ll get to it later.

You’ll also note that the function is wrapped in curly braces ({ and }). Rust requires these around all function bodies. It is also considered good style to put the opening curly brace on the same line as the function declaration, with one space in between.

Next up is this line:

    println!("Hello, world!");

This line does all of the work in our little program. There are a number of details that are important here. The first is that it’s indented with four spaces, not tabs. Please configure your editor of choice to insert four spaces with the tab key. We provide some sample configurations for various editors.

The second point is the println!() part. This is calling a Rust macro, which is how metaprogramming is done in Rust. If it were a function instead, it would look like this: println(). For our purposes, we don’t need to worry about this difference. Just know that sometimes, you’ll see a !, and that means that you’re calling a macro instead of a normal function. Rust implements println! as a macro rather than a function for good reasons, but that’s an advanced topic. One last thing to mention: Rust’s macros are significantly different from C macros, if you’ve used those. Don’t be scared of using macros. We’ll get to the details eventually, you’ll just have to trust us for now.

Next, "Hello, world!" is a ‘string’. Strings are a surprisingly complicated topic in a systems programming language, and this is a ‘statically allocated’ string. If you want to read further about allocation, check out the stack and the heap, but you don’t need to right now if you don’t want to. We pass this string as an argument to println!, which prints the string to the screen. Easy enough!

Finally, the line ends with a semicolon (;). Rust is an ‘expression oriented’ language, which means that most things are expressions, rather than statements. The ; is used to indicate that this expression is over, and the next one is ready to begin. Most lines of Rust code end with a ;.

Finally, actually compiling and running our program. We can compile with our compiler, rustc, by passing it the name of our source file:

$ rustc main.rs

This is similar to gcc or clang, if you come from a C or C++ background. Rust will output a binary executable. You can see it with ls:

$ ls
  main  main.rs

Or on Windows:

$ dir
  main.exe  main.rs

There are now two files: our source code, with the .rs extension, and the executable (main.exe on Windows, main everywhere else)

$ ./main  # or main.exe on Windows

This prints out our Hello, world! text to our terminal.

If you come from a dynamic language like Ruby, Python, or JavaScript, you may not be used to these two steps being separate. Rust is an ‘ahead-of-time compiled language’, which means that you can compile a program, give it to someone else, and they don’t need to have Rust installed. If you give someone a .rb or .py or .js file, they need to have a Ruby/Python/JavaScript implementation installed, but you just need one command to both compile and run your program. Everything is a tradeoff in language design, and Rust has made its choice.

Congratulations! You have officially written a Rust program. That makes you a Rust programmer! Welcome. 🎊🎉👍

Next, I’d like to introduce you to another tool, Cargo, which is used to write real-world Rust programs. Just using rustc is nice for simple things, but as your project grows, you’ll want something to help you manage all of the options that it has, and to make it easy to share your code with other people and projects.

Hello, Cargo!

Cargo is a tool that Rustaceans use to help manage their Rust projects. Cargo is currently in a pre-1.0 state, and so it is still a work in progress. However, it is already good enough to use for many Rust projects, and so it is assumed that Rust projects will use Cargo from the beginning.

Cargo manages three things: building your code, downloading the dependencies your code needs, and building those dependencies. At first, your program doesn’t have any dependencies, so we’ll only be using the first part of its functionality. Eventually, we’ll add more. Since we started off by using Cargo, it’ll be easy to add later.

If we installed Rust via the official installers we will also have Cargo. If we installed Rust some other way, we may want to check the Cargo README for specific instructions about installing it.

Converting to Cargo

Let’s convert Hello World to Cargo.

To Cargo-ify our project, we need to do three things: Make a Cargo.toml configuration file, put our source file in the right place, and get rid of the old executable (main.exe on Windows, main everywhere else). Let’s do that part first:

$ mkdir src
  $ mv main.rs src/main.rs
  $ rm main  # or main.exe on Windows

Note that since we’re creating an executable, we retain main.rs as the source filename. If we want to make a library instead, we should use lib.rs. This convention is used by Cargo to successfully compile our projects, but it can be overridden if we wish. Custom file locations for the entry point can be specified with a [lib] or [[bin]] key in the TOML file.

Cargo expects your source files to live inside a src directory. That leaves the top level for other things, like READMEs, license information, and anything not related to your code. Cargo helps us keep our projects nice and tidy. A place for everything, and everything in its place.

Next, our configuration file:

$ editor Cargo.toml

Make sure to get this name right: you need the capital C!

Put this inside:

[package]
  
  name = "hello_world"
  version = "0.0.1"
  authors = [ "Your name <you@example.com>" ]

This file is in the TOML format. TOML is similar to INI, but has some extra goodies. According to the TOML docs,

TOML aims to be a minimal configuration file format that’s easy to read due to obvious semantics. TOML is designed to map unambiguously to a hash table. TOML should be easy to parse into data structures in a wide variety of languages.

Once we have this file in place in our project’s root directory, we should be ready to build! To do so, run:

$ cargo build
     Compiling hello_world v0.0.1 (file:///home/yourname/projects/hello_world)
  $ ./target/debug/hello_world
  Hello, world!

Bam! We built our project with cargo build, and ran it with ./target/debug/hello_world. We can do both in one step with cargo run:

$ cargo run
       Running `target/debug/hello_world`
  Hello, world!

Notice that we didn’t re-build the project this time. Cargo figured out that we hadn’t changed the source file, and so it just ran the binary. If we had made a modification, we would have seen it do both:

$ cargo run
     Compiling hello_world v0.0.1 (file:///home/yourname/projects/hello_world)
       Running `target/debug/hello_world`
  Hello, world!

This hasn’t bought us a whole lot over our simple use of rustc, but think about the future: when our project gets more complex, we need to do more things to get all of the parts to properly compile. With Cargo, as our project grows, we can just run cargo build, and it’ll work the right way.

When your project is finally ready for release, you can use cargo build --release to compile your project with optimizations.

You’ll also notice that Cargo has created a new file: Cargo.lock.

[root]
  name = "hello_world"
  version = "0.0.1"

The Cargo.lock file is used by Cargo to keep track of dependencies in your application. Right now, we don’t have any, so it’s a bit sparse. You won’t ever need to touch this file yourself, just let Cargo handle it.

That’s it! We’ve successfully built hello_world with Cargo. Even though our program is simple, it’s using much of the real tooling that you’ll use for the rest of your Rust career. You can expect to do this to get started with virtually all Rust projects:

$ git clone someurl.com/foo
  $ cd foo
  $ cargo build

A New Project

You don’t have to go through this whole process every time you want to start a new project! Cargo has the ability to make a bare-bones project directory in which you can start developing right away.

To start a new project with Cargo, use cargo new:

$ cargo new hello_world --bin

We’re passing --bin because our goal is to get straight to making an executable application, as opposed to a library. Executables are often called ‘binaries.’ (as in /usr/bin, if you’re on a Unix system)

Let’s check out what Cargo has generated for us:

$ cd hello_world
  $ tree .
  .
  ├── Cargo.toml
  └── src
      └── main.rs
  
  1 directory, 2 files

If you don’t have the tree command, you can probably get it from your distribution’s package manager. It’s not necessary, but it’s certainly useful.

This is all we need to get started. First, let’s check out Cargo.toml:

[package]
  
  name = "hello_world"
  version = "0.1.0"
  authors = ["Your Name <you@example.com>"]

Cargo has populated this file with reasonable defaults based off the arguments you gave it and your git global configuration. You may notice that Cargo has also initialized the hello_world directory as a git repository.

Here’s what’s in src/main.rs:

fn main() {
      println!("Hello, world!");
  }

Cargo has generated a “Hello World!” for us, and you’re ready to start coding! Cargo has its own guide which covers Cargo’s features in much more depth.

Now that you’ve got the tools down, let’s actually learn more about the Rust language itself. These are the basics that will serve you well through the rest of your time with Rust.

You have two options: Dive into a project with ‘Learn Rust’, or start from the bottom and work your way up with ‘Syntax and Semantics’. More experienced systems programmers will probably prefer ‘Learn Rust’, while those from dynamic backgrounds may enjoy either. Different people learn differently! Choose whatever’s right for you.

Learn Rust

Welcome! This section has a few tutorials that teach you Rust through building projects. You’ll get a high-level overview, but we’ll skim over the details.

If you’d prefer a more ‘from the ground up’-style experience, check out Syntax and Semantics.

Guessing Game

For our first project, we’ll implement a classic beginner programming problem: the guessing game. Here’s how it works: Our program will generate a random integer between one and a hundred. It will then prompt us to enter a guess. Upon entering our guess, it will tell us if we’re too low or too high. Once we guess correctly, it will congratulate us. Sounds good?

Set up

Let’s set up a new project. Go to your projects directory. Remember how we had to create our directory structure and a Cargo.toml for hello_world? Cargo has a command that does that for us. Let’s give it a shot:

$ cd ~/projects
  $ cargo new guessing_game --bin
  $ cd guessing_game

We pass the name of our project to cargo new, and then the --bin flag, since we’re making a binary, rather than a library.

Check out the generated Cargo.toml:

[package]
  
  name = "guessing_game"
  version = "0.1.0"
  authors = ["Your Name <you@example.com>"]

Cargo gets this information from your environment. If it’s not correct, go ahead and fix that.

Finally, Cargo generated a ‘Hello, world!’ for us. Check out src/main.rs:

fn main() {
      println!("Hello, world!");
  }

Let’s try compiling what Cargo gave us:

$ cargo build
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)

Excellent! Open up your src/main.rs again. We’ll be writing all of our code in this file.

Before we move on, let me show you one more Cargo command: run. cargo run is kind of like cargo build, but it also then runs the produced executable. Try it out:

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/debug/guessing_game`
  Hello, world!

Great! The run command comes in handy when you need to rapidly iterate on a project. Our game is just such a project, we need to quickly test each iteration before moving on to the next one.

Processing a Guess

Let’s get to it! The first thing we need to do for our guessing game is allow our player to input a guess. Put this in your src/main.rs:

use std::io;
  
  fn main() {
      println!("Guess the number!");
  
      println!("Please input your guess.");
  
      let mut guess = String::new();
  
      io::stdin().read_line(&mut guess)
          .ok()
          .expect("Failed to read line");
  
      println!("You guessed: {}", guess);
  }

There’s a lot here! Let’s go over it, bit by bit.

use std::io;

We’ll need to take user input, and then print the result as output. As such, we need the io library from the standard library. Rust only imports a few things by default into every program, the ‘prelude’. If it’s not in the prelude, you’ll have to use it directly.

fn main() {

As you’ve seen before, the main() function is the entry point into your program. The fn syntax declares a new function, the ()s indicate that there are no arguments, and { starts the body of the function. Because we didn’t include a return type, it’s assumed to be (), an empty tuple.

    println!("Guess the number!");
  
      println!("Please input your guess.");

We previously learned that println!() is a macro that prints a string to the screen.

    let mut guess = String::new();

Now we’re getting interesting! There’s a lot going on in this little line. The first thing to notice is that this is a let statement, which is used to create ‘variable bindings’. They take this form:

let foo = bar;

This will create a new binding named foo, and bind it to the value bar. In many languages, this is called a ‘variable’, but Rust’s variable bindings have a few tricks up their sleeves.

For example, they’re immutable by default. That’s why our example uses mut: it makes a binding mutable, rather than immutable. let doesn’t take a name on the left hand side, it actually accepts a ‘pattern’. We’ll use patterns later. It’s easy enough to use for now:

let foo = 5; // immutable.
  let mut bar = 5; // mutable

Oh, and // will start a comment, until the end of the line. Rust ignores everything in comments.

So now we know that let mut guess will introduce a mutable binding named guess, but we have to look at the other side of the = for what it’s bound to: String::new().

String is a string type, provided by the standard library. A String is a growable, UTF-8 encoded bit of text.

The ::new() syntax uses :: because this is an ‘associated function’ of a particular type. That is to say, it’s associated with String itself, rather than a particular instance of a String. Some languages call this a ‘static method’.

This function is named new(), because it creates a new, empty String. You’ll find a new() function on many types, as it’s a common name for making a new value of some kind.

Let’s move forward:

    io::stdin().read_line(&mut guess)
          .ok()
          .expect("Failed to read line");

That’s a lot more! Let’s go bit-by-bit. The first line has two parts. Here’s the first:

io::stdin()

Remember how we used std::io on the first line of the program? We’re now calling an associated function on it. If we didn’t use std::io, we could have written this line as std::io::stdin().

This particular function returns a handle to the standard input for your terminal. More specifically, a std::io::Stdin.

The next part will use this handle to get input from the user:

.read_line(&mut guess)

Here, we call the read_line() method on our handle. Methods are like associated functions, but are only available on a particular instance of a type, rather than the type itself. We’re also passing one argument to read_line(): &mut guess.

Remember how we bound guess above? We said it was mutable. However, read_line doesn’t take a String as an argument: it takes a &mut String. Rust has a feature called ‘references’, which allows you to have multiple references to one piece of data, which can reduce copying. References are a complex feature, as one of Rust’s major selling points is how safe and easy it is to use references. We don’t need to know a lot of those details to finish our program right now, though. For now, all we need to know is that like let bindings, references are immutable by default. Hence, we need to write &mut guess, rather than &guess.

Why does read_line() take a mutable reference to a string? Its job is to take what the user types into standard input, and place that into a string. So it takes that string as an argument, and in order to add the input, it needs to be mutable.

But we’re not quite done with this line of code, though. While it’s a single line of text, it’s only the first part of the single logical line of code:

        .ok()
          .expect("Failed to read line");

When you call a method with the .foo() syntax, you may introduce a newline and other whitespace. This helps you split up long lines. We could have done:

    io::stdin().read_line(&mut guess).ok().expect("failed to read line");

But that gets hard to read. So we’ve split it up, three lines for three method calls. We already talked about read_line(), but what about ok() and expect()? Well, we already mentioned that read_line() puts what the user types into the &mut String we pass it. But it also returns a value: in this case, an io::Result. Rust has a number of types named Result in its standard library: a generic Result, and then specific versions for sub-libraries, like io::Result.

The purpose of these Result types is to encode error handling information. Values of the Result type, like any type, have methods defined on them. In this case, io::Result has an ok() method, which says ‘we want to assume this value is a successful one. If not, just throw away the error information’. Why throw it away? Well, for a basic program, we just want to print a generic error, as basically any issue means we can’t continue. The ok() method returns a value which has another method defined on it: expect(). The expect() method takes a value it’s called on, and if it isn’t a successful one, panic!s with a message you passed it. A panic! like this will cause our program to crash, displaying the message.

If we leave off calling these two methods, our program will compile, but we’ll get a warning:

$ cargo build
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
  src/main.rs:10:5: 10:39 warning: unused result which must be used,
  #[warn(unused_must_use)] on by default
  src/main.rs:10     io::stdin().read_line(&mut guess);
                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Rust warns us that we haven’t used the Result value. This warning comes from a special annotation that io::Result has. Rust is trying to tell you that you haven’t handled a possible error. The right way to suppress the error is to actually write error handling. Luckily, if we just want to crash if there’s a problem, we can use these two little methods. If we can recover from the error somehow, we’d do something else, but we’ll save that for a future project.

There’s just one line of this first example left:

    println!("You guessed: {}", guess);
  }

This prints out the string we saved our input in. The {}s are a placeholder, and so we pass it guess as an argument. If we had multiple {}s, we would pass multiple arguments:

let x = 5;
  let y = 10;
  
  println!("x and y: {} and {}", x, y);

Easy.

Anyway, that’s the tour. We can run what we have with cargo run:

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/debug/guessing_game`
  Guess the number!
  Please input your guess.
  6
  You guessed: 6

All right! Our first part is done: we can get input from the keyboard, and then print it back out.

Generating a secret number

Next, we need to generate a secret number. Rust does not yet include random number functionality in its standard library. The Rust team does, however, provide a rand crate. A ‘crate’ is a package of Rust code. We’ve been building a ‘binary crate’, which is an executable. rand is a ‘library crate’, which contains code that’s intended to be used with other programs.

Using external crates is where Cargo really shines. Before we can write the code using rand, we need to modify our Cargo.toml. Open it up, and add these few lines at the bottom:

[dependencies]
  
  rand="0.3.0"

The [dependencies] section of Cargo.toml is like the [package] section: everything that follows it is part of it, until the next section starts. Cargo uses the dependencies section to know what dependencies on external crates you have, and what versions you require. In this case, we’ve specified version 0.3.0, which Cargo understands to be any release that’s compatible with this specific version. Cargo understands Semantic Versioning, which is a standard for writing version numbers. If we wanted to use only 0.3.0 exactly, we could use =0.3.0. If we wanted to use the latest version we could use *; We could use a range of versions. Cargo’s documentation contains more details.

Now, without changing any of our code, let’s build our project:

$ cargo build
      Updating registry `https://github.com/rust-lang/crates.io-index`
   Downloading rand v0.3.8
   Downloading libc v0.1.6
     Compiling libc v0.1.6
     Compiling rand v0.3.8
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)

(You may see different versions, of course.)

Lots of new output! Now that we have an external dependency, Cargo fetches the latest versions of everything from the registry, which is a copy of data from Crates.io. Crates.io is where people in the Rust ecosystem post their open source Rust projects for others to use.

After updating the registry, Cargo checks our [dependencies] and downloads any we don’t have yet. In this case, while we only said we wanted to depend on rand, we’ve also grabbed a copy of libc. This is because rand depends on libc to work. After downloading them, it compiles them, and then compiles our project.

If we run cargo build again, we’ll get different output:

$ cargo build

That’s right, no output! Cargo knows that our project has been built, and that all of its dependencies are built, and so there’s no reason to do all that stuff. With nothing to do, it simply exits. If we open up src/main.rs again, make a trivial change, and then save it again, we’ll just see one line:

$ cargo build
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)

So, we told Cargo we wanted any 0.3.x version of rand, and so it fetched the latest version at the time this was written, v0.3.8. But what happens when next week, version v0.3.9 comes out, with an important bugfix? While getting bugfixes is important, what if 0.3.9 contains a regression that breaks our code?

The answer to this problem is the Cargo.lock file you’ll now find in your project directory. When you build your project for the first time, Cargo figures out all of the versions that fit your criteria, and then writes them to the Cargo.lock file. When you build your project in the future, Cargo will see that the Cargo.lock file exists, and then use that specific version rather than do all the work of figuring out versions again. This lets you have a repeatable build automatically. In other words, we’ll stay at 0.3.8 until we explicitly upgrade, and so will anyone who we share our code with, thanks to the lock file.

What about when we do want to use v0.3.9? Cargo has another command, update, which says ‘ignore the lock, figure out all the latest versions that fit what we’ve specified. If that works, write those versions out to the lock file’. But, by default, Cargo will only look for versions larger than 0.3.0 and smaller than 0.4.0. If we want to move to 0.4.x, we’d have to update the Cargo.toml directly. When we do, the next time we cargo build, Cargo will update the index and re-evaluate our rand requirements.

There’s a lot more to say about Cargo and its ecosystem, but for now, that’s all we need to know. Cargo makes it really easy to re-use libraries, and so Rustaceans tend to write smaller projects which are assembled out of a number of sub-packages.

Let’s get on to actually using rand. Here’s our next step:

extern crate rand;
  
  use std::io;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      println!("Please input your guess.");
  
      let mut guess = String::new();
  
      io::stdin().read_line(&mut guess)
          .ok()
          .expect("failed to read line");
  
      println!("You guessed: {}", guess);
  }

The first thing we’ve done is change the first line. It now says extern crate rand. Because we declared rand in our [dependencies], we can use extern crate to let Rust know we’ll be making use of it. This also does the equivalent of a use rand; as well, so we can make use of anything in the rand crate by prefixing it with rand::.

Next, we added another use line: use rand::Rng. We’re going to use a method in a moment, and it requires that Rng be in scope to work. The basic idea is this: methods are defined on something called ‘traits’, and for the method to work, it needs the trait to be in scope. For more about the details, read the traits section.

There are two other lines we added, in the middle:

    let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);

We use the rand::thread_rng() function to get a copy of the random number generator, which is local to the particular thread of execution we’re in. Because we use rand::Rng’d above, it has a gen_range() method available. This method takes two arguments, and generates a number between them. It’s inclusive on the lower bound, but exclusive on the upper bound, so we need 1 and 101 to get a number ranging from one to a hundred.

The second line just prints out the secret number. This is useful while we’re developing our program, so we can easily test it out. But we’ll be deleting it for the final version. It’s not much of a game if it prints out the answer when you start it up!

Try running our new program a few times:

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/debug/guessing_game`
  Guess the number!
  The secret number is: 7
  Please input your guess.
  4
  You guessed: 4
  $ cargo run
       Running `target/debug/guessing_game`
  Guess the number!
  The secret number is: 83
  Please input your guess.
  5
  You guessed: 5

Great! Next up: let’s compare our guess to the secret guess.

Comparing guesses

Now that we’ve got user input, let’s compare our guess to the random guess. Here’s our next step, though it doesn’t quite compile yet:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      println!("Please input your guess.");
  
      let mut guess = String::new();
  
      io::stdin().read_line(&mut guess)
          .ok()
          .expect("failed to read line");
  
      println!("You guessed: {}", guess);
  
      match guess.cmp(&secret_number) {
          Ordering::Less    => println!("Too small!"),
          Ordering::Greater => println!("Too big!"),
          Ordering::Equal   => println!("You win!"),
      }
  }

A few new bits here. The first is another use. We bring a type called std::cmp::Ordering into scope. Then, five new lines at the bottom that use it:

match guess.cmp(&secret_number) {
      Ordering::Less    => println!("Too small!"),
      Ordering::Greater => println!("Too big!"),
      Ordering::Equal   => println!("You win!"),
  }

The cmp() method can be called on anything that can be compared, and it takes a reference to the thing you want to compare it to. It returns the Ordering type we used earlier. We use a match statement to determine exactly what kind of Ordering it is. Ordering is an enum, short for ‘enumeration’, which looks like this:

enum Foo {
      Bar,
      Baz,
  }

With this definition, anything of type Foo can be either a Foo::Bar or a Foo::Baz. We use the :: to indicate the namespace for a particular enum variant.

The Ordering enum has three possible variants: Less, Equal, and Greater. The match statement takes a value of a type, and lets you create an ‘arm’ for each possible value. Since we have three types of Ordering, we have three arms:

match guess.cmp(&secret_number) {
      Ordering::Less    => println!("Too small!"),
      Ordering::Greater => println!("Too big!"),
      Ordering::Equal   => println!("You win!"),
  }

If it’s Less, we print Too small!, if it’s Greater, Too big!, and if Equal, You win!. match is really useful, and is used often in Rust.

I did mention that this won’t quite compile yet, though. Let’s try it:

$ cargo build
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
  src/main.rs:28:21: 28:35 error: mismatched types:
   expected `&collections::string::String`,
      found `&_`
  (expected struct `collections::string::String`,
      found integral variable) [E0308]
  src/main.rs:28     match guess.cmp(&secret_number) {
                                     ^~~~~~~~~~~~~~
  error: aborting due to previous error
  Could not compile `guessing_game`.

Whew! This is a big error. The core of it is that we have ‘mismatched types’. Rust has a strong, static type system. However, it also has type inference. When we wrote let guess = String::new(), Rust was able to infer that guess should be a String, and so it doesn’t make us write out the type. And with our secret_number, there are a number of types which can have a value between one and a hundred: i32, a thirty-two-bit number, or u32, an unsigned thirty-two-bit number, or i64, a sixty-four-bit number or others. So far, that hasn’t mattered, and so Rust defaults to an i32. However, here, Rust doesn’t know how to compare the guess and the secret_number. They need to be the same type. Ultimately, we want to convert the String we read as input into a real number type, for comparison. We can do that with three more lines. Here’s our new program:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      println!("Please input your guess.");
  
      let mut guess = String::new();
  
      io::stdin().read_line(&mut guess)
          .ok()
          .expect("failed to read line");
  
      let guess: u32 = guess.trim().parse()
          .ok()
          .expect("Please type a number!");
  
      println!("You guessed: {}", guess);
  
      match guess.cmp(&secret_number) {
          Ordering::Less    => println!("Too small!"),
          Ordering::Greater => println!("Too big!"),
          Ordering::Equal   => println!("You win!"),
      }
  }

The new three lines:

    let guess: u32 = guess.trim().parse()
          .ok()
          .expect("Please type a number!");

Wait a minute, I thought we already had a guess? We do, but Rust allows us to ‘shadow’ the previous guess with a new one. This is often used in this exact situation, where guess starts as a String, but we want to convert it to an u32. Shadowing lets us re-use the guess name, rather than forcing us to come up with two unique names like guess_str and guess, or something else.

We bind guess to an expression that looks like something we wrote earlier:

guess.trim().parse()

Followed by an ok().expect() invocation. Here, guess refers to the old guess, the one that was a String with our input in it. The trim() method on Strings will eliminate any white space at the beginning and end of our string. This is important, as we had to press the ‘return’ key to satisfy read_line(). This means that if we type 5 and hit return, guess looks like this: 5\n. The \n represents ‘newline’, the enter key. trim() gets rid of this, leaving our string with just the 5. The parse() method on strings parses a string into some kind of number. Since it can parse a variety of numbers, we need to give Rust a hint as to the exact type of number we want. Hence, let guess: u32. The colon (:) after guess tells Rust we’re going to annotate its type. u32 is an unsigned, thirty-two bit integer. Rust has a number of built-in number types, but we’ve chosen u32. It’s a good default choice for a small positive number.

Just like read_line(), our call to parse() could cause an error. What if our string contained A👍%? There’d be no way to convert that to a number. As such, we’ll do the same thing we did with read_line(): use the ok() and expect() methods to crash if there’s an error.

Let’s try our program out!

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/guessing_game`
  Guess the number!
  The secret number is: 58
  Please input your guess.
    76
  You guessed: 76
  Too big!

Nice! You can see I even added spaces before my guess, and it still figured out that I guessed 76. Run the program a few times, and verify that guessing the number works, as well as guessing a number too small.

Now we’ve got most of the game working, but we can only make one guess. Let’s change that by adding loops!

Looping

The loop keyword gives us an infinite loop. Let’s add that in:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      loop {
          println!("Please input your guess.");
  
          let mut guess = String::new();
  
          io::stdin().read_line(&mut guess)
              .ok()
              .expect("failed to read line");
  
          let guess: u32 = guess.trim().parse()
              .ok()
              .expect("Please type a number!");
  
          println!("You guessed: {}", guess);
  
          match guess.cmp(&secret_number) {
              Ordering::Less    => println!("Too small!"),
              Ordering::Greater => println!("Too big!"),
              Ordering::Equal   => println!("You win!"),
          }
      }
  }

And try it out. But wait, didn’t we just add an infinite loop? Yup. Remember our discussion about parse()? If we give a non-number answer, we’ll return and quit. Observe:

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/guessing_game`
  Guess the number!
  The secret number is: 59
  Please input your guess.
  45
  You guessed: 45
  Too small!
  Please input your guess.
  60
  You guessed: 60
  Too big!
  Please input your guess.
  59
  You guessed: 59
  You win!
  Please input your guess.
  quit
  thread '<main>' panicked at 'Please type a number!'

Ha! quit actually quits. As does any other non-number input. Well, this is suboptimal to say the least. First, let’s actually quit when you win the game:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      loop {
          println!("Please input your guess.");
  
          let mut guess = String::new();
  
          io::stdin().read_line(&mut guess)
              .ok()
              .expect("failed to read line");
  
          let guess: u32 = guess.trim().parse()
              .ok()
              .expect("Please type a number!");
  
          println!("You guessed: {}", guess);
  
          match guess.cmp(&secret_number) {
              Ordering::Less    => println!("Too small!"),
              Ordering::Greater => println!("Too big!"),
              Ordering::Equal   => {
                  println!("You win!");
                  break;
              }
          }
      }
  }

By adding the break line after the You win!, we’ll exit the loop when we win. Exiting the loop also means exiting the program, since it’s the last thing in main(). We have just one more tweak to make: when someone inputs a non-number, we don’t want to quit, we just want to ignore it. We can do that like this:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      println!("The secret number is: {}", secret_number);
  
      loop {
          println!("Please input your guess.");
  
          let mut guess = String::new();
  
          io::stdin().read_line(&mut guess)
              .ok()
              .expect("failed to read line");
  
          let guess: u32 = match guess.trim().parse() {
              Ok(num) => num,
              Err(_) => continue,
          };
  
          println!("You guessed: {}", guess);
  
          match guess.cmp(&secret_number) {
              Ordering::Less    => println!("Too small!"),
              Ordering::Greater => println!("Too big!"),
              Ordering::Equal   => {
                  println!("You win!");
                  break;
              }
          }
      }
  }

These are the lines that changed:

let guess: u32 = match guess.trim().parse() {
      Ok(num) => num,
      Err(_) => continue,
  };

This is how you generally move from ‘crash on error’ to ‘actually handle the error’, by switching from ok().expect() to a match statement. The Result returned by parse() is an enum just like Ordering, but in this case, each variant has some data associated with it: Ok is a success, and Err is a failure. Each contains more information: the successful parsed integer, or an error type. In this case, we match on Ok(num), which sets the inner value of the Ok to the name num, and then we just return it on the right-hand side. In the Err case, we don’t care what kind of error it is, so we just use _ instead of a name. This ignores the error, and continue causes us to go to the next iteration of the loop.

Now we should be good! Let’s try:

$ cargo run
     Compiling guessing_game v0.1.0 (file:///home/you/projects/guessing_game)
       Running `target/guessing_game`
  Guess the number!
  The secret number is: 61
  Please input your guess.
  10
  You guessed: 10
  Too small!
  Please input your guess.
  99
  You guessed: 99
  Too big!
  Please input your guess.
  foo
  Please input your guess.
  61
  You guessed: 61
  You win!

Awesome! With one tiny last tweak, we have finished the guessing game. Can you think of what it is? That’s right, we don’t want to print out the secret number. It was good for testing, but it kind of ruins the game. Here’s our final source:

extern crate rand;
  
  use std::io;
  use std::cmp::Ordering;
  use rand::Rng;
  
  fn main() {
      println!("Guess the number!");
  
      let secret_number = rand::thread_rng().gen_range(1, 101);
  
      loop {
          println!("Please input your guess.");
  
          let mut guess = String::new();
  
          io::stdin().read_line(&mut guess)
              .ok()
              .expect("failed to read line");
  
          let guess: u32 = match guess.trim().parse() {
              Ok(num) => num,
              Err(_) => continue,
          };
  
          println!("You guessed: {}", guess);
  
          match guess.cmp(&secret_number) {
              Ordering::Less    => println!("Too small!"),
              Ordering::Greater => println!("Too big!"),
              Ordering::Equal   => {
                  println!("You win!");
                  break;
              }
          }
      }
  }

Complete!

At this point, you have successfully built the Guessing Game! Congratulations!

This first project showed you a lot: let, match, methods, associated functions, using external crates, and more. Our next project will show off even more.

Dining Philosophers

For our second project, let’s look at a classic concurrency problem. It’s called ‘the dining philosophers’. It was originally conceived by Dijkstra in 1965, but we’ll use a lightly adapted version from this paper by Tony Hoare in 1985.

In ancient times, a wealthy philanthropist endowed a College to accommodate five eminent philosophers. Each philosopher had a room in which they could engage in their professional activity of thinking; there was also a common dining room, furnished with a circular table, surrounded by five chairs, each labelled by the name of the philosopher who was to sit in it. They sat anticlockwise around the table. To the left of each philosopher there was laid a golden fork, and in the centre stood a large bowl of spaghetti, which was constantly replenished. A philosopher was expected to spend most of their time thinking; but when they felt hungry, they went to the dining room, sat down in their own chair, picked up their own fork on their left, and plunged it into the spaghetti. But such is the tangled nature of spaghetti that a second fork is required to carry it to the mouth. The philosopher therefore had also to pick up the fork on their right. When they were finished they would put down both their forks, get up from their chair, and continue thinking. Of course, a fork can be used by only one philosopher at a time. If the other philosopher wants it, they just have to wait until the fork is available again.

This classic problem shows off a few different elements of concurrency. The reason is that it’s actually slightly tricky to implement: a simple implementation can deadlock. For example, let’s consider a simple algorithm that would solve this problem:

  1. A philosopher picks up the fork on their left.
  2. They then pick up the fork on their right.
  3. They eat.
  4. They return the forks.

Now, let’s imagine this sequence of events:

  1. Philosopher 1 begins the algorithm, picking up the fork on their left.
  2. Philosopher 2 begins the algorithm, picking up the fork on their left.
  3. Philosopher 3 begins the algorithm, picking up the fork on their left.
  4. Philosopher 4 begins the algorithm, picking up the fork on their left.
  5. Philosopher 5 begins the algorithm, picking up the fork on their left.
  6. … ? All the forks are taken, but nobody can eat!

There are different ways to solve this problem. We’ll get to our solution in the tutorial itself. For now, let’s get started modelling the problem itself. We’ll start with the philosophers:

struct Philosopher {
      name: String,
  }
  
  impl Philosopher {
      fn new(name: &str) -> Philosopher {
          Philosopher {
              name: name.to_string(),
          }
      }
  }
  
  fn main() {
      let p1 = Philosopher::new("Judith Butler");
      let p2 = Philosopher::new("Gilles Deleuze");
      let p3 = Philosopher::new("Karl Marx");
      let p4 = Philosopher::new("Emma Goldman");
      let p5 = Philosopher::new("Michel Foucault");
  }

Here, we make a struct to represent a philosopher. For now, a name is all we need. We choose the String type for the name, rather than &str. Generally speaking, working with a type which owns its data is easier than working with one that uses references.

Let’s continue:

impl Philosopher {
      fn new(name: &str) -> Philosopher {
          Philosopher {
              name: name.to_string(),
          }
      }
  }

This impl block lets us define things on Philosopher structs. In this case, we define an ‘associated function’ called new. The first line looks like this:

fn new(name: &str) -> Philosopher {

We take one argument, a name, of type &str. This is a reference to another string. It returns an instance of our Philosopher struct.

Philosopher {
      name: name.to_string(),
  }

This creates a new Philosopher, and sets its name to our name argument. Not just the argument itself, though, as we call .to_string() on it. This will create a copy of the string that our &str points to, and give us a new String, which is the type of the name field of Philosopher.

Why not accept a String directly? It’s nicer to call. If we took a String, but our caller had a &str, they’d have to call this method themselves. The downside of this flexibility is that we always make a copy. For this small program, that’s not particularly important, as we know we’ll just be using short strings anyway.

One last thing you’ll notice: we just define a Philosopher, and seemingly don’t do anything with it. Rust is an ‘expression based’ language, which means that almost everything in Rust is an expression which returns a value. This is true of functions as well, the last expression is automatically returned. Since we create a new Philosopher as the last expression of this function, we end up returning it.

This name, new(), isn’t anything special to Rust, but it is a convention for functions that create new instances of structs. Before we talk about why, let’s look at main() again:

#
  #
  fn main() {
      let p1 = Philosopher::new("Judith Butler");
      let p2 = Philosopher::new("Gilles Deleuze");
      let p3 = Philosopher::new("Karl Marx");
      let p4 = Philosopher::new("Emma Goldman");
      let p5 = Philosopher::new("Michel Foucault");
  }

Here, we create five variable bindings with five new philosophers. These are my favorite five, but you can substitute anyone you want. If we didn’t define that new() function, it would look like this:

fn main() {
      let p1 = Philosopher { name: "Judith Butler".to_string() };
      let p2 = Philosopher { name: "Gilles Deleuze".to_string() };
      let p3 = Philosopher { name: "Karl Marx".to_string() };
      let p4 = Philosopher { name: "Emma Goldman".to_string() };
      let p5 = Philosopher { name: "Michel Foucault".to_string() };
  }

That’s much noisier. Using new has other advantages too, but even in this simple case, it ends up being nicer to use.

Now that we’ve got the basics in place, there’s a number of ways that we can tackle the broader problem here. I like to start from the end first: let’s set up a way for each philosopher to finish eating. As a tiny step, let’s make a method, and then loop through all the philosophers, calling it:

struct Philosopher {
      name: String,
  }
  
  impl Philosopher {
      fn new(name: &str) -> Philosopher {
          Philosopher {
              name: name.to_string(),
          }
      }
  
      fn eat(&self) {
          println!("{} is done eating.", self.name);
      }
  }
  
  fn main() {
      let philosophers = vec![
          Philosopher::new("Judith Butler"),
          Philosopher::new("Gilles Deleuze"),
          Philosopher::new("Karl Marx"),
          Philosopher::new("Emma Goldman"),
          Philosopher::new("Michel Foucault"),
      ];
  
      for p in &philosophers {
          p.eat();
      }
  }

Let’s look at main() first. Rather than have five individual variable bindings for our philosophers, we make a Vec<T> of them instead. Vec<T> is also called a ‘vector’, and it’s a growable array type. We then use a for loop to iterate through the vector, getting a reference to each philosopher in turn.

In the body of the loop, we call p.eat(), which is defined above:

fn eat(&self) {
      println!("{} is done eating.", self.name);
  }

In Rust, methods take an explicit self parameter. That’s why eat() is a method, but new is an associated function: new() has no self. For our first version of eat(), we just print out the name of the philosopher, and mention they’re done eating. Running this program should give you the following output:

Judith Butler is done eating.
  Gilles Deleuze is done eating.
  Karl Marx is done eating.
  Emma Goldman is done eating.
  Michel Foucault is done eating.

Easy enough, they’re all done! We haven’t actually implemented the real problem yet, though, so we’re not done yet!

Next, we want to make our philosophers not just finish eating, but actually eat. Here’s the next version:

use std::thread;
  
  struct Philosopher {
      name: String,
  }
  
  impl Philosopher {
      fn new(name: &str) -> Philosopher {
          Philosopher {
              name: name.to_string(),
          }
      }
  
      fn eat(&self) {
          println!("{} is eating.", self.name);
  
          thread::sleep_ms(1000);
  
          println!("{} is done eating.", self.name);
      }
  }
  
  fn main() {
      let philosophers = vec![
          Philosopher::new("Judith Butler"),
          Philosopher::new("Gilles Deleuze"),
          Philosopher::new("Karl Marx"),
          Philosopher::new("Emma Goldman"),
          Philosopher::new("Michel Foucault"),
      ];
  
      for p in &philosophers {
          p.eat();
      }
  }

Just a few changes. Let’s break it down.

use std::thread;

use brings names into scope. We’re going to start using the thread module from the standard library, and so we need to use it.

    fn eat(&self) {
          println!("{} is eating.", self.name);
  
          thread::sleep_ms(1000);
  
          println!("{} is done eating.", self.name);
      }

We now print out two messages, with a sleep_ms() in the middle. This will simulate the time it takes a philosopher to eat.

If you run this program, you should see each philosopher eat in turn:

Judith Butler is eating.
  Judith Butler is done eating.
  Gilles Deleuze is eating.
  Gilles Deleuze is done eating.
  Karl Marx is eating.
  Karl Marx is done eating.
  Emma Goldman is eating.
  Emma Goldman is done eating.
  Michel Foucault is eating.
  Michel Foucault is done eating.

Excellent! We’re getting there. There’s just one problem: we aren’t actually operating in a concurrent fashion, which is a core part of the problem!

To make our philosophers eat concurrently, we need to make a small change. Here’s the next iteration:

use std::thread;
  
  struct Philosopher {
      name: String,
  }
  
  impl Philosopher {
      fn new(name: &str) -> Philosopher {
          Philosopher {
              name: name.to_string(),
          }
      }
  
      fn eat(&self) {
          println!("{} is eating.", self.name);
  
          thread::sleep_ms(1000);
  
          println!("{} is done eating.", self.name);
      }
  }
  
  fn main() {
      let philosophers = vec![
          Philosopher::new("Judith Butler"),
          Philosopher::new("Gilles Deleuze"),
          Philosopher::new("Karl Marx"),
          Philosopher::new("Emma Goldman"),
          Philosopher::new("Michel Foucault"),
      ];
  
      let handles: Vec<_> = philosophers.into_iter().map(|p| {
          thread::spawn(move || {
              p.eat();
          })
      }).collect();
  
      for h in handles {
          h.join().unwrap();
      }
  }

All we’ve done is change the loop in main(), and added a second one! Here’s the first change:

let handles: Vec<_> = philosophers.into_iter().map(|p| {
      thread::spawn(move || {
          p.eat();
      })
  }).collect();

While this is only five lines, they’re a dense five. Let’s break it down.

let handles: Vec<_> =

We introduce a new binding, called handles. We’ve given it this name because we are going to make some new threads, and that will return some handles to those threads that let us control their operation. We need to explicitly annotate the type here, though, due to an issue we’ll talk about later. The _ is a type placeholder. We’re saying “handles is a vector of something, but you can figure out what that something is, Rust.”

philosophers.into_iter().map(|p| {

We take our list of philosophers and call into_iter() on it. This creates an iterator that takes ownership of each philosopher. We need to do this to pass them to our threads. We take that iterator and call map on it, which takes a closure as an argument and calls that closure on each element in turn.

    thread::spawn(move || {
          p.eat();
      })

Here’s where the concurrency happens. The thread::spawn function takes a closure as an argument and executes that closure in a new thread. This closure needs an extra annotation, move, to indicate that the closure is going to take ownership of the values it’s capturing. Primarily, the p variable of the map function.

Inside the thread, all we do is call eat() on p. Also note that the call to thread::spawn lacks a trailing semicolon, making this an expression. This distinction is important, yielding the correct return value. For more details, read Expressions vs. Statements.

}).collect();

Finally, we take the result of all those map calls and collect them up. collect() will make them into a collection of some kind, which is why we needed to annotate the return type: we want a Vec<T>. The elements are the return values of the thread::spawn calls, which are handles to those threads. Whew!

for h in handles {
      h.join().unwrap();
  }

At the end of main(), we loop through the handles and call join() on them, which blocks execution until the thread has completed execution. This ensures that the threads complete their work before the program exits.

If you run this program, you’ll see that the philosophers eat out of order! We have multi-threading!

Judith Butler is eating.
  Gilles Deleuze is eating.
  Karl Marx is eating.
  Emma Goldman is eating.
  Michel Foucault is eating.
  Judith Butler is done eating.
  Gilles Deleuze is done eating.
  Karl Marx is done eating.
  Emma Goldman is done eating.
  Michel Foucault is done eating.

But what about the forks? We haven’t modeled them at all yet.

To do that, let’s make a new struct:

use std::sync::Mutex;
  
  struct Table {
      forks: Vec<Mutex<()>>,
  }

This Table has a vector of Mutexes. A mutex is a way to control concurrency: only one thread can access the contents at once. This is exactly the property we need with our forks. We use an empty tuple, (), inside the mutex, since we’re not actually going to use the value, just hold onto it.

Let’s modify the program to use the Table:

use std::thread;
  use std::sync::{Mutex, Arc};
  
  struct Philosopher {
      name: String,
      left: usize,
      right: usize,
  }
  
  impl Philosopher {
      fn new(name: &str, left: usize, right: usize) -> Philosopher {
          Philosopher {
              name: name.to_string(),
              left: left,
              right: right,
          }
      }
  
      fn eat(&self, table: &Table) {
          let _left = table.forks[self.left].lock().unwrap();
          let _right = table.forks[self.right].lock().unwrap();
  
          println!("{} is eating.", self.name);
  
          thread::sleep_ms(1000);
  
          println!("{} is done eating.", self.name);
      }
  }
  
  struct Table {
      forks: Vec<Mutex<()>>,
  }
  
  fn main() {
      let table = Arc::new(Table { forks: vec![
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
      ]});
  
      let philosophers = vec![
          Philosopher::new("Judith Butler", 0, 1),
          Philosopher::new("Gilles Deleuze", 1, 2),
          Philosopher::new("Karl Marx", 2, 3),
          Philosopher::new("Emma Goldman", 3, 4),
          Philosopher::new("Michel Foucault", 0, 4),
      ];
  
      let handles: Vec<_> = philosophers.into_iter().map(|p| {
          let table = table.clone();
  
          thread::spawn(move || {
              p.eat(&table);
          })
      }).collect();
  
      for h in handles {
          h.join().unwrap();
      }
  }

Lots of changes! However, with this iteration, we’ve got a working program. Let’s go over the details:

use std::sync::{Mutex, Arc};

We’re going to use another structure from the std::sync package: Arc<T>. We’ll talk more about it when we use it.

struct Philosopher {
      name: String,
      left: usize,
      right: usize,
  }

We need to add two more fields to our Philosopher. Each philosopher is going to have two forks: the one on their left, and the one on their right. We’ll use the usize type to indicate them, as it’s the type that you index vectors with. These two values will be the indexes into the forks our Table has.

fn new(name: &str, left: usize, right: usize) -> Philosopher {
      Philosopher {
          name: name.to_string(),
          left: left,
          right: right,
      }
  }

We now need to construct those left and right values, so we add them to new().

fn eat(&self, table: &Table) {
      let _left = table.forks[self.left].lock().unwrap();
      let _right = table.forks[self.right].lock().unwrap();
  
      println!("{} is eating.", self.name);
  
      thread::sleep_ms(1000);
  
      println!("{} is done eating.", self.name);
  }

We have two new lines. We’ve also added an argument, table. We access the Table’s list of forks, and then use self.left and self.right to access the fork at that particular index. That gives us access to the Mutex at that index, and we call lock() on it. If the mutex is currently being accessed by someone else, we’ll block until it becomes available.

The call to lock() might fail, and if it does, we want to crash. In this case, the error that could happen is that the mutex is ‘poisoned’, which is what happens when the thread panics while the lock is held. Since this shouldn’t happen, we just use unwrap().

One other odd thing about these lines: we’ve named the results _left and _right. What’s up with that underscore? Well, we aren’t planning on using the value inside the lock. We just want to acquire it. As such, Rust will warn us that we never use the value. By using the underscore, we tell Rust that this is what we intended, and it won’t throw a warning.

What about releasing the lock? Well, that will happen when _left and _right go out of scope, automatically.

    let table = Arc::new(Table { forks: vec![
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
          Mutex::new(()),
      ]});

Next, in main(), we make a new Table and wrap it in an Arc<T>. ‘arc’ stands for ‘atomic reference count’, and we need that to share our Table across multiple threads. As we share it, the reference count will go up, and when each thread ends, it will go back down.

let philosophers = vec![
      Philosopher::new("Judith Butler", 0, 1),
      Philosopher::new("Gilles Deleuze", 1, 2),
      Philosopher::new("Karl Marx", 2, 3),
      Philosopher::new("Emma Goldman", 3, 4),
      Philosopher::new("Michel Foucault", 0, 4),
  ];

We need to pass in our left and right values to the constructors for our Philosophers. But there’s one more detail here, and it’s very important. If you look at the pattern, it’s all consistent until the very end. Monsieur Foucault should have 4, 0 as arguments, but instead, has 0, 4. This is what prevents deadlock, actually: one of our philosophers is left handed! This is one way to solve the problem, and in my opinion, it’s the simplest.

let handles: Vec<_> = philosophers.into_iter().map(|p| {
      let table = table.clone();
  
      thread::spawn(move || {
          p.eat(&table);
      })
  }).collect();

Finally, inside of our map()/collect() loop, we call table.clone(). The clone() method on Arc<T> is what bumps up the reference count, and when it goes out of scope, it decrements the count. This is needed so that we know how many references to table exist across our threads. If we didn’t have a count, we wouldn’t know how to deallocate it.

You’ll notice we can introduce a new binding to table here, and it will shadow the old one. This is often used so that you don’t need to come up with two unique names.

With this, our program works! Only two philosophers can eat at any one time, and so you’ll get some output like this:

Gilles Deleuze is eating.
  Emma Goldman is eating.
  Emma Goldman is done eating.
  Gilles Deleuze is done eating.
  Judith Butler is eating.
  Karl Marx is eating.
  Judith Butler is done eating.
  Michel Foucault is eating.
  Karl Marx is done eating.
  Michel Foucault is done eating.

Congrats! You’ve implemented a classic concurrency problem in Rust.

Rust Inside Other Languages

For our third project, we’re going to choose something that shows off one of Rust’s greatest strengths: a lack of a substantial runtime.

As organizations grow, they increasingly rely on a multitude of programming languages. Different programming languages have different strengths and weaknesses, and a polyglot stack lets you use a particular language where its strengths make sense and a different one where it’s weak.

A very common area where many programming languages are weak is in runtime performance of programs. Often, using a language that is slower, but offers greater programmer productivity, is a worthwhile trade-off. To help mitigate this, they provide a way to write some of your system in C and then call that C code as though it were written in the higher-level language. This is called a ‘foreign function interface’, often shortened to ‘FFI’.

Rust has support for FFI in both directions: it can call into C code easily, but crucially, it can also be called into as easily as C. Combined with Rust’s lack of a garbage collector and low runtime requirements, this makes Rust a great candidate to embed inside of other languages when you need that extra oomph.

There is a whole chapter devoted to FFI and its specifics elsewhere in the book, but in this chapter, we’ll examine this particular use-case of FFI, with examples in Ruby, Python, and JavaScript.

The problem

There are many different projects we could choose here, but we’re going to pick an example where Rust has a clear advantage over many other languages: numeric computing and threading.

Many languages, for the sake of consistency, place numbers on the heap, rather than on the stack. Especially in languages that focus on object-oriented programming and use garbage collection, heap allocation is the default. Sometimes optimizations can stack allocate particular numbers, but rather than relying on an optimizer to do its job, we may want to ensure that we’re always using primitive number types rather than some sort of object type.

Second, many languages have a ‘global interpreter lock’ (GIL), which limits concurrency in many situations. This is done in the name of safety, which is a positive effect, but it limits the amount of work that can be done at the same time, which is a big negative.

To emphasize these two aspects, we’re going to create a little project that uses these two aspects heavily. Since the focus of the example is to embed Rust into other languages, rather than the problem itself, we’ll just use a toy example:

Start ten threads. Inside each thread, count from one to five million. After all ten threads are finished, print out ‘done!’.

I chose five million based on my particular computer. Here’s an example of this code in Ruby:

threads = []
  
  10.times do
    threads << Thread.new do
      count = 0
  
      5_000_000.times do
        count += 1
      end
  
      count
    end
  end
  
  threads.each do |t|
    puts "Thread finished with count=#{t.value}"
  end
  puts "done!"

Try running this example, and choose a number that runs for a few seconds. Depending on your computer’s hardware, you may have to increase or decrease the number.

On my system, running this program takes 2.156 seconds. And, if I use some sort of process monitoring tool, like top, I can see that it only uses one core on my machine. That’s the GIL kicking in.

While it’s true that this is a synthetic program, one can imagine many problems that are similar to this in the real world. For our purposes, spinning up a few busy threads represents some sort of parallel, expensive computation.

A Rust library

Let’s rewrite this problem in Rust. First, let’s make a new project with Cargo:

$ cargo new embed
  $ cd embed

This program is fairly easy to write in Rust:

use std::thread;
  
  fn process() {
      let handles: Vec<_> = (0..10).map(|_| {
          thread::spawn(|| {
              let mut x = 0;
              for _ in (0..5_000_000) {
                  x += 1
              }
              x
          })
      }).collect();
  
      for h in handles {
          println!("Thread finished with count={}",
          h.join().map_err(|_| "Could not join a thread!").unwrap());
      }
      println!("done!");
  }

Some of this should look familiar from previous examples. We spin up ten threads, collecting them into a handles vector. Inside of each thread, we loop five million times, and add one to x each time. Finally, we join on each thread.

Right now, however, this is a Rust library, and it doesn’t expose anything that’s callable from C. If we tried to hook this up to another language right now, it wouldn’t work. We only need to make two small changes to fix this, though. The first is to modify the beginning of our code:

#[no_mangle]
  pub extern fn process() {

We have to add a new attribute, no_mangle. When you create a Rust library, it changes the name of the function in the compiled output. The reasons for this are outside the scope of this tutorial, but in order for other languages to know how to call the function, we can’t do that. This attribute turns that behavior off.

The other change is the pub extern. The pub means that this function should be callable from outside of this module, and the extern says that it should be able to be called from C. That’s it! Not a whole lot of change.

The second thing we need to do is to change a setting in our Cargo.toml. Add this at the bottom:

[lib]
  name = "embed"
  crate-type = ["dylib"]

This tells Rust that we want to compile our library into a standard dynamic library. By default, Rust compiles an ‘rlib’, a Rust-specific format.

Let’s build the project now:

$ cargo build --release
     Compiling embed v0.1.0 (file:///home/steve/src/embed)

We’ve chosen cargo build --release, which builds with optimizations on. We want this to be as fast as possible! You can find the output of the library in target/release:

$ ls target/release/
  build  deps  examples  libembed.so  native

That libembed.so is our ‘shared object’ library. We can use this file just like any shared object library written in C! As an aside, this may be embed.dll or libembed.dylib, depending on the platform.

Now that we’ve got our Rust library built, let’s use it from our Ruby.

Ruby

Open up an embed.rb file inside of our project, and do this:

require 'ffi'
  
  module Hello
    extend FFI::Library
    ffi_lib 'target/release/libembed.so'
    attach_function :process, [], :void
  end
  
  Hello.process
  
  puts 'done!'

Before we can run this, we need to install the ffi gem:

$ gem install ffi # this may need sudo
  Fetching: ffi-1.9.8.gem (100%)
  Building native extensions.  This could take a while...
  Successfully installed ffi-1.9.8
  Parsing documentation for ffi-1.9.8
  Installing ri documentation for ffi-1.9.8
  Done installing documentation for ffi after 0 seconds
  1 gem installed

And finally, we can try running it:

$ ruby embed.rb
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  Thread finished with count=5000000
  done!
  done!
  $

Whoa, that was fast! On my system, this took 0.086 seconds, rather than the two seconds the pure Ruby version took. Let’s break down this Ruby code:

require 'ffi'

We first need to require the ffi gem. This lets us interface with our Rust library like a C library.

module Hello
    extend FFI::Library
    ffi_lib 'target/release/libembed.so'

The Hello module is used to attach the native functions from the shared library. Inside, we extend the necessary FFI::Library module and then call ffi_lib to load up our shared object library. We just pass it the path that our library is stored, which, as we saw before, is target/release/libembed.so.

attach_function :process, [], :void

The attach_function method is provided by the FFI gem. It’s what connects our process() function in Rust to a Ruby function of the same name. Since process() takes no arguments, the second parameter is an empty array, and since it returns nothing, we pass :void as the final argument.

Hello.process

This is the actual call into Rust. The combination of our module and the call to attach_function sets this all up. It looks like a Ruby function but is actually Rust!

puts 'done!'

Finally, as per our project’s requirements, we print out done!.

That’s it! As we’ve seen, bridging between the two languages is really easy, and buys us a lot of performance.

Next, let’s try Python!

Python

Create an embed.py file in this directory, and put this in it:

from ctypes import cdll
  
  lib = cdll.LoadLibrary("target/release/libembed.so")
  
  lib.process()
  
  print("done!")

Even easier! We use cdll from the ctypes module. A quick call to LoadLibrary later, and we can call process().

On my system, this takes 0.017 seconds. Speedy!

Node.js

Node isn’t a language, but it’s currently the dominant implementation of server-side JavaScript.

In order to do FFI with Node, we first need to install the library:

$ npm install ffi

After that installs, we can use it:

var ffi = require('ffi');
  
  var lib = ffi.Library('target/release/libembed', {
    'process': ['void', []]
  });
  
  lib.process();
  
  console.log("done!");

It looks more like the Ruby example than the Python example. We use the ffi module to get access to ffi.Library(), which loads up our shared object. We need to annotate the return type and argument types of the function, which are void for return and an empty array to signify no arguments. From there, we just call it and print the result.

On my system, this takes a quick 0.092 seconds.

Conclusion

As you can see, the basics of doing this are very easy. Of course, there’s a lot more that we could do here. Check out the FFI chapter for more details.

Effective Rust

So you’ve learned how to write some Rust code. But there’s a difference between writing any Rust code and writing good Rust code.

This section consists of relatively independent tutorials which show you how to take your Rust to the next level. Common patterns and standard library features will be introduced. Read these sections in any order of your choosing.

The Stack and the Heap

As a systems language, Rust operates at a low level. If you’re coming from a high-level language, there are some aspects of systems programming that you may not be familiar with. The most important one is how memory works, with a stack and a heap. If you’re familiar with how C-like languages use stack allocation, this chapter will be a refresher. If you’re not, you’ll learn about this more general concept, but with a Rust-y focus.

Memory management

These two terms are about memory management. The stack and the heap are abstractions that help you determine when to allocate and deallocate memory.

Here’s a high-level comparison:

The stack is very fast, and is where memory is allocated in Rust by default. But the allocation is local to a function call, and is limited in size. The heap, on the other hand, is slower, and is explicitly allocated by your program. But it’s effectively unlimited in size, and is globally accessible.

The Stack

Let’s talk about this Rust program:

fn main() {
      let x = 42;
  }

This program has one variable binding, x. This memory needs to be allocated from somewhere. Rust ‘stack allocates’ by default, which means that basic values ‘go on the stack’. What does that mean?

Well, when a function gets called, some memory gets allocated for all of its local variables and some other information. This is called a ‘stack frame’, and for the purpose of this tutorial, we’re going to ignore the extra information and just consider the local variables we’re allocating. So in this case, when main() is run, we’ll allocate a single 32-bit integer for our stack frame. This is automatically handled for you, as you can see; we didn’t have to write any special Rust code or anything.

When the function is over, its stack frame gets deallocated. This happens automatically, we didn’t have to do anything special here.

That’s all there is for this simple program. The key thing to understand here is that stack allocation is very, very fast. Since we know all the local variables we have ahead of time, we can grab the memory all at once. And since we’ll throw them all away at the same time as well, we can get rid of it very fast too.

The downside is that we can’t keep values around if we need them for longer than a single function. We also haven’t talked about what the word, ‘stack’, means. To do that, we need a slightly more complicated example:

fn foo() {
      let y = 5;
      let z = 100;
  }
  
  fn main() {
      let x = 42;
  
      foo();
  }

This program has three variables total: two in foo(), one in main(). Just as before, when main() is called, a single integer is allocated for its stack frame. But before we can show what happens when foo() is called, we need to visualize what’s going on with memory. Your operating system presents a view of memory to your program that’s pretty simple: a huge list of addresses, from 0 to a large number, representing how much RAM your computer has. For example, if you have a gigabyte of RAM, your addresses go from 0 to 1,073,741,823. That number comes from 230, the number of bytes in a gigabyte.

This memory is kind of like a giant array: addresses start at zero and go up to the final number. So here’s a diagram of our first stack frame:

Address Name Value
0 x 42

We’ve got x located at address 0, with the value 42.

When foo() is called, a new stack frame is allocated:

Address Name Value
2 z 100
1 y 5
0 x 42

Because 0 was taken by the first frame, 1 and 2 are used for foo()’s stack frame. It grows upward, the more functions we call.

There’s some important things we have to take note of here. The numbers 0, 1, and 2 are all solely for illustrative purposes, and bear no relationship to the actual numbers the computer will actually use. In particular, the series of addresses are in reality going to be separated by some number of bytes that separate each address, and that separation may even exceed the size of the value being stored.

After foo() is over, its frame is deallocated:

Address Name Value
0 x 42

And then, after main(), even this last value goes away. Easy!

It’s called a ‘stack’ because it works like a stack of dinner plates: the first plate you put down is the last plate to pick back up. Stacks are sometimes called ‘last in, first out queues’ for this reason, as the last value you put on the stack is the first one you retrieve from it.

Let’s try a three-deep example:

fn bar() {
      let i = 6;
  }
  
  fn foo() {
      let a = 5;
      let b = 100;
      let c = 1;
  
      bar();
  }
  
  fn main() {
      let x = 42;
  
      foo();
  }

Okay, first, we call main():

Address Name Value
0 x 42

Next up, main() calls foo():

Address Name Value
3 c 1
2 b 100
1 a 5
0 x 42

And then foo() calls bar():

Address Name Value
4 i 6
3 c 1
2 b 100
1 a 5
0 x 42

Whew! Our stack is growing tall.

After bar() is over, its frame is deallocated, leaving just foo() and main():

Address Name Value
3 c 1
2 b 100
1 a 5
0 x 42

And then foo() ends, leaving just main():

Address Name Value
0 x 42

And then we’re done. Getting the hang of it? It’s like piling up dishes: you add to the top, you take away from the top.

The Heap

Now, this works pretty well, but not everything can work like this. Sometimes, you need to pass some memory between different functions, or keep it alive for longer than a single function’s execution. For this, we can use the heap.

In Rust, you can allocate memory on the heap with the Box<T> type. Here’s an example:

fn main() {
      let x = Box::new(5);
      let y = 42;
  }

Here’s what happens in memory when main() is called:

Address Name Value
1 y 42
0 x ??????

We allocate space for two variables on the stack. y is 42, as it always has been, but what about x? Well, x is a Box<i32>, and boxes allocate memory on the heap. The actual value of the box is a structure which has a pointer to ‘the heap’. When we start executing the function, and Box::new() is called, it allocates some memory for the heap, and puts 5 there. The memory now looks like this:

Address Name Value
230 5
1 y 42
0 x → 230

We have 230 in our hypothetical computer with 1GB of RAM. And since our stack grows from zero, the easiest place to allocate memory is from the other end. So our first value is at the highest place in memory. And the value of the struct at x has a raw pointer to the place we’ve allocated on the heap, so the value of x is 230, the memory location we’ve asked for.

We haven’t really talked too much about what it actually means to allocate and deallocate memory in these contexts. Getting into very deep detail is out of the scope of this tutorial, but what’s important to point out here is that the heap isn’t just a stack that grows from the opposite end. We’ll have an example of this later in the book, but because the heap can be allocated and freed in any order, it can end up with ‘holes’. Here’s a diagram of the memory layout of a program which has been running for a while now:

Address Name Value
230 5
(230) - 1
(230) - 2
(230) - 3 42
3 y → (230) - 3
2 y 42
1 y 42
0 x → 230

In this case, we’ve allocated four things on the heap, but deallocated two of them. There’s a gap between 230 and (230) - 3 which isn’t currently being used. The specific details of how and why this happens depends on what kind of strategy you use to manage the heap. Different programs can use different ‘memory allocators’, which are libraries that manage this for you. Rust programs use jemalloc for this purpose.

Anyway, back to our example. Since this memory is on the heap, it can stay alive longer than the function which allocates the box. In this case, however, it doesn’t.1 When the function is over, we need to free the stack frame for main(). Box<T>, though, has a trick up its sleeve: Drop. The implementation of Drop for Box deallocates the memory that was allocated when it was created. Great! So when x goes away, it first frees the memory allocated on the heap:

Address Name Value
1 y 42
0 x ??????

And then the stack frame goes away, freeing all of our memory.

Arguments and borrowing

We’ve got some basic examples with the stack and the heap going, but what about function arguments and borrowing? Here’s a small Rust program:

fn foo(i: &i32) {
      let z = 42;
  }
  
  fn main() {
      let x = 5;
      let y = &x;
  
      foo(y);
  }

When we enter main(), memory looks like this:

Address Name Value
1 y → 0
0 x 5

x is a plain old 5, and y is a reference to x. So its value is the memory location that x lives at, which in this case is 0.

What about when we call foo(), passing y as an argument?

Address Name Value
3 z 42
2 i → 0
1 y → 0
0 x 5

Stack frames aren’t just for local bindings, they’re for arguments too. So in this case, we need to have both i, our argument, and z, our local variable binding. i is a copy of the argument, y. Since y’s value is 0, so is i’s.

This is one reason why borrowing a variable doesn’t deallocate any memory: the value of a reference is just a pointer to a memory location. If we got rid of the underlying memory, things wouldn’t work very well.

A complex example

Okay, let’s go through this complex program step-by-step:

fn foo(x: &i32) {
      let y = 10;
      let z = &y;
  
      baz(z);
      bar(x, z);
  }
  
  fn bar(a: &i32, b: &i32) {
      let c = 5;
      let d = Box::new(5);
      let e = &d;
  
      baz(e);
  }
  
  fn baz(f: &i32) {
      let g = 100;
  }
  
  fn main() {
      let h = 3;
      let i = Box::new(20);
      let j = &h;
  
      foo(j);
  }

First, we call main():

Address Name Value
230 20
2 j → 0
1 i → 230
0 h 3

We allocate memory for j, i, and h. i is on the heap, and so has a value pointing there.

Next, at the end of main(), foo() gets called:

Address Name Value
230 20
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

Space gets allocated for x, y, and z. The argument x has the same value as j, since that’s what we passed it in. It’s a pointer to the 0 address, since j points at h.

Next, foo() calls baz(), passing z:

Address Name Value
230 20
7 g 100
6 f → 4
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

We’ve allocated memory for f and g. baz() is very short, so when it’s over, we get rid of its stack frame:

Address Name Value
230 20
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

Next, foo() calls bar() with x and z:

Address Name Value
230 20
(230) - 1 5
10 e → 9
9 d → (230) - 1
8 c 5
7 b → 4
6 a → 0
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

We end up allocating another value on the heap, and so we have to subtract one from 230. It’s easier to just write that than 1,073,741,823. In any case, we set up the variables as usual.

At the end of bar(), it calls baz():

Address Name Value
230 20
(230) - 1 5
12 g 100
11 f → 9
10 e → 9
9 d → (230) - 1
8 c 5
7 b → 4
6 a → 0
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

With this, we’re at our deepest point! Whew! Congrats for following along this far.

After baz() is over, we get rid of f and g:

Address Name Value
230 20
(230) - 1 5
10 e → 9
9 d → (230) - 1
8 c 5
7 b → 4
6 a → 0
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

Next, we return from bar(). d in this case is a Box<T>, so it also frees what it points to: (230) - 1.

Address Name Value
230 20
5 z → 4
4 y 10
3 x → 0
2 j → 0
1 i → 230
0 h 3

And after that, foo() returns:

Address Name Value
230 20
2 j → 0
1 i → 230
0 h 3

And then, finally, main(), which cleans the rest up. When i is Dropped, it will clean up the last of the heap too.

What do other languages do?

Most languages with a garbage collector heap-allocate by default. This means that every value is boxed. There are a number of reasons why this is done, but they’re out of scope for this tutorial. There are some possible optimizations that don’t make it true 100% of the time, too. Rather than relying on the stack and Drop to clean up memory, the garbage collector deals with the heap instead.

Which to use?

So if the stack is faster and easier to manage, why do we need the heap? A big reason is that Stack-allocation alone means you only have LIFO semantics for reclaiming storage. Heap-allocation is strictly more general, allowing storage to be taken from and returned to the pool in arbitrary order, but at a complexity cost.

Generally, you should prefer stack allocation, and so, Rust stack-allocates by default. The LIFO model of the stack is simpler, at a fundamental level. This has two big impacts: runtime efficiency and semantic impact.

Runtime Efficiency

Managing the memory for the stack is trivial: The machine just increments or decrements a single value, the so-called “stack pointer”. Managing memory for the heap is non-trivial: heap-allocated memory is freed at arbitrary points, and each block of heap-allocated memory can be of arbitrary size, the memory manager must generally work much harder to identify memory for reuse.

If you’d like to dive into this topic in greater detail, this paper is a great introduction.

Semantic impact

Stack-allocation impacts the Rust language itself, and thus the developer’s mental model. The LIFO semantics is what drives how the Rust language handles automatic memory management. Even the deallocation of a uniquely-owned heap-allocated box can be driven by the stack-based LIFO semantics, as discussed throughout this chapter. The flexibility (i.e. expressiveness) of non LIFO-semantics means that in general the compiler cannot automatically infer at compile-time where memory should be freed; it has to rely on dynamic protocols, potentially from outside the language itself, to drive deallocation (reference counting, as used by Rc<T> and Arc<T>, is one example of this).

When taken to the extreme, the increased expressive power of heap allocation comes at the cost of either significant runtime support (e.g. in the form of a garbage collector) or significant programmer effort (in the form of explicit memory management calls that require verification not provided by the Rust compiler).

Testing

Program testing can be a very effective way to show the presence of bugs, but it is hopelessly inadequate for showing their absence.

Edsger W. Dijkstra, “The Humble Programmer” (1972)

Let’s talk about how to test Rust code. What we will not be talking about is the right way to test Rust code. There are many schools of thought regarding the right and wrong way to write tests. All of these approaches use the same basic tools, and so we’ll show you the syntax for using them.

The test attribute

At its simplest, a test in Rust is a function that’s annotated with the test attribute. Let’s make a new project with Cargo called adder:

$ cargo new adder
  $ cd adder

Cargo will automatically generate a simple test when you make a new project. Here’s the contents of src/lib.rs:

#[test]
  fn it_works() {
  }

Note the #[test]. This attribute indicates that this is a test function. It currently has no body. That’s good enough to pass! We can run the tests with cargo test:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

Cargo compiled and ran our tests. There are two sets of output here: one for the test we wrote, and another for documentation tests. We’ll talk about those later. For now, see this line:

test it_works ... ok

Note the it_works. This comes from the name of our function:

fn it_works() {

We also get a summary line:

test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured

So why does our do-nothing test pass? Any test which doesn’t panic! passes, and any test that does panic! fails. Let’s make our test fail:

#[test]
  fn it_works() {
      assert!(false);
  }

assert! is a macro provided by Rust which takes one argument: if the argument is true, nothing happens. If the argument is false, it panic!s. Let’s run our tests again:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test it_works ... FAILED
  
  failures:
  
  ---- it_works stdout ----
          thread 'it_works' panicked at 'assertion failed: false', /home/steve/tmp/adder
  ↳ /src/lib.rs:3
  
  
  
  failures:
      it_works
  
  test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured
  
  thread '<main>' panicked at 'Some tests failed', /home/steve/src/rust/src/libtest/lib.
  ↳ rs:247

Rust indicates that our test failed:

test it_works ... FAILED

And that’s reflected in the summary line:

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured

We also get a non-zero status code. We can use $? on OS X and Linux:

$ echo $?
  101

On Windows, if you’re using cmd:

> echo %ERRORLEVEL%

And if you’re using PowerShell:

> echo $LASTEXITCODE # the code itself
  > echo $? # a boolean, fail or succeed

This is useful if you want to integrate cargo test into other tooling.

We can invert our test’s failure with another attribute: should_panic:

#[test]
  #[should_panic]
  fn it_works() {
      assert!(false);
  }

This test will now succeed if we panic! and fail if we complete. Let’s try it:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

Rust provides another macro, assert_eq!, that compares two arguments for equality:

#[test]
  #[should_panic]
  fn it_works() {
      assert_eq!("Hello", "world");
  }

Does this test pass or fail? Because of the should_panic attribute, it passes:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

should_panic tests can be fragile, as it’s hard to guarantee that the test didn’t fail for an unexpected reason. To help with this, an optional expected parameter can be added to the should_panic attribute. The test harness will make sure that the failure message contains the provided text. A safer version of the example above would be:

#[test]
  #[should_panic(expected = "assertion failed")]
  fn it_works() {
      assert_eq!("Hello", "world");
  }

That’s all there is to the basics! Let’s write one ‘real’ test:

pub fn add_two(a: i32) -> i32 {
      a + 2
  }
  
  #[test]
  fn it_works() {
      assert_eq!(4, add_two(2));
  }

This is a very common use of assert_eq!: call some function with some known arguments and compare it to the expected output.

The ignore attribute

Sometimes a few specific tests can be very time-consuming to execute. These can be disabled by default by using the ignore attribute:

#[test]
  fn it_works() {
      assert_eq!(4, add_two(2));
  }
  
  #[test]
  #[ignore]
  fn expensive_test() {
      // code that takes an hour to run
  }

Now we run our tests and see that it_works is run, but expensive_test is not:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 2 tests
  test expensive_test ... ignored
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 1 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

The expensive tests can be run explicitly using cargo test -- --ignored:

$ cargo test -- --ignored
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test expensive_test ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

The --ignored argument is an argument to the test binary, and not to cargo, which is why the command is cargo test -- --ignored.

The tests module

There is one way in which our existing example is not idiomatic: it’s missing the tests module. The idiomatic way of writing our example looks like this:

pub fn add_two(a: i32) -> i32 {
      a + 2
  }
  
  #[cfg(test)]
  mod tests {
      use super::add_two;
  
      #[test]
      fn it_works() {
          assert_eq!(4, add_two(2));
      }
  }

There’s a few changes here. The first is the introduction of a mod tests with a cfg attribute. The module allows us to group all of our tests together, and to also define helper functions if needed, that don’t become a part of the rest of our crate. The cfg attribute only compiles our test code if we’re currently trying to run the tests. This can save compile time, and also ensures that our tests are entirely left out of a normal build.

The second change is the use declaration. Because we’re in an inner module, we need to bring our test function into scope. This can be annoying if you have a large module, and so this is a common use of globs. Let’s change our src/lib.rs to make use of it:

pub fn add_two(a: i32) -> i32 {
      a + 2
  }
  
  #[cfg(test)]
  mod tests {
      use super::*;
  
      #[test]
      fn it_works() {
          assert_eq!(4, add_two(2));
      }
  }

Note the different use line. Now we run our tests:

$ cargo test
      Updating registry `https://github.com/rust-lang/crates.io-index`
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test tests::it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

It works!

The current convention is to use the tests module to hold your “unit-style” tests. Anything that just tests one small bit of functionality makes sense to go here. But what about “integration-style” tests instead? For that, we have the tests directory

The tests directory

To write an integration test, let’s make a tests directory, and put a tests/lib.rs file inside, with this as its contents:

extern crate adder;
  
  #[test]
  fn it_works() {
      assert_eq!(4, adder::add_two(2));
  }

This looks similar to our previous tests, but slightly different. We now have an extern crate adder at the top. This is because the tests in the tests directory are an entirely separate crate, and so we need to import our library. This is also why tests is a suitable place to write integration-style tests: they use the library like any other consumer of it would.

Let’s run them:

$ cargo test
     Compiling adder v0.0.1 (file:///home/you/projects/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test tests::it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
       Running target/lib-c18e7d3494509e74
  
  running 1 test
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 0 tests
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 0 measured

Now we have three sections: our previous test is also run, as well as our new one.

That’s all there is to the tests directory. The tests module isn’t needed here, since the whole thing is focused on tests.

Let’s finally check out that third section: documentation tests.

Documentation tests

Nothing is better than documentation with examples. Nothing is worse than examples that don’t actually work, because the code has changed since the documentation has been written. To this end, Rust supports automatically running examples in your documentation (note: this only works in library crates, not binary crates). Here’s a fleshed-out src/lib.rs with examples:

//! The `adder` crate provides functions that add numbers to other numbers.
  //!
  //! # Examples
  //!
  //! ```
  //! assert_eq!(4, adder::add_two(2));
  //! ```
  
  /// This function adds two to its argument.
  ///
  /// # Examples
  ///
  /// ```
  /// use adder::add_two;
  ///
  /// assert_eq!(4, add_two(2));
  /// ```
  pub fn add_two(a: i32) -> i32 {
      a + 2
  }
  
  #[cfg(test)]
  mod tests {
      use super::*;
  
      #[test]
      fn it_works() {
          assert_eq!(4, add_two(2));
      }
  }

Note the module-level documentation with //! and the function-level documentation with ///. Rust’s documentation supports Markdown in comments, and so triple graves mark code blocks. It is conventional to include the # Examples section, exactly like that, with examples following.

Let’s run the tests again:

$ cargo test
     Compiling adder v0.0.1 (file:///home/steve/tmp/adder)
       Running target/adder-91b3e234d4ed382a
  
  running 1 test
  test tests::it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
       Running target/lib-c18e7d3494509e74
  
  running 1 test
  test it_works ... ok
  
  test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured
  
     Doc-tests adder
  
  running 2 tests
  test add_two_0 ... ok
  test _0 ... ok
  
  test result: ok. 2 passed; 0 failed; 0 ignored; 0 measured

Now we have all three kinds of tests running! Note the names of the documentation tests: the _0 is generated for the module test, and add_two_0 for the function test. These will auto increment with names like add_two_1 as you add more examples.

Conditional Compilation

Rust has a special attribute, #[cfg], which allows you to compile code based on a flag passed to the compiler. It has two forms:

#[cfg(foo)]
  
  #[cfg(bar = "baz")]

They also have some helpers:

#[cfg(any(unix, windows))]
  
  #[cfg(all(unix, target_pointer_width = "32"))]
  
  #[cfg(not(foo))]

These can nest arbitrarily:

#[cfg(any(not(unix), all(target_os="macos", target_arch = "powerpc")))]

As for how to enable or disable these switches, if you’re using Cargo, they get set in the [features] section of your Cargo.toml:

[features]
  # no features by default
  default = []
  
  # The “secure-password” feature depends on the bcrypt package.
  secure-password = ["bcrypt"]

When you do this, Cargo passes along a flag to rustc:

--cfg feature="${feature_name}"

The sum of these cfg flags will determine which ones get activated, and therefore, which code gets compiled. Let’s take this code:

#[cfg(feature = "foo")]
  mod foo {
  }

If we compile it with cargo build --features "foo", it will send the --cfg feature="foo" flag to rustc, and the output will have the mod foo in it. If we compile it with a regular cargo build, no extra flags get passed on, and so, no foo module will exist.

cfg_attr

You can also set another attribute based on a cfg variable with cfg_attr:

#[cfg_attr(a, b)]

Will be the same as #[b] if a is set by cfg attribute, and nothing otherwise.

cfg!

The cfg! syntax extension lets you use these kinds of flags elsewhere in your code, too:

if cfg!(target_os = "macos") || cfg!(target_os = "ios") {
      println!("Think Different!");
  }

These will be replaced by a true or false at compile-time, depending on the configuration settings.

Documentation

Documentation is an important part of any software project, and it’s first-class in Rust. Let’s talk about the tooling Rust gives you to document your project.

About rustdoc

The Rust distribution includes a tool, rustdoc, that generates documentation. rustdoc is also used by Cargo through cargo doc.

Documentation can be generated in two ways: from source code, and from standalone Markdown files.

Documenting source code

The primary way of documenting a Rust project is through annotating the source code. You can use documentation comments for this purpose:

/// Constructs a new `Rc<T>`.
  ///
  /// # Examples
  ///
  /// ```
  /// use std::rc::Rc;
  ///
  /// let five = Rc::new(5);
  /// ```
  pub fn new(value: T) -> Rc<T> {
      // implementation goes here
  }

This code generates documentation that looks like this. I’ve left the implementation out, with a regular comment in its place.

The first thing to notice about this annotation is that it uses /// instead of //. The triple slash indicates a documentation comment.

Documentation comments are written in Markdown.

Rust keeps track of these comments, and uses them when generating documentation. This is important when documenting things like enums:

/// The `Option` type. See [the module level documentation](../) for more.
  enum Option<T> {
      /// No value
      None,
      /// Some value `T`
      Some(T),
  }

The above works, but this does not:

/// The `Option` type. See [the module level documentation](../) for more.
  enum Option<T> {
      None, /// No value
      Some(T), /// Some value `T`
  }

You’ll get an error:

hello.rs:4:1: 4:2 error: expected ident, found `}`
  hello.rs:4 }
             ^

This unfortunate error is correct: documentation comments apply to the thing after them, and there’s nothing after that last comment.

Writing documentation comments

Anyway, let’s cover each part of this comment in detail:

/// Constructs a new `Rc<T>`.

The first line of a documentation comment should be a short summary of its functionality. One sentence. Just the basics. High level.

///
  /// Other details about constructing `Rc<T>`s, maybe describing complicated
  /// semantics, maybe additional options, all kinds of stuff.
  ///

Our original example had just a summary line, but if we had more things to say, we could have added more explanation in a new paragraph.

Special sections

Next, are special sections. These are indicated with a header, #. There are four kinds of headers that are commonly used. They aren’t special syntax, just convention, for now.

/// # Panics

Unrecoverable misuses of a function (i.e. programming errors) in Rust are usually indicated by panics, which kill the whole current thread at the very least. If your function has a non-trivial contract like this, that is detected/enforced by panics, documenting it is very important.

/// # Failures

If your function or method returns a Result<T, E>, then describing the conditions under which it returns Err(E) is a nice thing to do. This is slightly less important than Panics, because failure is encoded into the type system, but it’s still a good thing to do.

/// # Safety

If your function is unsafe, you should explain which invariants the caller is responsible for upholding.

/// # Examples
  ///
  /// ```
  /// use std::rc::Rc;
  ///
  /// let five = Rc::new(5);
  /// ```

Fourth, Examples. Include one or more examples of using your function or method, and your users will love you for it. These examples go inside of code block annotations, which we’ll talk about in a moment, and can have more than one section:

/// # Examples
  ///
  /// Simple `&str` patterns:
  ///
  /// ```
  /// let v: Vec<&str> = "Mary had a little lamb".split(' ').collect();
  /// assert_eq!(v, vec!["Mary", "had", "a", "little", "lamb"]);
  /// ```
  ///
  /// More complex patterns with a lambda:
  ///
  /// ```
  /// let v: Vec<&str> = "abc1def2ghi".split(|c: char| c.is_numeric()).collect();
  /// assert_eq!(v, vec!["abc", "def", "ghi"]);
  /// ```

Let’s discuss the details of these code blocks.

Code block annotations

To write some Rust code in a comment, use the triple graves:

/// ```
  /// println!("Hello, world");
  /// ```

If you want something that’s not Rust code, you can add an annotation:

/// ```c
  /// printf("Hello, world\n");
  /// ```

This will highlight according to whatever language you’re showing off. If you’re just showing plain text, choose text.

It’s important to choose the correct annotation here, because rustdoc uses it in an interesting way: It can be used to actually test your examples in a library crate, so that they don’t get out of date. If you have some C code but rustdoc thinks it’s Rust because you left off the annotation, rustdoc will complain when trying to generate the documentation.

Documentation as tests

Let’s discuss our sample example documentation:

/// ```
  /// println!("Hello, world");
  /// ```

You’ll notice that you don’t need a fn main() or anything here. rustdoc will automatically add a main() wrapper around your code, and in the right place. For example:

/// ```
  /// use std::rc::Rc;
  ///
  /// let five = Rc::new(5);
  /// ```

This will end up testing:

fn main() {
      use std::rc::Rc;
      let five = Rc::new(5);
  }

Here’s the full algorithm rustdoc uses to postprocess examples:

  1. Any leading #![foo] attributes are left intact as crate attributes.
  2. Some common allow attributes are inserted, including unused_variables, unused_assignments, unused_mut, unused_attributes, and dead_code. Small examples often trigger these lints.
  3. If the example does not contain extern crate, then extern crate <mycrate>; is inserted.
  4. Finally, if the example does not contain fn main, the remainder of the text is wrapped in fn main() { your_code }

Sometimes, this isn’t enough, though. For example, all of these code samples with /// we’ve been talking about? The raw text:

/// Some documentation.
  # fn foo() {}

looks different than the output:

/// Some documentation.

Yes, that’s right: you can add lines that start with #, and they will be hidden from the output, but will be used when compiling your code. You can use this to your advantage. In this case, documentation comments need to apply to some kind of function, so if I want to show you just a documentation comment, I need to add a little function definition below it. At the same time, it’s just there to satisfy the compiler, so hiding it makes the example more clear. You can use this technique to explain longer examples in detail, while still preserving the testability of your documentation. For example, this code:

let x = 5;
  let y = 6;
  println!("{}", x + y);

Here’s an explanation, rendered:

First, we set x to five:

let x = 5;

Next, we set y to six:

let y = 6;

Finally, we print the sum of x and y:

println!("{}", x + y);

Here’s the same explanation, in raw text:

First, we set x to five:

let x = 5;
  # let y = 6;
  # println!("{}", x + y);

Next, we set y to six:

# let x = 5;
  let y = 6;
  # println!("{}", x + y);

Finally, we print the sum of x and y:

# let x = 5;
  # let y = 6;
  println!("{}", x + y);

By repeating all parts of the example, you can ensure that your example still compiles, while only showing the parts that are relevant to that part of your explanation.

Documenting macros

Here’s an example of documenting a macro:

/// Panic with a given message unless an expression evaluates to true.
  ///
  /// # Examples
  ///
  /// ```
  /// # #[macro_use] extern crate foo;
  /// # fn main() {
  /// panic_unless!(1 + 1 == 2, “Math is broken.”);
  /// # }
  /// ```
  ///
  /// ```should_panic
  /// # #[macro_use] extern crate foo;
  /// # fn main() {
  /// panic_unless!(true == false, “I’m broken.”);
  /// # }
  /// ```
  #[macro_export]
  macro_rules! panic_unless {
      ($condition:expr, $($rest:expr),+) => ({ if ! $condition { panic!($($rest),+); } }
  ↳ );
  }

You’ll note three things: we need to add our own extern crate line, so that we can add the #[macro_use] attribute. Second, we’ll need to add our own main() as well. Finally, a judicious use of # to comment out those two things, so they don’t show up in the output.

Running documentation tests

To run the tests, either

$ rustdoc --test path/to/my/crate/root.rs
  # or
  $ cargo test

That’s right, cargo test tests embedded documentation too. However, cargo test will not test binary crates, only library ones. This is due to the way rustdoc works: it links against the library to be tested, but with a binary, there’s nothing to link to.

There are a few more annotations that are useful to help rustdoc do the right thing when testing your code:

/// ```ignore
  /// fn foo() {
  /// ```

The ignore directive tells Rust to ignore your code. This is almost never what you want, as it’s the most generic. Instead, consider annotating it with text if it’s not code, or using #s to get a working example that only shows the part you care about.

/// ```should_panic
  /// assert!(false);
  /// ```

should_panic tells rustdoc that the code should compile correctly, but not actually pass as a test.

/// ```no_run
  /// loop {
  ///     println!("Hello, world");
  /// }
  /// ```

The no_run attribute will compile your code, but not run it. This is important for examples such as “Here’s how to start up a network service,” which you would want to make sure compile, but might run in an infinite loop!

Documenting modules

Rust has another kind of doc comment, //!. This comment doesn’t document the next item, but the enclosing item. In other words:

mod foo {
      //! This is documentation for the `foo` module.
      //!
      //! # Examples
  
      // ...
  }

This is where you’ll see //! used most often: for module documentation. If you have a module in foo.rs, you’ll often open its code and see this:

//! A module for using `foo`s.
  //!
  //! The `foo` module contains a lot of useful functionality blah blah blah
Documentation comment style

Check out RFC 505 for full conventions around the style and format of documentation.

Other documentation

All of this behavior works in non-Rust source files too. Because comments are written in Markdown, they’re often .md files.

When you write documentation in Markdown files, you don’t need to prefix the documentation with comments. For example:

/// # Examples
  ///
  /// ```
  /// use std::rc::Rc;
  ///
  /// let five = Rc::new(5);
  /// ```

is just

### Examples
  
  ```
  use std::rc::Rc;
  
  let five = Rc::new(5);
  ```

when it’s in a Markdown file. There is one wrinkle though: Markdown files need to have a title like this:

% The title
  
  This is the example documentation.

This % line needs to be the very first line of the file.

doc attributes

At a deeper level, documentation comments are sugar for documentation attributes:

/// this
  
  #[doc="this"]

are the same, as are these:

//! this
  
  #![doc="/// this"]

You won’t often see this attribute used for writing documentation, but it can be useful when changing some options, or when writing a macro.

Re-exports

rustdoc will show the documentation for a public re-export in both places:

extern crate foo;
  
  pub use foo::bar;

This will create documentation for bar both inside the documentation for the crate foo, as well as the documentation for your crate. It will use the same documentation in both places.

This behavior can be suppressed with no_inline:

extern crate foo;
  
  #[doc(no_inline)]
  pub use foo::bar;
Controlling HTML

You can control a few aspects of the HTML that rustdoc generates through the #![doc] version of the attribute:

#![doc(html_logo_url = "https://www.rust-lang.org/logos/rust-logo-128x128-blk-v2.png",
         html_favicon_url = "https://www.rust-lang.org/favicon.ico",
         html_root_url = "https://doc.rust-lang.org/")]

This sets a few different options, with a logo, favicon, and a root URL.

Generation options

rustdoc also contains a few other options on the command line, for further customization:

  • --html-in-header FILE: includes the contents of FILE at the end of the <head>...</head> section.
  • --html-before-content FILE: includes the contents of FILE directly after <body>, before the rendered content (including the search bar).
  • --html-after-content FILE: includes the contents of FILE after all the rendered content.

Security note

The Markdown in documentation comments is placed without processing into the final webpage. Be careful with literal HTML:

/// <script>alert(document.cookie)</script>

Iterators

Let’s talk about loops.

Remember Rust’s for loop? Here’s an example:

for x in 0..10 {
      println!("{}", x);
  }

Now that you know more Rust, we can talk in detail about how this works. Ranges (the 0..10) are ‘iterators’. An iterator is something that we can call the .next() method on repeatedly, and it gives us a sequence of things.

Like this:

let mut range = 0..10;
  
  loop {
      match range.next() {
          Some(x) => {
              println!("{}", x);
          },
          None => { break }
      }
  }

We make a mutable binding to the range, which is our iterator. We then loop, with an inner match. This match is used on the result of range.next(), which gives us a reference to the next value of the iterator. next returns an Option<i32>, in this case, which will be Some(i32) when we have a value and None once we run out. If we get Some(i32), we print it out, and if we get None, we break out of the loop.

This code sample is basically the same as our for loop version. The for loop is just a handy way to write this loop/match/break construct.

for loops aren’t the only thing that uses iterators, however. Writing your own iterator involves implementing the Iterator trait. While doing that is outside of the scope of this guide, Rust provides a number of useful iterators to accomplish various tasks. Before we talk about those, we should talk about a Rust anti-pattern. And that’s using ranges like this.

Yes, we just talked about how ranges are cool. But ranges are also very primitive. For example, if you needed to iterate over the contents of a vector, you may be tempted to write this:

let nums = vec![1, 2, 3];
  
  for i in 0..nums.len() {
      println!("{}", nums[i]);
  }

This is strictly worse than using an actual iterator. You can iterate over vectors directly, so write this:

let nums = vec![1, 2, 3];
  
  for num in &nums {
      println!("{}", num);
  }

There are two reasons for this. First, this more directly expresses what we mean. We iterate through the entire vector, rather than iterating through indexes, and then indexing the vector. Second, this version is more efficient: the first version will have extra bounds checking because it used indexing, nums[i]. But since we yield a reference to each element of the vector in turn with the iterator, there’s no bounds checking in the second example. This is very common with iterators: we can ignore unnecessary bounds checks, but still know that we’re safe.

There’s another detail here that’s not 100% clear because of how println! works. num is actually of type &i32. That is, it’s a reference to an i32, not an i32 itself. println! handles the dereferencing for us, so we don’t see it. This code works fine too:

let nums = vec![1, 2, 3];
  
  for num in &nums {
      println!("{}", *num);
  }

Now we’re explicitly dereferencing num. Why does &nums give us references? Firstly, because we explicitly asked it to with &. Secondly, if it gave us the data itself, we would have to be its owner, which would involve making a copy of the data and giving us the copy. With references, we’re just borrowing a reference to the data, and so it’s just passing a reference, without needing to do the move.

So, now that we’ve established that ranges are often not what you want, let’s talk about what you do want instead.

There are three broad classes of things that are relevant here: iterators, iterator adapters, and consumers. Here’s some definitions:

  • iterators give you a sequence of values.
  • iterator adapters operate on an iterator, producing a new iterator with a different output sequence.
  • consumers operate on an iterator, producing some final set of values.

Let’s talk about consumers first, since you’ve already seen an iterator, ranges.

Consumers

A consumer operates on an iterator, returning some kind of value or values. The most common consumer is collect(). This code doesn’t quite compile, but it shows the intention:

let one_to_one_hundred = (1..101).collect();

As you can see, we call collect() on our iterator. collect() takes as many values as the iterator will give it, and returns a collection of the results. So why won’t this compile? Rust can’t determine what type of things you want to collect, and so you need to let it know. Here’s the version that does compile:

let one_to_one_hundred = (1..101).collect::<Vec<i32>>();

If you remember, the ::<> syntax allows us to give a type hint, and so we tell it that we want a vector of integers. You don’t always need to use the whole type, though. Using a _ will let you provide a partial hint:

let one_to_one_hundred = (1..101).collect::<Vec<_>>();

This says “Collect into a Vec<T>, please, but infer what the T is for me.” _ is sometimes called a “type placeholder” for this reason.

collect() is the most common consumer, but there are others too. find() is one:

let greater_than_forty_two = (0..100)
                               .find(|x| *x > 42);
  
  match greater_than_forty_two {
      Some(_) => println!("We got some numbers!"),
      None => println!("No numbers found :("),
  }

find takes a closure, and works on a reference to each element of an iterator. This closure returns true if the element is the element we’re looking for, and false otherwise. Because we might not find a matching element, find returns an Option rather than the element itself.

Another important consumer is fold. Here’s what it looks like:

let sum = (1..4).fold(0, |sum, x| sum + x);

fold() is a consumer that looks like this: fold(base, |accumulator, element| ...). It takes two arguments: the first is an element called the base. The second is a closure that itself takes two arguments: the first is called the accumulator, and the second is an element. Upon each iteration, the closure is called, and the result is the value of the accumulator on the next iteration. On the first iteration, the base is the value of the accumulator.

Okay, that’s a bit confusing. Let’s examine the values of all of these things in this iterator:

base accumulator element closure result
0 0 1 1
0 1 2 3
0 3 3 6

We called fold() with these arguments:

.fold(0, |sum, x| sum + x);

So, 0 is our base, sum is our accumulator, and x is our element. On the first iteration, we set sum to 0, and x is the first element of nums, 1. We then add sum and x, which gives us 0 + 1 = 1. On the second iteration, that value becomes our accumulator, sum, and the element is the second element of the array, 2. 1 + 2 = 3, and so that becomes the value of the accumulator for the last iteration. On that iteration, x is the last element, 3, and 3 + 3 = 6, which is our final result for our sum. 1 + 2 + 3 = 6, and that’s the result we got.

Whew. fold can be a bit strange the first few times you see it, but once it clicks, you can use it all over the place. Any time you have a list of things, and you want a single result, fold is appropriate.

Consumers are important due to one additional property of iterators we haven’t talked about yet: laziness. Let’s talk some more about iterators, and you’ll see why consumers matter.

Iterators

As we’ve said before, an iterator is something that we can call the .next() method on repeatedly, and it gives us a sequence of things. Because you need to call the method, this means that iterators can be lazy and not generate all of the values upfront. This code, for example, does not actually generate the numbers 1-99, instead creating a value that merely represents the sequence:

let nums = 1..100;

Since we didn’t do anything with the range, it didn’t generate the sequence. Let’s add the consumer:

let nums = (1..100).collect::<Vec<i32>>();

Now, collect() will require that the range gives it some numbers, and so it will do the work of generating the sequence.

Ranges are one of two basic iterators that you’ll see. The other is iter(). iter() can turn a vector into a simple iterator that gives you each element in turn:

let nums = vec![1, 2, 3];
  
  for num in nums.iter() {
     println!("{}", num);
  }

These two basic iterators should serve you well. There are some more advanced iterators, including ones that are infinite.

That’s enough about iterators. Iterator adapters are the last concept we need to talk about with regards to iterators. Let’s get to it!

Iterator adapters

Iterator adapters take an iterator and modify it somehow, producing a new iterator. The simplest one is called map:

(1..100).map(|x| x + 1);

map is called upon another iterator, and produces a new iterator where each element reference has the closure it’s been given as an argument called on it. So this would give us the numbers from 2-100. Well, almost! If you compile the example, you’ll get a warning:

warning: unused result which must be used: iterator adaptors are lazy and
           do nothing unless consumed, #[warn(unused_must_use)] on by default
  (1..100).map(|x| x + 1);
   ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Laziness strikes again! That closure will never execute. This example doesn’t print any numbers:

(1..100).map(|x| println!("{}", x));

If you are trying to execute a closure on an iterator for its side effects, just use for instead.

There are tons of interesting iterator adapters. take(n) will return an iterator over the next n elements of the original iterator. Note that this has no side effect on the original iterator. Let’s try it out with our infinite iterator from before:

for i in (1..).take(5) {
      println!("{}", i);
  }

This will print

1
  2
  3
  4
  5

filter() is an adapter that takes a closure as an argument. This closure returns true or false. The new iterator filter() produces only the elements that that closure returns true for:

for i in (1..100).filter(|&x| x % 2 == 0) {
      println!("{}", i);
  }

This will print all of the even numbers between one and a hundred. (Note that because filter doesn’t consume the elements that are being iterated over, it is passed a reference to each element, and thus the filter predicate uses the &x pattern to extract the integer itself.)

You can chain all three things together: start with an iterator, adapt it a few times, and then consume the result. Check it out:

(1..)
      .filter(|&x| x % 2 == 0)
      .filter(|&x| x % 3 == 0)
      .take(5)
      .collect::<Vec<i32>>();

This will give you a vector containing 6, 12, 18, 24, and 30.

This is just a small taste of what iterators, iterator adapters, and consumers can help you with. There are a number of really useful iterators, and you can write your own as well. Iterators provide a safe, efficient way to manipulate all kinds of lists. They’re a little unusual at first, but if you play with them, you’ll get hooked. For a full list of the different iterators and consumers, check out the iterator module documentation.

Concurrency

Concurrency and parallelism are incredibly important topics in computer science, and are also a hot topic in industry today. Computers are gaining more and more cores, yet many programmers aren’t prepared to fully utilize them.

Rust’s memory safety features also apply to its concurrency story too. Even concurrent Rust programs must be memory safe, having no data races. Rust’s type system is up to the task, and gives you powerful ways to reason about concurrent code at compile time.

Before we talk about the concurrency features that come with Rust, it’s important to understand something: Rust is low-level enough that the vast majority of this is provided by the standard library, not by the language. This means that if you don’t like some aspect of the way Rust handles concurrency, you can implement an alternative way of doing things. mio is a real-world example of this principle in action.

Background: Send and Sync

Concurrency is difficult to reason about. In Rust, we have a strong, static type system to help us reason about our code. As such, Rust gives us two traits to help us make sense of code that can possibly be concurrent.

Send

The first trait we’re going to talk about is Send. When a type T implements Send, it indicates to the compiler that something of this type is able to have ownership transferred safely between threads.

This is important to enforce certain restrictions. For example, if we have a channel connecting two threads, we would want to be able to send some data down the channel and to the other thread. Therefore, we’d ensure that Send was implemented for that type.

In the opposite way, if we were wrapping a library with FFI that isn’t threadsafe, we wouldn’t want to implement Send, and so the compiler will help us enforce that it can’t leave the current thread.

Sync

The second of these traits is called Sync. When a type T implements Sync, it indicates to the compiler that something of this type has no possibility of introducing memory unsafety when used from multiple threads concurrently.

For example, sharing immutable data with an atomic reference count is threadsafe. Rust provides a type like this, Arc<T>, and it implements Sync, so it is safe to share between threads.

These two traits allow you to use the type system to make strong guarantees about the properties of your code under concurrency. Before we demonstrate why, we need to learn how to create a concurrent Rust program in the first place!

Threads

Rust’s standard library provides a library for threads, which allow you to run Rust code in parallel. Here’s a basic example of using std::thread:

use std::thread;
  
  fn main() {
      thread::spawn(|| {
          println!("Hello from a thread!");
      });
  }

The thread::spawn() method accepts a closure, which is executed in a new thread. It returns a handle to the thread, that can be used to wait for the child thread to finish and extract its result:

use std::thread;
  
  fn main() {
      let handle = thread::spawn(|| {
          "Hello from a thread!"
      });
  
      println!("{}", handle.join().unwrap());
  }

Many languages have the ability to execute threads, but it’s wildly unsafe. There are entire books about how to prevent errors that occur from shared mutable state. Rust helps out with its type system here as well, by preventing data races at compile time. Let’s talk about how you actually share things between threads.

Safe Shared Mutable State

Due to Rust’s type system, we have a concept that sounds like a lie: “safe shared mutable state.” Many programmers agree that shared mutable state is very, very bad.

Someone once said this:

Shared mutable state is the root of all evil. Most languages attempt to deal with this problem through the ‘mutable’ part, but Rust deals with it by solving the ‘shared’ part.

The same ownership system that helps prevent using pointers incorrectly also helps rule out data races, one of the worst kinds of concurrency bugs.

As an example, here is a Rust program that would have a data race in many languages. It will not compile:

use std::thread;
  
  fn main() {
      let mut data = vec![1, 2, 3];
  
      for i in 0..3 {
          thread::spawn(move || {
              data[i] += 1;
          });
      }
  
      thread::sleep_ms(50);
  }

This gives us an error:

8:17 error: capture of moved value: `data`
          data[i] += 1;
          ^~~~

Rust knows this wouldn’t be safe! If we had a reference to data in each thread, and the thread takes ownership of the reference, we’d have three owners!

So, we need some type that lets us have more than one reference to a value and that we can share between threads, that is it must implement Sync.

We’ll use Arc<T>, rust’s standard atomic reference count type, which wraps a value up with some extra runtime bookkeeping which allows us to share the ownership of the value between multiple references at the same time.

The bookkeeping consists of a count of how many of these references exist to the value, hence the reference count part of the name.

The Atomic part means Arc<T> can safely be accessed from multiple threads. To do this the compiler guarantees that mutations of the internal count use indivisible operations which can’t have data races.

use std::thread;
  use std::sync::Arc;
  
  fn main() {
      let mut data = Arc::new(vec![1, 2, 3]);
  
      for i in 0..3 {
          let data = data.clone();
          thread::spawn(move || {
              data[i] += 1;
          });
      }
  
      thread::sleep_ms(50);
  }

We now call clone() on our Arc<T>, which increases the internal count. This handle is then moved into the new thread.

And… still gives us an error.

<anon>:11:24 error: cannot borrow immutable borrowed content as mutable
  <anon>:11                    data[i] += 1;
                               ^~~~

Arc<T> assumes one more property about its contents to ensure that it is safe to share across threads: it assumes its contents are Sync. This is true for our value if it’s immutable, but we want to be able to mutate it, so we need something else to persuade the borrow checker we know what we’re doing.

It looks like we need some type that allows us to safely mutate a shared value, for example a type that that can ensure only one thread at a time is able to mutate the value inside it at any one time.

For that, we can use the Mutex<T> type!

Here’s the working version:

use std::sync::{Arc, Mutex};
  use std::thread;
  
  fn main() {
      let data = Arc::new(Mutex::new(vec![1, 2, 3]));
  
      for i in 0..3 {
          let data = data.clone();
          thread::spawn(move || {
              let mut data = data.lock().unwrap();
              data[i] += 1;
          });
      }
  
      thread::sleep_ms(50);
  }

If we’d tried to use Mutex<T> without wrapping it in an Arc<T> we would have seen another error like:

error: the trait `core::marker::Send` is not implemented for the type `std::sync::mute
  ↳ x::MutexGuard<'_, collections::vec::Vec<u32>>` [E0277]
   thread::spawn(move || {
                    ^~~~~~~~~~~~~
  note: `std::sync::mutex::MutexGuard<'_, collections::vec::Vec<u32>>` cannot be sent be
  ↳ tween threads safely
   thread::spawn(move || {
                    ^~~~~~~~~~~~~

You see, Mutex has a lock method which has this signature:

fn lock(&self) -> LockResult<MutexGuard<T>>

and because Send is not implemented for MutexGuard<T>, we couldn’t have transferred the guard across thread boundaries on it’s own.

Let’s examine the body of the thread more closely:

thread::spawn(move || {
      let mut data = data.lock().unwrap();
      data[i] += 1;
  });

First, we call lock(), which acquires the mutex’s lock. Because this may fail, it returns an Result<T, E>, and because this is just an example, we unwrap() it to get a reference to the data. Real code would have more robust error handling here. We’re then free to mutate it, since we have the lock.

Lastly, while the threads are running, we wait on a short timer. But this is not ideal: we may have picked a reasonable amount of time to wait but it’s more likely we’ll either be waiting longer than necessary or not long enough, depending on just how much time the threads actually take to finish computing when the program runs.

A more precise alternative to the timer would be to use one of the mechanisms provided by the Rust standard library for synchronizing threads with each other. Let’s talk about one of them: channels.

Channels

Here’s a version of our code that uses channels for synchronization, rather than waiting for a specific time:

use std::sync::{Arc, Mutex};
  use std::thread;
  use std::sync::mpsc;
  
  fn main() {
      let data = Arc::new(Mutex::new(0));
  
      let (tx, rx) = mpsc::channel();
  
      for _ in 0..10 {
          let (data, tx) = (data.clone(), tx.clone());
  
          thread::spawn(move || {
              let mut data = data.lock().unwrap();
              *data += 1;
  
              tx.send(());
          });
      }
  
      for _ in 0..10 {
          rx.recv();
      }
  }

We use the mpsc::channel() method to construct a new channel. We just send a simple () down the channel, and then wait for ten of them to come back.

While this channel is just sending a generic signal, we can send any data that is Send over the channel!

use std::thread;
  use std::sync::mpsc;
  
  fn main() {
      let (tx, rx) = mpsc::channel();
  
      for _ in 0..10 {
          let tx = tx.clone();
  
          thread::spawn(move || {
              let answer = 42;
  
              tx.send(answer);
          });
      }
  
     rx.recv().ok().expect("Could not receive answer");
  }

A u32 is Send because we can make a copy. So we create a thread, ask it to calculate the answer, and then it send()s us the answer over the channel.

Panics

A panic! will crash the currently executing thread. You can use Rust’s threads as a simple isolation mechanism:

use std::thread;
  
  let handle = thread::spawn(move || {
      panic!("oops!");
  });
  
  let result = handle.join();
  
  assert!(result.is_err());

Thread.join() gives us a Result back, which allows us to check if the thread has panicked or not.

Error Handling

Like most programming languages, Rust encourages the programmer to handle errors in a particular way. Generally speaking, error handling is divided into two broad categories: exceptions and return values. Rust opts for return values.

In this chapter, we intend to provide a comprehensive treatment of how to deal with errors in Rust. More than that, we will attempt to introduce error handling one piece at a time so that you’ll come away with a solid working knowledge of how everything fits together.

When done naïvely, error handling in Rust can be verbose and annoying. This chapter will explore those stumbling blocks and demonstrate how to use the standard library to make error handling concise and ergonomic.

Table of Contents

This chapter is very long, mostly because we start at the very beginning with sum types and combinators, and try to motivate the way Rust does error handling incrementally. As such, programmers with experience in other expressive type systems may want to jump around.

The Basics

You can think of error handling as using case analysis to determine whether a computation was successful or not. As you will see, the key to ergonomic error handling is reducing the amount of explicit case analysis the programmer has to do while keeping code composable.

Keeping code composable is important, because without that requirement, we could panic whenever we come across something unexpected. (panic causes the current task to unwind, and in most cases, the entire program aborts.) Here’s an example:

// Guess a number between 1 and 10.
  // If it matches the number we had in mind, return true. Else, return false.
  fn guess(n: i32) -> bool {
      if n < 1 || n > 10 {
          panic!("Invalid number: {}", n);
      }
      n == 5
  }
  
  fn main() {
      guess(11);
  }

If you try running this code, the program will crash with a message like this:

thread '<main>' panicked at 'Invalid number: 11', src/bin/panic-simple.rs:5

Here’s another example that is slightly less contrived. A program that accepts an integer as an argument, doubles it and prints it.

use std::env;
  
  fn main() {
      let mut argv = env::args();
      let arg: String = argv.nth(1).unwrap(); // error 1
      let n: i32 = arg.parse().unwrap(); // error 2
      println!("{}", 2 * n);
  }

If you give this program zero arguments (error 1) or if the first argument isn’t an integer (error 2), the program will panic just like in the first example.

You can think of this style of error handling as similar to a bull running through a china shop. The bull will get to where it wants to go, but it will trample everything in the process.

Unwrapping explained

In the previous example, we claimed that the program would simply panic if it reached one of the two error conditions, yet, the program does not include an explicit call to panic like the first example. This is because the panic is embedded in the calls to unwrap.

To “unwrap” something in Rust is to say, “Give me the result of the computation, and if there was an error, just panic and stop the program.” It would be better if we just showed the code for unwrapping because it is so simple, but to do that, we will first need to explore the Option and Result types. Both of these types have a method called unwrap defined on them.

The Option type

The Option type is defined in the standard library:

enum Option<T> {
      None,
      Some(T),
  }

The Option type is a way to use Rust’s type system to express the possibility of absence. Encoding the possibility of absence into the type system is an important concept because it will cause the compiler to force the programmer to handle that absence. Let’s take a look at an example that tries to find a character in a string:

// Searches `haystack` for the Unicode character `needle`. If one is found, the
  // byte offset of the character is returned. Otherwise, `None` is returned.
  fn find(haystack: &str, needle: char) -> Option<usize> {
      for (offset, c) in haystack.char_indices() {
          if c == needle {
              return Some(offset);
          }
      }
      None
  }

Notice that when this function finds a matching character, it doen’t just return the offset. Instead, it returns Some(offset). Some is a variant or a value constructor for the Option type. You can think of it as a function with the type fn<T>(value: T) -> Option<T>. Correspondingly, None is also a value constructor, except it has no arguments. You can think of None as a function with the type fn<T>() -> Option<T>.

This might seem like much ado about nothing, but this is only half of the story. The other half is using the find function we’ve written. Let’s try to use it to find the extension in a file name.

fn main() {
      let file_name = "foobar.rs";
      match find(file_name, '.') {
          None => println!("No file extension found."),
          Some(i) => println!("File extension: {}", &file_name[i+1..]),
      }
  }

This code uses pattern matching to do case analysis on the Option<usize> returned by the find function. In fact, case analysis is the only way to get at the value stored inside an Option<T>. This means that you, as the programmer, must handle the case when an Option<T> is None instead of Some(t).

But wait, what about unwrap used in unwrap-double? There was no case analysis there! Instead, the case analysis was put inside the unwrap method for you. You could define it yourself if you want:

enum Option<T> {
      None,
      Some(T),
  }
  
  impl<T> Option<T> {
      fn unwrap(self) -> T {
          match self {
              Option::Some(val) => val,
              Option::None =>
                panic!("called `Option::unwrap()` on a `None` value"),
          }
      }
  }

The unwrap method abstracts away the case analysis. This is precisely the thing that makes unwrap ergonomic to use. Unfortunately, that panic! means that unwrap is not composable: it is the bull in the china shop.

Composing Option<T> values

In option-ex-string-find we saw how to use find to discover the extension in a file name. Of course, not all file names have a . in them, so it’s possible that the file name has no extension. This possibility of absence is encoded into the types using Option<T>. In other words, the compiler will force us to address the possibility that an extension does not exist. In our case, we just print out a message saying as such.

Getting the extension of a file name is a pretty common operation, so it makes sense to put it into a function:

// Returns the extension of the given file name, where the extension is defined
  // as all characters proceding the first `.`.
  // If `file_name` has no `.`, then `None` is returned.
  fn extension_explicit(file_name: &str) -> Option<&str> {
      match find(file_name, '.') {
          None => None,
          Some(i) => Some(&file_name[i+1..]),
      }
  }

(Pro-tip: don’t use this code. Use the extension method in the standard library instead.)

The code stays simple, but the important thing to notice is that the type of find forces us to consider the possibility of absence. This is a good thing because it means the compiler won’t let us accidentally forget about the case where a file name doesn’t have an extension. On the other hand, doing explicit case analysis like we’ve done in extension_explicit every time can get a bit tiresome.

In fact, the case analysis in extension_explicit follows a very common pattern: map a function on to the value inside of an Option<T>, unless the option is None, in which case, just return None.

Rust has parametric polymorphism, so it is very easy to define a combinator that abstracts this pattern:

fn map<F, T, A>(option: Option<T>, f: F) -> Option<A> where F: FnOnce(T) -> A {
      match option {
          None => None,
          Some(value) => Some(f(value)),
      }
  }

Indeed, map is defined as a method on Option<T> in the standard library.

Armed with our new combinator, we can rewrite our extension_explicit method to get rid of the case analysis:

// Returns the extension of the given file name, where the extension is defined
  // as all characters proceding the first `.`.
  // If `file_name` has no `.`, then `None` is returned.
  fn extension(file_name: &str) -> Option<&str> {
      find(file_name, '.').map(|i| &file_name[i+1..])
  }

One other pattern that we find is very common is assigning a default value to the case when an Option value is None. For example, maybe your program assumes that the extension of a file is rs even if none is present. As you might imagine, the case analysis for this is not specific to file extensions - it can work with any Option<T>:

fn unwrap_or<T>(option: Option<T>, default: T) -> T {
      match option {
          None => default,
          Some(value) => value,
      }
  }

The trick here is that the default value must have the same type as the value that might be inside the Option<T>. Using it is dead simple in our case:

#
  fn main() {
      assert_eq!(extension("foobar.csv").unwrap_or("rs"), "csv");
      assert_eq!(extension("foobar").unwrap_or("rs"), "rs");
  }

(Note that unwrap_or is defined as a method on Option<T> in the standard library, so we use that here instead of the free-standing function we defined above. Don’t forget to check out the more general unwrap_or_else method.)

There is one more combinator that we think is worth paying special attention to: and_then. It makes it easy to compose distinct computations that admit the possibility of absence. For example, much of the code in this section is about finding an extension given a file name. In order to do this, you first need the file name which is typically extracted from a file path. While most file paths have a file name, not all of them do. For example, ., .. or /.

So, we are tasked with the challenge of finding an extension given a file path. Let’s start with explicit case analysis:

fn file_path_ext_explicit(file_path: &str) -> Option<&str> {
      match file_name(file_path) {
          None => None,
          Some(name) => match extension(name) {
              None => None,
              Some(ext) => Some(ext),
          }
      }
  }
  
  fn file_name(file_path: &str) -> Option<&str> {
    // implementation elided
    unimplemented!()
  }

You might think that we could just use the map combinator to reduce the case analysis, but its type doesn’t quite fit. Namely, map takes a function that does something only with the inner value. The result of that function is then always rewrapped with Some. Instead, we need something like map, but which allows the caller to return another Option. Its generic implementation is even simpler than map:

fn and_then<F, T, A>(option: Option<T>, f: F) -> Option<A>
          where F: FnOnce(T) -> Option<A> {
      match option {
          None => None,
          Some(value) => f(value),
      }
  }

Now we can rewrite our file_path_ext function without explicit case analysis:

fn file_path_ext(file_path: &str) -> Option<&str> {
      file_name(file_path).and_then(extension)
  }

The Option type has many other combinators defined in the standard library. It is a good idea to skim this list and familiarize yourself with what’s available—they can often reduce case analysis for you. Familiarizing yourself with these combinators will pay dividends because many of them are also defined (with similar semantics) for Result, which we will talk about next.

Combinators make using types like Option ergonomic because they reduce explicit case analysis. They are also composable because they permit the caller to handle the possibility of absence in their own way. Methods like unwrap remove choices because they will panic if Option<T> is None.

The Result type

The Result type is also defined in the standard library:

enum Result<T, E> {
      Ok(T),
      Err(E),
  }

The Result type is a richer version of Option. Instead of expressing the possibility of absence like Option does, Result expresses the possibility of error. Usually, the error is used to explain why the result of some computation failed. This is a strictly more general form of Option. Consider the following type alias, which is semantically equivalent to the real Option<T> in every way:

type Option<T> = Result<T, ()>;

This fixes the second type parameter of Result to always be () (pronounced “unit” or “empty tuple”). Exactly one value inhabits the () type: (). (Yup, the type and value level terms have the same notation!)

The Result type is a way of representing one of two possible outcomes in a computation. By convention, one outcome is meant to be expected or “Ok” while the other outcome is meant to be unexpected or “Err”.

Just like Option, the Result type also has an unwrap method defined in the standard library. Let’s define it:

impl<T, E: ::std::fmt::Debug> Result<T, E> {
      fn unwrap(self) -> T {
          match self {
              Result::Ok(val) => val,
              Result::Err(err) =>
                panic!("called `Result::unwrap()` on an `Err` value: {:?}", err),
          }
      }
  }

This is effectively the same as our definition for Option::unwrap, except it includes the error value in the panic! message. This makes debugging easier, but it also requires us to add a Debug constraint on the E type parameter (which represents our error type). Since the vast majority of types should satisfy the Debug constraint, this tends to work out in practice. (Debug on a type simply means that there’s a reasonable way to print a human readable description of values with that type.)

OK, let’s move on to an example.

Parsing integers

The Rust standard library makes converting strings to integers dead simple. It’s so easy in fact, that it is very tempting to write something like the following:

fn double_number(number_str: &str) -> i32 {
      2 * number_str.parse::<i32>().unwrap()
  }
  
  fn main() {
      let n: i32 = double_number("10");
      assert_eq!(n, 20);
  }

At this point, you should be skeptical of calling unwrap. For example, if the string doesn’t parse as a number, you’ll get a panic:

thread '<main>' panicked at 'called `Result::unwrap()` on an `Err` value: ParseIntErro
  ↳ r { kind: InvalidDigit }', /home/rustbuild/src/rust-buildbot/slave/beta-dist-rustc-l
  ↳ inux/build/src/libcore/result.rs:729

This is rather unsightly, and if this happened inside a library you’re using, you might be understandably annoyed. Instead, we should try to handle the error in our function and let the caller decide what to do. This means changing the return type of double_number. But to what? Well, that requires looking at the signature of the parse method in the standard library:

impl str {
      fn parse<F: FromStr>(&self) -> Result<F, F::Err>;
  }

Hmm. So we at least know that we need to use a Result. Certainly, it’s possible that this could have returned an Option. After all, a string either parses as a number or it doesn’t, right? That’s certainly a reasonable way to go, but the implementation internally distinguishes why the string didn’t parse as an integer. (Whether it’s an empty string, an invalid digit, too big or too small.) Therefore, using a Result makes sense because we want to provide more information than simply “absence.” We want to say why the parsing failed. You should try to emulate this line of reasoning when faced with a choice between Option and Result. If you can provide detailed error information, then you probably should. (We’ll see more on this later.)

OK, but how do we write our return type? The parse method as defined above is generic over all the different number types defined in the standard library. We could (and probably should) also make our function generic, but let’s favor explicitness for the moment. We only care about i32, so we need to find its implementation of FromStr (do a CTRL-F in your browser for “FromStr”) and look at its associated type Err. We did this so we can find the concrete error type. In this case, it’s std::num::ParseIntError. Finally, we can rewrite our function:

use std::num::ParseIntError;
  
  fn double_number(number_str: &str) -> Result<i32, ParseIntError> {
      match number_str.parse::<i32>() {
          Ok(n) => Ok(2 * n),
          Err(err) => Err(err),
      }
  }
  
  fn main() {
      match double_number("10") {
          Ok(n) => assert_eq!(n, 20),
          Err(err) => println!("Error: {:?}", err),
      }
  }

This is a little better, but now we’ve written a lot more code! The case analysis has once again bitten us.

Combinators to the rescue! Just like Option, Result has lots of combinators defined as methods. There is a large intersection of common combinators between Result and Option. In particular, map is part of that intersection:

use std::num::ParseIntError;
  
  fn double_number(number_str: &str) -> Result<i32, ParseIntError> {
      number_str.parse::<i32>().map(|n| 2 * n)
  }
  
  fn main() {
      match double_number("10") {
          Ok(n) => assert_eq!(n, 20),
          Err(err) => println!("Error: {:?}", err),
      }
  }

The usual suspects are all there for Result, including unwrap_or and and_then. Additionally, since Result has a second type parameter, there are combinators that affect only the error type, such as map_err (instead of map) and or_else (instead of and_then).

The Result type alias idiom

In the standard library, you may frequently see types like Result<i32>. But wait, we defined Result to have two type parameters. How can we get away with only specifying one? The key is to define a Result type alias that fixes one of the type parameters to a particular type. Usually the fixed type is the error type. For example, our previous example parsing integers could be rewritten like this:

use std::num::ParseIntError;
  use std::result;
  
  type Result<T> = result::Result<T, ParseIntError>;
  
  fn double_number(number_str: &str) -> Result<i32> {
      unimplemented!();
  }

Why would we do this? Well, if we have a lot of functions that could return ParseIntError, then it’s much more convenient to define an alias that always uses ParseIntError so that we don’t have to write it out all the time.

The most prominent place this idiom is used in the standard library is with io::Result. Typically, one writes io::Result<T>, which makes it clear that you’re using the io module’s type alias instead of the plain definition from std::result. (This idiom is also used for fmt::Result.)

A brief interlude: unwrapping isn’t evil

If you’ve been following along, you might have noticed that I’ve taken a pretty hard line against calling methods like unwrap that could panic and abort your program. Generally speaking, this is good advice.

However, unwrap can still be used judiciously. What exactly justifies use of unwrap is somewhat of a grey area and reasonable people can disagree. I’ll summarize some of my opinions on the matter.

  • In examples and quick ‘n’ dirty code. Sometimes you’re writing examples or a quick program, and error handling simply isn’t important. Beating the convenience of unwrap can be hard in such scenarios, so it is very appealing.
  • When panicking indicates a bug in the program. When the invariants of your code should prevent a certain case from happening (like, say, popping from an empty stack), then panicking can be permissible. This is because it exposes a bug in your program. This can be explicit, like from an assert! failing, or it could be because your index into an array was out of bounds.

This is probably not an exhaustive list. Moreover, when using an Option, it is often better to use its expect method. expect does exactly the same thing as unwrap, except it prints a message you give to expect. This makes the resulting panic a bit nicer to deal with, since it will show your message instead of “called unwrap on a None value.”

My advice boils down to this: use good judgment. There’s a reason why the words “never do X” or “Y is considered harmful” don’t appear in my writing. There are trade offs to all things, and it is up to you as the programmer to determine what is acceptable for your use cases. My goal is only to help you evaluate trade offs as accurately as possible.

Now that we’ve covered the basics of error handling in Rust, and explained unwrapping, let’s start exploring more of the standard library.

Working with multiple error types

Thus far, we’ve looked at error handling where everything was either an Option<T> or a Result<T, SomeError>. But what happens when you have both an Option and a Result? Or what if you have a Result<T, Error1> and a Result<T, Error2>? Handling composition of distinct error types is the next challenge in front of us, and it will be the major theme throughout the rest of this chapter.

Composing Option and Result

So far, I’ve talked about combinators defined for Option and combinators defined for Result. We can use these combinators to compose results of different computations without doing explicit case analysis.

Of course, in real code, things aren’t always as clean. Sometimes you have a mix of Option and Result types. Must we resort to explicit case analysis, or can we continue using combinators?

For now, let’s revisit one of the first examples in this chapter:

use std::env;
  
  fn main() {
      let mut argv = env::args();
      let arg: String = argv.nth(1).unwrap(); // error 1
      let n: i32 = arg.parse().unwrap(); // error 2
      println!("{}", 2 * n);
  }

Given our new found knowledge of Option, Result and their various combinators, we should try to rewrite this so that errors are handled properly and the program doesn’t panic if there’s an error.

The tricky aspect here is that argv.nth(1) produces an Option while arg.parse() produces a Result. These aren’t directly composable. When faced with both an Option and a Result, the solution is usually to convert the Option to a Result. In our case, the absence of a command line parameter (from env::args()) means the user didn’t invoke the program correctly. We could just use a String to describe the error. Let’s try:

use std::env;
  
  fn double_arg(mut argv: env::Args) -> Result<i32, String> {
      argv.nth(1)
          .ok_or("Please give at least one argument".to_owned())
          .and_then(|arg| arg.parse::<i32>().map_err(|err| err.to_string()))
  }
  
  fn main() {
      match double_arg(env::args()) {
          Ok(n) => println!("{}", n),
          Err(err) => println!("Error: {}", err),
      }
  }

There are a couple new things in this example. The first is the use of the Option::ok_or combinator. This is one way to convert an Option into a Result. The conversion requires you to specify what error to use if Option is None. Like the other combinators we’ve seen, its definition is very simple:

fn ok_or<T, E>(option: Option<T>, err: E) -> Result<T, E> {
      match option {
          Some(val) => Ok(val),
          None => Err(err),
      }
  }

The other new combinator used here is Result::map_err. This is just like Result::map, except it maps a function on to the error portion of a Result value. If the Result is an Ok(...) value, then it is returned unmodified.

We use map_err here because it is necessary for the error types to remain the same (because of our use of and_then). Since we chose to convert the Option<String> (from argv.nth(1)) to a Result<String, String>, we must also convert the ParseIntError from arg.parse() to a String.

The limits of combinators

Doing IO and parsing input is a very common task, and it’s one that I personally have done a lot of in Rust. Therefore, we will use (and continue to use) IO and various parsing routines to exemplify error handling.

Let’s start simple. We are tasked with opening a file, reading all of its contents and converting its contents to a number. Then we multiply it by 2 and print the output.

Although I’ve tried to convince you not to use unwrap, it can be useful to first write your code using unwrap. It allows you to focus on your problem instead of the error handling, and it exposes the points where proper error handling need to occur. Let’s start there so we can get a handle on the code, and then refactor it to use better error handling.

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> i32 {
      let mut file = File::open(file_path).unwrap(); // error 1
      let mut contents = String::new();
      file.read_to_string(&mut contents).unwrap(); // error 2
      let n: i32 = contents.trim().parse().unwrap(); // error 3
      2 * n
  }
  
  fn main() {
      let doubled = file_double("foobar");
      println!("{}", doubled);
  }

(N.B. The AsRef<Path> is used because those are the same bounds used on std::fs::File::open. This makes it ergnomic to use any kind of string as a file path.)

There are three different errors that can occur here:

  1. A problem opening the file.
  2. A problem reading data from the file.
  3. A problem parsing the data as a number.

The first two problems are described via the std::io::Error type. We know this because of the return types of std::fs::File::open and std::io::Read::read_to_string. (Note that they both use the Result type alias idiom described previously. If you click on the Result type, you’ll see the type alias, and consequently, the underlying io::Error type.) The third problem is described by the std::num::ParseIntError type. The io::Error type in particular is pervasive throughout the standard library. You will see it again and again.

Let’s start the process of refactoring the file_double function. To make this function composable with other components of the program, it should not panic if any of the above error conditions are met. Effectively, this means that the function should return an error if any of its operations fail. Our problem is that the return type of file_double is i32, which does not give us any useful way of reporting an error. Thus, we must start by changing the return type from i32 to something else.

The first thing we need to decide: should we use Option or Result? We certainly could use Option very easily. If any of the three errors occur, we could simply return None. This will work and it is better than panicking, but we can do a lot better. Instead, we should pass some detail about the error that occurred. Since we want to express the possibility of error, we should use Result<i32, E>. But what should E be? Since two different types of errors can occur, we need to convert them to a common type. One such type is String. Let’s see how that impacts our code:

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, String> {
      File::open(file_path)
           .map_err(|err| err.to_string())
           .and_then(|mut file| {
                let mut contents = String::new();
                file.read_to_string(&mut contents)
                    .map_err(|err| err.to_string())
                    .map(|_| contents)
           })
           .and_then(|contents| {
                contents.trim().parse::<i32>()
                        .map_err(|err| err.to_string())
           })
           .map(|n| 2 * n)
  }
  
  fn main() {
      match file_double("foobar") {
          Ok(n) => println!("{}", n),
          Err(err) => println!("Error: {}", err),
      }
  }

This code looks a bit hairy. It can take quite a bit of practice before code like this becomes easy to write. The way we write it is by following the types. As soon as we changed the return type of file_double to Result<i32, String>, we had to start looking for the right combinators. In this case, we only used three different combinators: and_then, map and map_err.

and_then is used to chain multiple computations where each computation could return an error. After opening the file, there are two more computations that could fail: reading from the file and parsing the contents as a number. Correspondingly, there are two calls to and_then.

map is used to apply a function to the Ok(...) value of a Result. For example, the very last call to map multiplies the Ok(...) value (which is an i32) by 2. If an error had occurred before that point, this operation would have been skipped because of how map is defined.

map_err is the trick the makes all of this work. map_err is just like map, except it applies a function to the Err(...) value of a Result. In this case, we want to convert all of our errors to one type: String. Since both io::Error and num::ParseIntError implement ToString, we can call the to_string() method to convert them.

With all of that said, the code is still hairy. Mastering use of combinators is important, but they have their limits. Let’s try a different approach: early returns.

Early returns

I’d like to take the code from the previous section and rewrite it using early returns. Early returns let you exit the function early. We can’t return early in file_double from inside another closure, so we’ll need to revert back to explicit case analysis.

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, String> {
      let mut file = match File::open(file_path) {
          Ok(file) => file,
          Err(err) => return Err(err.to_string()),
      };
      let mut contents = String::new();
      if let Err(err) = file.read_to_string(&mut contents) {
          return Err(err.to_string());
      }
      let n: i32 = match contents.trim().parse() {
          Ok(n) => n,
          Err(err) => return Err(err.to_string()),
      };
      Ok(2 * n)
  }
  
  fn main() {
      match file_double("foobar") {
          Ok(n) => println!("{}", n),
          Err(err) => println!("Error: {}", err),
      }
  }

Reasonable people can disagree over whether this code is better that the code that uses combinators, but if you aren’t familiar with the combinator approach, this code looks simpler to read to me. It uses explicit case analysis with match and if let. If an error occurs, it simply stops executing the function and returns the error (by converting it to a string).

Isn’t this a step backwards though? Previously, we said that the key to ergonomic error handling is reducing explicit case analysis, yet we’ve reverted back to explicit case analysis here. It turns out, there are multiple ways to reduce explicit case analysis. Combinators aren’t the only way.

The try! macro

A cornerstone of error handling in Rust is the try! macro. The try! macro abstracts case analysis just like combinators, but unlike combinators, it also abstracts control flow. Namely, it can abstract the early return pattern seen above.

Here is a simplified definition of a try! macro:

macro_rules! try {
      ($e:expr) => (match $e {
          Ok(val) => val,
          Err(err) => return Err(err),
      });
  }

(The real definition is a bit more sophisticated. We will address that later.)

Using the try! macro makes it very easy to simplify our last example. Since it does the case analysis and the early return for us, we get tighter code that is easier to read:

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, String> {
      let mut file = try!(File::open(file_path).map_err(|e| e.to_string()));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents).map_err(|e| e.to_string()));
      let n = try!(contents.trim().parse::<i32>().map_err(|e| e.to_string()));
      Ok(2 * n)
  }
  
  fn main() {
      match file_double("foobar") {
          Ok(n) => println!("{}", n),
          Err(err) => println!("Error: {}", err),
      }
  }

The map_err calls are still necessary given our definition of try!. This is because the error types still need to be converted to String. The good news is that we will soon learn how to remove those map_err calls! The bad news is that we will need to learn a bit more about a couple important traits in the standard library before we can remove the map_err calls.

Defining your own error type

Before we dive into some of the standard library error traits, I’d like to wrap up this section by removing the use of String as our error type in the previous examples.

Using String as we did in our previous examples is convenient because it’s easy to convert errors to strings, or even make up your own errors as strings on the spot. However, using String for your errors has some downsides.

The first downside is that the error messages tend to clutter your code. It’s possible to define the error messages elsewhere, but unless you’re unusually disciplined, it is very tempting to embed the error message into your code. Indeed, we did exactly this in a previous example.

The second and more important downside is that Strings are lossy. That is, if all errors are converted to strings, then the errors we pass to the caller become completely opaque. The only reasonable thing the caller can do with a String error is show it to the user. Certainly, inspecting the string to determine the type of error is not robust. (Admittedly, this downside is far more important inside of a library as opposed to, say, an application.)

For example, the io::Error type embeds an io::ErrorKind, which is structured data that represents what went wrong during an IO operation. This is important because you might want to react differently depending on the error. (e.g., A BrokenPipe error might mean quitting your program gracefully while a NotFound error might mean exiting with an error code and showing an error to the user.) With io::ErrorKind, the caller can examine the type of an error with case analysis, which is strictly superior to trying to tease out the details of an error inside of a String.

Instead of using a String as an error type in our previous example of reading an integer from a file, we can define our own error type that represents errors with structured data. We endeavor to not drop information from underlying errors in case the caller wants to inspect the details.

The ideal way to represent one of many possibilities is to define our own sum type using enum. In our case, an error is either an io::Error or a num::ParseIntError, so a natural definition arises:

use std::io;
  use std::num;
  
  // We derive `Debug` because all types should probably derive `Debug`.
  // This gives us a reasonable human readable description of `CliError` values.
  #[derive(Debug)]
  enum CliError {
      Io(io::Error),
      Parse(num::ParseIntError),
  }

Tweaking our code is very easy. Instead of converting errors to strings, we simply convert them to our CliError type using the corresponding value constructor:

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, CliError> {
      let mut file = try!(File::open(file_path).map_err(CliError::Io));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents).map_err(CliError::Io));
      let n: i32 = try!(contents.trim().parse().map_err(CliError::Parse));
      Ok(2 * n)
  }
  
  fn main() {
      match file_double("foobar") {
          Ok(n) => println!("{}", n),
          Err(err) => println!("Error: {:?}", err),
      }
  }

The only change here is switching map_err(|e| e.to_string()) (which converts errors to strings) to map_err(CliError::Io) or map_err(CliError::Parse). The caller gets to decide the level of detail to report to the user. In effect, using a String as an error type removes choices from the caller while using a custom enum error type like CliError gives the caller all of the conveniences as before in addition to structured data describing the error.

A rule of thumb is to define your own error type, but a String error type will do in a pinch, particularly if you’re writing an application. If you’re writing a library, defining your own error type should be strongly preferred so that you don’t remove choices from the caller unnecessarily.

Standard library traits used for error handling

The standard library defines two integral traits for error handling: std::error::Error and std::convert::From. While Error is designed specifically for generically describing errors, the From trait serves a more general role for converting values between two distinct types.

The Error trait

The Error trait is defined in the standard library:

use std::fmt::{Debug, Display};
  
  trait Error: Debug + Display {
    /// A short description of the error.
    fn description(&self) -> &str;
  
    /// The lower level cause of this error, if any.
    fn cause(&self) -> Option<&Error> { None }
  }

This trait is super generic because it is meant to be implemented for all types that represent errors. This will prove useful for writing composable code as we’ll see later. Otherwise, the trait allows you to do at least the following things:

  • Obtain a Debug representation of the error.
  • Obtain a user-facing Display representation of the error.
  • Obtain a short description of the error (via the description method).
  • Inspect the causal chain of an error, if one exists (via the cause method).

The first two are a result of Error requiring impls for both Debug and Display. The latter two are from the two methods defined on Error. The power of Error comes from the fact that all error types impl Error, which means errors can be existentially quantified as a trait object. This manifests as either Box<Error> or &Error. Indeed, the cause method returns an &Error, which is itself a trait object. We’ll revisit the Error trait’s utility as a trait object later.

For now, it suffices to show an example implementing the Error trait. Let’s use the error type we defined in the previous section:

use std::io;
  use std::num;
  
  // We derive `Debug` because all types should probably derive `Debug`.
  // This gives us a reasonable human readable description of `CliError` values.
  #[derive(Debug)]
  enum CliError {
      Io(io::Error),
      Parse(num::ParseIntError),
  }

This particular error type represents the possibility of two types of errors occurring: an error dealing with I/O or an error converting a string to a number. The error could represent as many error types as you want by adding new variants to the enum definition.

Implementing Error is pretty straight-forward. It’s mostly going to be a lot explicit case analysis.

use std::error;
  use std::fmt;
  
  impl fmt::Display for CliError {
      fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
          match *self {
              // Both underlying errors already impl `Display`, so we defer to
              // their implementations.
              CliError::Io(ref err) => write!(f, "IO error: {}", err),
              CliError::Parse(ref err) => write!(f, "Parse error: {}", err),
          }
      }
  }
  
  impl error::Error for CliError {
      fn description(&self) -> &str {
          // Both underlying errors already impl `Error`, so we defer to their
          // implementations.
          match *self {
              CliError::Io(ref err) => err.description(),
              // Normally we can just write `err.description()`, but the error
              // type has a concrete method called `description`, which conflicts
              // with the trait method. For now, we must explicitly call
              // `description` through the `Error` trait.
              CliError::Parse(ref err) => error::Error::description(err),
          }
      }
  
      fn cause(&self) -> Option<&error::Error> {
          match *self {
              // N.B. Both of these implicitly cast `err` from their concrete
              // types (either `&io::Error` or `&num::ParseIntError`)
              // to a trait object `&Error`. This works because both error types
              // implement `Error`.
              CliError::Io(ref err) => Some(err),
              CliError::Parse(ref err) => Some(err),
          }
      }
  }

We note that this is a very typical implementation of Error: match on your different error types and satisfy the contracts defined for description and cause.

The From trait

The std::convert::From trait is defined in the standard library:

trait From<T> {
      fn from(T) -> Self;
  }

Deliciously simple, yes? From is very useful because it gives us a generic way to talk about conversion from a particular type T to some other type (in this case, “some other type” is the subject of the impl, or Self). The crux of From is the set of implementations provided by the standard library.

Here are a few simple examples demonstrating how From works:

let string: String = From::from("foo");
  let bytes: Vec<u8> = From::from("foo");
  let cow: ::std::borrow::Cow<str> = From::from("foo");

OK, so From is useful for converting between strings. But what about errors? It turns out, there is one critical impl:

impl<'a, E: Error + 'a> From<E> for Box<Error + 'a>

This impl says that for any type that impls Error, we can convert it to a trait object Box<Error>. This may not seem terribly surprising, but it is useful in a generic context.

Remember the two errors we were dealing with previously? Specifically, io::Error and num::ParseIntError. Since both impl Error, they work with From:

use std::error::Error;
  use std::fs;
  use std::io;
  use std::num;
  
  // We have to jump through some hoops to actually get error values.
  let io_err: io::Error = io::Error::last_os_error();
  let parse_err: num::ParseIntError = "not a number".parse::<i32>().unwrap_err();
  
  // OK, here are the conversions.
  let err1: Box<Error> = From::from(io_err);
  let err2: Box<Error> = From::from(parse_err);

There is a really important pattern to recognize here. Both err1 and err2 have the same type. This is because they are existentially quantified types, or trait objects. In particularly, their underlying type is erased from the compiler’s knowledge, so it truly sees err1 and err2 as exactly the same. Additionally, we constructed err1 and err2 using precisely the same function call: From::from. This is because From::from is overloaded on both its argument and its return type.

This pattern is important because it solves a problem we had earlier: it gives us a way to reliably convert errors to the same type using the same function.

Time to revisit an old friend; the try! macro.

The real try! macro

Previously, we presented this definition of try!:

macro_rules! try {
      ($e:expr) => (match $e {
          Ok(val) => val,
          Err(err) => return Err(err),
      });
  }

This is not it’s real definition. It’s real definition is in the standard library:

macro_rules! try {
      ($e:expr) => (match $e {
          Ok(val) => val,
          Err(err) => return Err(::std::convert::From::from(err)),
      });
  }

There’s one tiny but powerful change: the error value is passed through From::from. This makes the try! macro a lot more powerful because it gives you automatic type conversion for free.

Armed with our more powerful try! macro, let’s take a look at code we wrote previously to read a file and convert its contents to an integer:

use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, String> {
      let mut file = try!(File::open(file_path).map_err(|e| e.to_string()));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents).map_err(|e| e.to_string()));
      let n = try!(contents.trim().parse::<i32>().map_err(|e| e.to_string()));
      Ok(2 * n)
  }

Earlier, we promised that we could get rid of the map_err calls. Indeed, all we have to do is pick a type that From works with. As we saw in the previous section, From has an impl that let’s it convert any error type into a Box<Error>:

use std::error::Error;
  use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, Box<Error>> {
      let mut file = try!(File::open(file_path));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents));
      let n = try!(contents.trim().parse::<i32>());
      Ok(2 * n)
  }

We are getting very close to ideal error handling. Our code has very little overhead as a result from error handling because the try! macro encapsulates three things simultaneously:

  1. Case analysis.
  2. Control flow.
  3. Error type conversion.

When all three things are combined, we get code that is unencumbered by combinators, calls to unwrap or case analysis.

There’s one little nit left: the Box<Error> type is opaque. If we return a Box<Error> to the caller, the caller can’t (easily) inspect underlying error type. The situation is certainly better than String because the caller can call methods like description and cause, but the limitation remains: Box<Error> is opaque. (N.B. This isn’t entirely true because Rust does have runtime reflection, which is useful in some scenarios that are beyond the scope of this chapter.)

It’s time to revisit our custom CliError type and tie everything together.

Composing custom error types

In the last section, we looked at the real try! macro and how it does automatic type conversion for us by calling From::from on the error value. In particular, we converted errors to Box<Error>, which works, but the type is opaque to callers.

To fix this, we use the same remedy that we’re already familiar with: a custom error type. Once again, here is the code that reads the contents of a file and converts it to an integer:

use std::fs::File;
  use std::io::{self, Read};
  use std::num;
  use std::path::Path;
  
  // We derive `Debug` because all types should probably derive `Debug`.
  // This gives us a reasonable human readable description of `CliError` values.
  #[derive(Debug)]
  enum CliError {
      Io(io::Error),
      Parse(num::ParseIntError),
  }
  
  fn file_double_verbose<P: AsRef<Path>>(file_path: P) -> Result<i32, CliError> {
      let mut file = try!(File::open(file_path).map_err(CliError::Io));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents).map_err(CliError::Io));
      let n: i32 = try!(contents.trim().parse().map_err(CliError::Parse));
      Ok(2 * n)
  }

Notice that we still have the calls to map_err. Why? Well, recall the definitions of try! and From. The problem is that there is no From impl that allows us to convert from error types like io::Error and num::ParseIntError to our own custom CliError. Of course, it is easy to fix this! Since we defined CliError, we can impl From with it:

use std::io;
  use std::num;
  
  impl From<io::Error> for CliError {
      fn from(err: io::Error) -> CliError {
          CliError::Io(err)
      }
  }
  
  impl From<num::ParseIntError> for CliError {
      fn from(err: num::ParseIntError) -> CliError {
          CliError::Parse(err)
      }
  }

All these impls are doing is teaching From how to create a CliError from other error types. In our case, construction is as simple as invoking the corresponding value constructor. Indeed, it is typically this easy.

We can finally rewrite file_double:


  use std::fs::File;
  use std::io::Read;
  use std::path::Path;
  
  fn file_double<P: AsRef<Path>>(file_path: P) -> Result<i32, CliError> {
      let mut file = try!(File::open(file_path));
      let mut contents = String::new();
      try!(file.read_to_string(&mut contents));
      let n: i32 = try!(contents.trim().parse());
      Ok(2 * n)
  }

The only thing we did here was remove the calls to map_err. They are no longer needed because the try! macro invokes From::from on the error value. This works because we’ve provided From impls for all the error types that could appear.

If we modified our file_double function to perform some other operation, say, convert a string to a float, then we’d need to add a new variant to our error type:

use std::io;
  use std::num;
  
  enum CliError {
      Io(io::Error),
      ParseInt(num::ParseIntError),
      ParseFloat(num::ParseFloatError),
  }

And add a new From impl:


  use std::num;
  
  impl From<num::ParseFloatError> for CliError {
      fn from(err: num::ParseFloatError) -> CliError {
          CliError::ParseFloat(err)
      }
  }

And that’s it!

Advice for library writers

If your library needs to report custom errors, then you should probably define your own error type. It’s up to you whether or not to expose its representation (like ErrorKind) or keep it hidden (like ParseIntError). Regardless of how you do it, it’s usually good practice to at least provide some information about the error beyond just its String representation. But certainly, this will vary depending on use cases.

At a minimum, you should probably implement the Error trait. This will give users of your library some minimum flexibility for composing errors. Implementing the Error trait also means that users are guaranteed the ability to obtain a string representation of an error (because it requires impls for both fmt::Debug and fmt::Display).

Beyond that, it can also be useful to provide implementations of From on your error types. This allows you (the library author) and your users to compose more detailed errors. For example, csv::Error provides From impls for both io::Error and byteorder::Error.

Finally, depending on your tastes, you may also want to define a Result type alias, particularly if your library defines a single error type. This is used in the standard library for io::Result and fmt::Result.

Case study: A program to read population data

This chapter was long, and depending on your background, it might be rather dense. While there is plenty of example code to go along with the prose, most of it was specifically designed to be pedagogical. So, we’re going to do something new: a case study.

For this, we’re going to build up a command line program that lets you query world population data. The objective is simple: you give it a location and it will tell you the population. Despite the simplicity, there is a lot that can go wrong!

The data we’ll be using comes from the Data Science Toolkit. I’ve prepared some data from it for this exercise. You can either grab the world population data (41MB gzip compressed, 145MB uncompressed) or just the US population data (2.2MB gzip compressed, 7.2MB uncompressed).

Up until now, we’ve kept the code limited to Rust’s standard library. For a real task like this though, we’ll want to at least use something to parse CSV data, parse the program arguments and decode that stuff into Rust types automatically. For that, we’ll use the csv, and rustc-serialize crates.

Initial setup

We’re not going to spend a lot of time on setting up a project with Cargo because it is already covered well in the Cargo chapter and Cargo’s documentation.

To get started from scratch, run cargo new --bin city-pop and make sure your Cargo.toml looks something like this:

[package]
  name = "city-pop"
  version = "0.1.0"
  authors = ["Andrew Gallant <jamslam@gmail.com>"]
  
  [[bin]]
  name = "city-pop"
  
  [dependencies]
  csv = "0.*"
  rustc-serialize = "0.*"
  getopts = "0.*"

You should already be able to run:

cargo build --release
  ./target/release/city-pop
  # Outputs: Hello, world!

Argument parsing

Let’s get argument parsing out of the way. we won’t go into too much detail on Getopts, but there is some good documentation describing it. The short story is that Getopts generates an argument parser and a help message from a vector of options (The fact that it is a vector is hidden behind a struct and a set of methods). Once the parsing is done, we can decode the program arguments into a Rust struct. From there, we can get information about the flags, for instance, wether they were passed in, and what arguments they had. Here’s our program with the appropriate extern crate statements, and the basic argument setup for Getopts:

extern crate getopts;
  extern crate rustc_serialize;
  
  use getopts::Options;
  use std::env;
  
  fn print_usage(program: &str, opts: Options) {
      println!("{}", opts.usage(&format!("Usage: {} [options] <data-path> <city>", progr
  ↳ am)));
  }
  
  fn main() {
      let args: Vec<String> = env::args().collect();
      let program = args[0].clone();
  
      let mut opts = Options::new();
      opts.optflag("h", "help", "Show this usage message.");
      
      let matches = match opts.parse(&args[1..]) {
          Ok(m)  => { m }
      Err(e) => { panic!(e.to_string()) }
      };
      if matches.opt_present("h") {
          print_usage(&program, opts);
      return;
      }
      let data_path = args[1].clone();
      let city = args[2].clone();
      
      // Do stuff with information
  }

First, we get a vector of the arguments passed into our program. We then store the first one, knowing that it is our program’s name. Once that’s done, we set up our argument flags, in this case a simplistic help message flag. Once we have the argument flags set up, we use Options.parse to parse the argument vector (starting from index one, becouse index 0 is the program name). If this was successful, we assign matches to the parsed object, if not, we panic. Once past that, we test if the user passed in the help flag, and if so print the usage message. The option help messages are constructed by Getopts, so all we have to do to print the usage message is tell it what we want it to print for the program name and template. If the user has not passed in the help flag, we assign the proper variables to their corresponding arguments.

Writing the logic

We’re all different in how we write code, but error handling is usually the last thing we want to think about. This isn’t very good practice for good design, but it can be useful for rapidly prototyping. In our case, because Rust forces us to be explicit about error handling, it will also make it obvious what parts of our program can cause errors. Why? Because Rust will make us call unwrap! This can give us a nice bird’s eye view of how we need to approach error handling.

In this case study, the logic is really simple. All we need to do is parse the CSV data given to us and print out a field in matching rows. Let’s do it. (Make sure to add extern crate csv; to the top of your file.)

// This struct represents the data in each row of the CSV file.
  // Type based decoding absolves us of a lot of the nitty gritty error
  // handling, like parsing strings as integers or floats.
  #[derive(Debug, RustcDecodable)]
  struct Row {
      country: String,
      city: String,
      accent_city: String,
      region: String,
  
      // Not every row has data for the population, latitude or longitude!
      // So we express them as `Option` types, which admits the possibility of
      // absence. The CSV parser will fill in the correct value for us.
      population: Option<u64>,
      latitude: Option<f64>,
      longitude: Option<f64>,
  }
  
  fn print_usage(program: &str, opts: Options) {
      println!("{}", opts.usage(&format!("Usage: {} [options] <data-path> <city>", progr
  ↳ am)));
  }
  
  fn main() {
      let args: Vec<String> = env::args().collect();
      let program = args[0].clone();
  
      let mut opts = Options::new();
      opts.optflag("h", "help", "Show this usage message.");
      
      let matches = match opts.parse(&args[1..]) {
          Ok(m)  => { m }
          Err(e) => { panic!(e.to_string()) }
      };
      
      if matches.opt_present("h") {
          print_usage(&program, opts);
          return;
      }
          
      let data_file = args[1].clone();
      let data_path = Path::new(&data_file);
      let city = args[2].clone();
      
      let file = fs::File::open(data_path).unwrap();
      let mut rdr = csv::Reader::from_reader(file);
      
      for row in rdr.decode::<Row>() {
          let row = row.unwrap();
      
          if row.city == city {
              println!("{}, {}: {:?}",
                  row.city, row.country,
                  row.population.expect("population count"));
          }
      }
  }

Let’s outline the errors. We can start with the obvious: the three places that unwrap is called:

  1. fs::File::open can return an io::Error.
  2. csv::Reader::decode decodes one record at a time, and decoding a record (look at the Item associated type on the Iterator impl) can produce a csv::Error.
  3. If row.population is None, then calling expect will panic.

Are there any others? What if we can’t find a matching city? Tools like grep will return an error code, so we probably should too. So we have logic errors specific to our problem, IO errors and CSV parsing errors. We’re going to explore two different ways to approach handling these errors.

I’d like to start with Box<Error>. Later, we’ll see how defining our own error type can be useful.

Error handling with Box<Error>

Box<Error> is nice because it just works. You don’t need to define your own error types and you don’t need any From implementations. The downside is that since Box<Error> is a trait object, it erases the type, which means the compiler can no longer reason about its underlying type.

Previously we started refactoring our code by changing the type of our function from T to Result<T, OurErrorType>. In this case, OurErrorType is just Box<Error>. But what’s T? And can we add a return type to main?

The answer to the second question is no, we can’t. That means we’ll need to write a new function. But what is T? The simplest thing we can do is to return a list of matching Row values as a Vec<Row>. (Better code would return an iterator, but that is left as an exercise to the reader.)

Let’s refactor our code into its own function, but keep the calls to unwrap. Note that we opt to handle the possibility of a missing population count by simply ignoring that row.

struct Row {
      // unchanged
  }
  
  struct PopulationCount {
      city: String,
      country: String,
      // This is no longer an `Option` because values of this type are only
      // constructed if they have a population count.
      count: u64,
  }
  
  fn print_usage(program: &str, opts: Options) {
      println!("{}", opts.usage(&format!("Usage: {} [options] <data-path> <city>", progr
  ↳ am)));
  }
  
  fn search<P: AsRef<Path>>(file_path: P, city: &str) -> Vec<PopulationCount> {
      let mut found = vec![];
      let file = fs::File::open(file_path).unwrap();
      let mut rdr = csv::Reader::from_reader(file);
      for row in rdr.decode::<Row>() {
          let row = row.unwrap();
          match row.population {
              None => { } // skip it
              Some(count) => if row.city == city {
                  found.push(PopulationCount {
                      city: row.city,
                      country: row.country,
                      count: count,
                  });
              },
          }
      }
      found
  }
  
  fn main() {
      let args: Vec<String> = env::args().collect();
      let program = args[0].clone();
  
      let mut opts = Options::new();
      opts.optflag("h", "help", "Show this usage message.");
  
      let matches = match opts.parse(&args[1..]) {
          Ok(m)  => { m }
          Err(e) => { panic!(e.to_string()) }
      };
      if matches.opt_present("h") {
          print_usage(&program, opts);
          return;
      }
          
      let data_file = args[1].clone();
      let data_path = Path::new(&data_file);
      let city = args[2].clone();
      for pop in search(&data_path, &city) {
          println!("{}, {}: {:?}", pop.city, pop.country, pop.count);
      }
  }

While we got rid of one use of expect (which is a nicer variant of unwrap), we still should handle the absence of any search results.

To convert this to proper error handling, we need to do the following:

  1. Change the return type of search to be Result<Vec<PopulationCount>, Box<Error>>.
  2. Use the try! macro so that errors are returned to the caller instead of panicking the program.
  3. Handle the error in main.

Let’s try it:

fn search<P: AsRef<Path>>
           (file_path: P, city: &str)
           -> Result<Vec<PopulationCount>, Box<Error+Send+Sync>> {
      let mut found = vec![];
      let file = try!(fs::File::open(file_path));
      let mut rdr = csv::Reader::from_reader(file);
      for row in rdr.decode::<Row>() {
          let row = try!(row);
          match row.population {
              None => { } // skip it
              Some(count) => if row.city == city {
                  found.push(PopulationCount {
                      city: row.city,
                      country: row.country,
                      count: count,
                  });
              },
          }
      }
      if found.is_empty() {
          Err(From::from("No matching cities with a population were found."))
      } else {
          Ok(found)
      }
  }

Instead of x.unwrap(), we now have try!(x). Since our function returns a Result<T, E>, the try! macro will return early from the function if an error occurs.

There is one big gotcha in this code: we used Box<Error + Send + Sync> instead of Box<Error>. We did this so we could convert a plain string to an error type. We need these extra bounds so that we can use the corresponding From impls:

// We are making use of this impl in the code above, since we call `From::from`
  // on a `&'static str`.
  impl<'a, 'b> From<&'b str> for Box<Error + Send + Sync + 'a>
  
  // But this is also useful when you need to allocate a new string for an
  // error message, usually with `format!`.
  impl From<String> for Box<Error + Send + Sync>

Now that we’ve seen how to do proper error handling with Box<Error>, let’s try a different approach with our own custom error type. But first, let’s take a quick break from error handling and add support for reading from stdin.

Reading from stdin

In our program, we accept a single file for input and do one pass over the data. This means we probably should be able to accept input on stdin. But maybe we like the current format too—so let’s have both!

Adding support for stdin is actually quite easy. There are only two things we have to do:

  1. Tweak the program arguments so that a single parameter—the city—can be accepted while the population data is read from stdin.
  2. Modify the program so that an option -f can take the file, if it is not passed into stdin.
  3. Modify the search function to take an optional file path. When None, it should know to read from stdin.

First, here’s the new usage:

fn print_usage(program: &str, opts: Options) {
      println!("{}", opts.usage(&format!("Usage: {} [options] <city>", program)));
  }

The next part is going to be only a little harder:

...
  let mut opts = Options::new();
  opts.optopt("f", "file", "Choose an input file, instead of using STDIN.", "NAME");
  opts.optflag("h", "help", "Show this usage message.");
  ...
  let file = matches.opt_str("f");
  let data_file = file.as_ref().map(Path::new);
      
  let city = if !matches.free.is_empty() {
      matches.free[0].clone()
  } else {
      print_usage(&program, opts);
      return;
  };
  
  for pop in search(&data_file, &city) {
      println!("{}, {}: {:?}", pop.city, pop.country, pop.count);
  }
  ...

In this peice of code, we take file (which has the type Option<String>), and convert it to a type that search can use, in this case, &Option<AsRef<Path>>. Do do this, we take a reference of file, and map Path::new onto it. In this case, as_ref() converts the Option<String> into an Option<&str>, and from there, we can execute Path::new to the content of the optional, and return the optional of the new value. Once we have that, it is a simple matter of getting the city argument and executing search.

Modifying search is slightly trickier. The csv crate can build a parser out of any type that implements io::Read. But how can we use the same code over both types? There’s actually a couple ways we could go about this. One way is to write search such that it is generic on some type parameter R that satisfies io::Read. Another way is to just use trait objects:

fn search<P: AsRef<Path>>
           (file_path: &Option<P>, city: &str)
           -> Result<Vec<PopulationCount>, Box<Error+Send+Sync>> {
      let mut found = vec![];
      let input: Box<io::Read> = match *file_path {
          None => Box::new(io::stdin()),
          Some(ref file_path) => Box::new(try!(fs::File::open(file_path))),
      };
      let mut rdr = csv::Reader::from_reader(input);
      // The rest remains unchanged!
  }

Error handling with a custom type

Previously, we learned how to compose errors using a custom error type. We did this by defining our error type as an enum and implementing Error and From.

Since we have three distinct errors (IO, CSV parsing and not found), let’s define an enum with three variants:

#[derive(Debug)]
  enum CliError {
      Io(io::Error),
      Csv(csv::Error),
      NotFound,
  }

And now for impls on Display and Error:

impl fmt::Display for CliError {
      fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
          match *self {
              CliError::Io(ref err) => err.fmt(f),
              CliError::Csv(ref err) => err.fmt(f),
              CliError::NotFound => write!(f, "No matching cities with a \
                                               population were found."),
          }
      }
  }
  
  impl Error for CliError {
      fn description(&self) -> &str {
          match *self {
              CliError::Io(ref err) => err.description(),
              CliError::Csv(ref err) => err.description(),
              CliError::NotFound => "not found",
          }
      }
  }

Before we can use our CliError type in our search function, we need to provide a couple From impls. How do we know which impls to provide? Well, we’ll need to convert from both io::Error and csv::Error to CliError. Those are the only external errors, so we’ll only need two From impls for now:

impl From<io::Error> for CliError {
      fn from(err: io::Error) -> CliError {
          CliError::Io(err)
      }
  }
  
  impl From<csv::Error> for CliError {
      fn from(err: csv::Error) -> CliError {
          CliError::Csv(err)
      }
  }

The From impls are important because of how try! is defined. In particular, if an error occurs, From::from is called on the error, which in this case, will convert it to our own error type CliError.

With the From impls done, we only need to make two small tweaks to our search function: the return type and the “not found” error. Here it is in full:

fn search<P: AsRef<Path>>
           (file_path: &Option<P>, city: &str)
           -> Result<Vec<PopulationCount>, CliError> {
      let mut found = vec![];
      let input: Box<io::Read> = match *file_path {
          None => Box::new(io::stdin()),
          Some(ref file_path) => Box::new(try!(fs::File::open(file_path))),
      };
      let mut rdr = csv::Reader::from_reader(input);
      for row in rdr.decode::<Row>() {
          let row = try!(row);
          match row.population {
              None => { } // skip it
              Some(count) => if row.city == city {
                  found.push(PopulationCount {
                      city: row.city,
                      country: row.country,
                      count: count,
                  });
              },
          }
      }
      if found.is_empty() {
          Err(CliError::NotFound)
      } else {
          Ok(found)
      }
  }

No other changes are necessary.

Adding functionality

Writing generic code is great, because generalizing stuff is cool, and it can then be useful later. But sometimes, the juice isn’t worth the squeeze. Look at what we just did in the previous step:

  1. Defined a new error type.
  2. Added impls for Error, Display and two for From.

The big downside here is that our program didn’t improve a whole lot. There is quite a bit of overhead to representing errors with enums, especially in short programs like this.

One useful aspect of using a custom error type like we’ve done here is that the main function can now choose to handle errors differently. Previously, with Box<Error>, it didn’t have much of a choice: just print the message. We’re still doing that here, but what if we wanted to, say, add a --quiet flag? The --quiet flag should silence any verbose output.

Right now, if the program doesn’t find a match, it will output a message saying so. This can be a little clumsy, especially if you intend for the program to be used in shell scripts.

So let’s start by adding the flags. Like before, we need to tweak the usage string and add a flag to the Option variable. Once were done that, Getopts does the rest:

...
  let mut opts = Options::new();
  opts.optopt("f", "file", "Choose an input file, instead of using STDIN.", "NAME");
  opts.optflag("h", "help", "Show this usage message.");
  opts.optflag("q", "quit", "Silences errors and warnings.");
  ...

Now we just need to implement our “quiet” functionality. This requires us to tweak the case analysis in main:

match search(&args.arg_data_path, &args.arg_city) {
      Err(CliError::NotFound) if args.flag_quiet => process::exit(1),
      Err(err) => fatal!("{}", err),
      Ok(pops) => for pop in pops {
          println!("{}, {}: {:?}", pop.city, pop.country, pop.count);
      }
  }

Certainly, we don’t want to be quiet if there was an IO error or if the data failed to parse. Therefore, we use case analysis to check if the error type is NotFound and if --quiet has been enabled. If the search failed, we still quit with an exit code (following grep’s convention).

If we had stuck with Box<Error>, then it would be pretty tricky to implement the --quiet functionality.

This pretty much sums up our case study. From here, you should be ready to go out into the world and write your own programs and libraries with proper error handling.

The Short Story

Since this chapter is long, it is useful to have a quick summary for error handling in Rust. These are some good “rules of thumb.” They are emphatically not commandments. There are probably good reasons to break every one of these heuristics!

  • If you’re writing short example code that would be overburdened by error handling, it’s probably just fine to use unwrap (whether that’s Result::unwrap, Option::unwrap or preferably Option::expect). Consumers of your code should know to use proper error handling. (If they don’t, send them here!)
  • If you’re writing a quick ‘n’ dirty program, don’t feel ashamed if you use unwrap. Be warned: if it winds up in someone else’s hands, don’t be surprised if they are agitated by poor error messages!
  • If you’re writing a quick ‘n’ dirty program and feel ashamed about panicking anyway, then using either a String or a Box<Error + Send + Sync> for your error type (the Box<Error + Send + Sync> type is because of the available From impls).
  • Otherwise, in a program, define your own error types with appropriate From and Error impls to make the try! macro more ergnomic.
  • If you’re writing a library and your code can produce errors, define your own error type and implement the std::error::Error trait. Where appropriate, implement From to make both your library code and the caller’s code easier to write. (Because of Rust’s coherence rules, callers will not be able to impl From on your error type, so your library should do it.)
  • Learn the combinators defined on Option and Result. Using them exclusively can be a bit tiring at times, but I’ve personally found a healthy mix of try! and combinators to be quite appealing. and_then, map and unwrap_or are my favorites.

Choosing your Guarantees

One important feature of Rust is that it lets us control the costs and guarantees of a program.

There are various “wrapper type” abstractions in the Rust standard library which embody a multitude of tradeoffs between cost, ergonomics, and guarantees. Many let one choose between run time and compile time enforcement. This section will explain a few selected abstractions in detail.

Before proceeding, it is highly recommended that one reads about ownership and borrowing in Rust.

Basic pointer types

Box<T>

Box<T> is an “owned” pointer, or a “box”. While it can hand out references to the contained data, it is the only owner of the data. In particular, consider the following:

let x = Box::new(1);
  let y = x;
  // x no longer accessible here

Here, the box was moved into y. As x no longer owns it, the compiler will no longer allow the programmer to use x after this. A box can similarly be moved out of a function by returning it.

When a box (that hasn’t been moved) goes out of scope, destructors are run. These destructors take care of deallocating the inner data.

This is a zero-cost abstraction for dynamic allocation. If you want to allocate some memory on the heap and safely pass around a pointer to that memory, this is ideal. Note that you will only be allowed to share references to this by the regular borrowing rules, checked at compile time.

&T and &mut T

These are immutable and mutable references respectively. They follow the “read-write lock” pattern, such that one may either have only one mutable reference to some data, or any number of immutable ones, but not both. This guarantee is enforced at compile time, and has no visible cost at runtime. In most cases these two pointer types suffice for sharing cheap references between sections of code.

These pointers cannot be copied in such a way that they outlive the lifetime associated with them.

*const T and *mut T

These are C-like raw pointers with no lifetime or ownership attached to them. They just point to some location in memory with no other restrictions. The only guarantee that these provide is that they cannot be dereferenced except in code marked unsafe.

These are useful when building safe, low cost abstractions like Vec<T>, but should be avoided in safe code.

Rc<T>

This is the first wrapper we will cover that has a runtime cost.

Rc<T> is a reference counted pointer. In other words, this lets us have multiple “owning” pointers to the same data, and the data will be dropped (destructors will be run) when all pointers are out of scope.

Internally, it contains a shared “reference count” (also called “refcount”), which is incremented each time the Rc is cloned, and decremented each time one of the Rcs goes out of scope. The main responsibility of Rc<T> is to ensure that destructors are called for shared data.

The internal data here is immutable, and if a cycle of references is created, the data will be leaked. If we want data that doesn’t leak when there are cycles, we need a garbage collector.

Guarantees

The main guarantee provided here is that the data will not be destroyed until all references to it are out of scope.

This should be used when we wish to dynamically allocate and share some data (read-only) between various portions of your program, where it is not certain which portion will finish using the pointer last. It’s a viable alternative to &T when &T is either impossible to statically check for correctness, or creates extremely unergonomic code where the programmer does not wish to spend the development cost of working with.

This pointer is not thread safe, and Rust will not let it be sent or shared with other threads. This lets one avoid the cost of atomics in situations where they are unnecessary.

There is a sister smart pointer to this one, Weak<T>. This is a non-owning, but also non-borrowed, smart pointer. It is also similar to &T, but it is not restricted in lifetime—a Weak<T> can be held on to forever. However, it is possible that an attempt to access the inner data may fail and return None, since this can outlive the owned Rcs. This is useful for cyclic data structures and other things.

Cost

As far as memory goes, Rc<T> is a single allocation, though it will allocate two extra words (i.e. two usize values) as compared to a regular Box<T> (for “strong” and “weak” refcounts).

Rc<T> has the computational cost of incrementing/decrementing the refcount whenever it is cloned or goes out of scope respectively. Note that a clone will not do a deep copy, rather it will simply increment the inner reference count and return a copy of the Rc<T>.

Cell types

Cells provide interior mutability. In other words, they contain data which can be manipulated even if the type cannot be obtained in a mutable form (for example, when it is behind an &-ptr or Rc<T>).

The documentation for the cell module has a pretty good explanation for these.

These types are generally found in struct fields, but they may be found elsewhere too.

Cell<T>

Cell<T> is a type that provides zero-cost interior mutability, but only for Copy types. Since the compiler knows that all the data owned by the contained value is on the stack, there’s no worry of leaking any data behind references (or worse!) by simply replacing the data.

It is still possible to violate your own invariants using this wrapper, so be careful when using it. If a field is wrapped in Cell, it’s a nice indicator that the chunk of data is mutable and may not stay the same between the time you first read it and when you intend to use it.

use std::cell::Cell;
  
  let x = Cell::new(1);
  let y = &x;
  let z = &x;
  x.set(2);
  y.set(3);
  z.set(4);
  println!("{}", x.get());

Note that here we were able to mutate the same value from various immutable references.

This has the same runtime cost as the following:

let mut x = 1;
  let y = &mut x;
  let z = &mut x;
  x = 2;
  *y = 3;
  *z = 4;
  println!("{}", x);

but it has the added benefit of actually compiling successfully.

Guarantees

This relaxes the “no aliasing with mutability” restriction in places where it’s unnecessary. However, this also relaxes the guarantees that the restriction provides; so if your invariants depend on data stored within Cell, you should be careful.

This is useful for mutating primitives and other Copy types when there is no easy way of doing it in line with the static rules of & and &mut.

Cell does not let you obtain interior references to the data, which makes it safe to freely mutate.

Cost

There is no runtime cost to using Cell<T>, however if you are using it to wrap larger (Copy) structs, it might be worthwhile to instead wrap individual fields in Cell<T> since each write is otherwise a full copy of the struct.

RefCell<T>

RefCell<T> also provides interior mutability, but isn’t restricted to Copy types.

Instead, it has a runtime cost. RefCell<T> enforces the read-write lock pattern at runtime (it’s like a single-threaded mutex), unlike &T/&mut T which do so at compile time. This is done by the borrow() and borrow_mut() functions, which modify an internal reference count and return smart pointers which can be dereferenced immutably and mutably respectively. The refcount is restored when the smart pointers go out of scope. With this system, we can dynamically ensure that there are never any other borrows active when a mutable borrow is active. If the programmer attempts to make such a borrow, the thread will panic.

use std::cell::RefCell;
  
  let x = RefCell::new(vec![1,2,3,4]);
  {
      println!("{:?}", *x.borrow())
  }
  
  {
      let mut my_ref = x.borrow_mut();
      my_ref.push(1);
  }

Similar to Cell, this is mainly useful for situations where it’s hard or impossible to satisfy the borrow checker. Generally we know that such mutations won’t happen in a nested form, but it’s good to check.

For large, complicated programs, it becomes useful to put some things in RefCells to make things simpler. For example, a lot of the maps in the ctxt struct in the rust compiler internals are inside this wrapper. These are only modified once (during creation, which is not right after initialization) or a couple of times in well-separated places. However, since this struct is pervasively used everywhere, juggling mutable and immutable pointers would be hard (perhaps impossible) and probably form a soup of &-ptrs which would be hard to extend. On the other hand, the RefCell provides a cheap (not zero-cost) way of safely accessing these. In the future, if someone adds some code that attempts to modify the cell when it’s already borrowed, it will cause a (usually deterministic) panic which can be traced back to the offending borrow.

Similarly, in Servo’s DOM there is a lot of mutation, most of which is local to a DOM type, but some of which crisscrosses the DOM and modifies various things. Using RefCell and Cell to guard all mutation lets us avoid worrying about mutability everywhere, and it simultaneously highlights the places where mutation is actually happening.

Note that RefCell should be avoided if a mostly simple solution is possible with & pointers.

Guarantees

RefCell relaxes the static restrictions preventing aliased mutation, and replaces them with dynamic ones. As such the guarantees have not changed.

Cost

RefCell does not allocate, but it contains an additional “borrow state” indicator (one word in size) along with the data.

At runtime each borrow causes a modification/check of the refcount.

Synchronous types

Many of the types above cannot be used in a threadsafe manner. Particularly, Rc<T> and RefCell<T>, which both use non-atomic reference counts (atomic reference counts are those which can be incremented from multiple threads without causing a data race), cannot be used this way. This makes them cheaper to use, but we need thread safe versions of these too. They exist, in the form of Arc<T> and Mutex<T>/RwLock<T>

Note that the non-threadsafe types cannot be sent between threads, and this is checked at compile time.

There are many useful wrappers for concurrent programming in the sync module, but only the major ones will be covered below.

Arc<T>

Arc<T> is just a version of Rc<T> that uses an atomic reference count (hence, “Arc”). This can be sent freely between threads.

C++’s shared_ptr is similar to Arc, however in the case of C++ the inner data is always mutable. For semantics similar to that from C++, we should use Arc<Mutex<T>>, Arc<RwLock<T>>, or Arc<UnsafeCell<T>>2 (UnsafeCell<T> is a cell type that can be used to hold any data and has no runtime cost, but accessing it requires unsafe blocks). The last one should only be used if we are certain that the usage won’t cause any memory unsafety. Remember that writing to a struct is not an atomic operation, and many functions like vec.push() can reallocate internally and cause unsafe behavior, so even monotonicity may not be enough to justify UnsafeCell.

Guarantees

Like Rc, this provides the (thread safe) guarantee that the destructor for the internal data will be run when the last Arc goes out of scope (barring any cycles).

Cost

This has the added cost of using atomics for changing the refcount (which will happen whenever it is cloned or goes out of scope). When sharing data from an Arc in a single thread, it is preferable to share & pointers whenever possible.

Mutex<T> and RwLock<T>

Mutex<T> and RwLock<T> provide mutual-exclusion via RAII guards (guards are objects which maintain some state, like a lock, until their destructor is called). For both of these, the mutex is opaque until we call lock() on it, at which point the thread will block until a lock can be acquired, and then a guard will be returned. This guard can be used to access the inner data (mutably), and the lock will be released when the guard goes out of scope.

{
      let guard = mutex.lock();
      // guard dereferences mutably to the inner type
      *guard += 1;
  } // lock released when destructor runs

RwLock has the added benefit of being efficient for multiple reads. It is always safe to have multiple readers to shared data as long as there are no writers; and RwLock lets readers acquire a “read lock”. Such locks can be acquired concurrently and are kept track of via a reference count. Writers must obtain a “write lock” which can only be obtained when all readers have gone out of scope.

Guarantees

Both of these provide safe shared mutability across threads, however they are prone to deadlocks. Some level of additional protocol safety can be obtained via the type system.

Costs

These use internal atomic-like types to maintain the locks, which are pretty costly (they can block all memory reads across processors till they’re done). Waiting on these locks can also be slow when there’s a lot of concurrent access happening.

Composition

A common gripe when reading Rust code is with types like Rc<RefCell<Vec<T>>> (or even more more complicated compositions of such types). It’s not always clear what the composition does, or why the author chose one like this (and when one should be using such a composition in one’s own code)

Usually, it’s a case of composing together the guarantees that you need, without paying for stuff that is unnecessary.

For example, Rc<RefCell<T>> is one such composition. Rc<T> itself can’t be dereferenced mutably; because Rc<T> provides sharing and shared mutability can lead to unsafe behavior, so we put RefCell<T> inside to get dynamically verified shared mutability. Now we have shared mutable data, but it’s shared in a way that there can only be one mutator (and no readers) or multiple readers.

Now, we can take this a step further, and have Rc<RefCell<Vec<T>>> or Rc<Vec<RefCell<T>>>. These are both shareable, mutable vectors, but they’re not the same.

With the former, the RefCell<T> is wrapping the Vec<T>, so the Vec<T> in its entirety is mutable. At the same time, there can only be one mutable borrow of the whole Vec at a given time. This means that your code cannot simultaneously work on different elements of the vector from different Rc handles. However, we are able to push and pop from the Vec<T> at will. This is similar to an &mut Vec<T> with the borrow checking done at runtime.

With the latter, the borrowing is of individual elements, but the overall vector is immutable. Thus, we can independently borrow separate elements, but we cannot push or pop from the vector. This is similar to an &mut [T]3, but, again, the borrow checking is at runtime.

In concurrent programs, we have a similar situation with Arc<Mutex<T>>, which provides shared mutability and ownership.

When reading code that uses these, go in step by step and look at the guarantees/costs provided.

When choosing a composed type, we must do the reverse; figure out which guarantees we want, and at which point of the composition we need them. For example, if there is a choice between Vec<RefCell<T>> and RefCell<Vec<T>>, we should figure out the tradeoffs as done above and pick one.

FFI

Introduction

This guide will use the snappy compression/decompression library as an introduction to writing bindings for foreign code. Rust is currently unable to call directly into a C++ library, but snappy includes a C interface (documented in snappy-c.h).

The following is a minimal example of calling a foreign function which will compile if snappy is installed:

# #![feature(libc)]
  extern crate libc;
  use libc::size_t;
  
  #[link(name = "snappy")]
  extern {
      fn snappy_max_compressed_length(source_length: size_t) -> size_t;
  }
  
  fn main() {
      let x = unsafe { snappy_max_compressed_length(100) };
      println!("max compressed length of a 100 byte buffer: {}", x);
  }

The extern block is a list of function signatures in a foreign library, in this case with the platform’s C ABI. The #[link(...)] attribute is used to instruct the linker to link against the snappy library so the symbols are resolved.

Foreign functions are assumed to be unsafe so calls to them need to be wrapped with unsafe {} as a promise to the compiler that everything contained within truly is safe. C libraries often expose interfaces that aren’t thread-safe, and almost any function that takes a pointer argument isn’t valid for all possible inputs since the pointer could be dangling, and raw pointers fall outside of Rust’s safe memory model.

When declaring the argument types to a foreign function, the Rust compiler can not check if the declaration is correct, so specifying it correctly is part of keeping the binding correct at runtime.

The extern block can be extended to cover the entire snappy API:

# #![feature(libc)]
  extern crate libc;
  use libc::{c_int, size_t};
  
  #[link(name = "snappy")]
  extern {
      fn snappy_compress(input: *const u8,
                         input_length: size_t,
                         compressed: *mut u8,
                         compressed_length: *mut size_t) -> c_int;
      fn snappy_uncompress(compressed: *const u8,
                           compressed_length: size_t,
                           uncompressed: *mut u8,
                           uncompressed_length: *mut size_t) -> c_int;
      fn snappy_max_compressed_length(source_length: size_t) -> size_t;
      fn snappy_uncompressed_length(compressed: *const u8,
                                    compressed_length: size_t,
                                    result: *mut size_t) -> c_int;
      fn snappy_validate_compressed_buffer(compressed: *const u8,
                                           compressed_length: size_t) -> c_int;
  }
  # fn main() {}

Creating a safe interface

The raw C API needs to be wrapped to provide memory safety and make use of higher-level concepts like vectors. A library can choose to expose only the safe, high-level interface and hide the unsafe internal details.

Wrapping the functions which expect buffers involves using the slice::raw module to manipulate Rust vectors as pointers to memory. Rust’s vectors are guaranteed to be a contiguous block of memory. The length is number of elements currently contained, and the capacity is the total size in elements of the allocated memory. The length is less than or equal to the capacity.

pub fn validate_compressed_buffer(src: &[u8]) -> bool {
      unsafe {
          snappy_validate_compressed_buffer(src.as_ptr(), src.len() as size_t) == 0
      }
  }

The validate_compressed_buffer wrapper above makes use of an unsafe block, but it makes the guarantee that calling it is safe for all inputs by leaving off unsafe from the function signature.

The snappy_compress and snappy_uncompress functions are more complex, since a buffer has to be allocated to hold the output too.

The snappy_max_compressed_length function can be used to allocate a vector with the maximum required capacity to hold the compressed output. The vector can then be passed to the snappy_compress function as an output parameter. An output parameter is also passed to retrieve the true length after compression for setting the length.

pub fn compress(src: &[u8]) -> Vec<u8> {
      unsafe {
          let srclen = src.len() as size_t;
          let psrc = src.as_ptr();
  
          let mut dstlen = snappy_max_compressed_length(srclen);
          let mut dst = Vec::with_capacity(dstlen as usize);
          let pdst = dst.as_mut_ptr();
  
          snappy_compress(psrc, srclen, pdst, &mut dstlen);
          dst.set_len(dstlen as usize);
          dst
      }
  }

Decompression is similar, because snappy stores the uncompressed size as part of the compression format and snappy_uncompressed_length will retrieve the exact buffer size required.

pub fn uncompress(src: &[u8]) -> Option<Vec<u8>> {
      unsafe {
          let srclen = src.len() as size_t;
          let psrc = src.as_ptr();
  
          let mut dstlen: size_t = 0;
          snappy_uncompressed_length(psrc, srclen, &mut dstlen);
  
          let mut dst = Vec::with_capacity(dstlen as usize);
          let pdst = dst.as_mut_ptr();
  
          if snappy_uncompress(psrc, srclen, pdst, &mut dstlen) == 0 {
              dst.set_len(dstlen as usize);
              Some(dst)
          } else {
              None // SNAPPY_INVALID_INPUT
          }
      }
  }

For reference, the examples used here are also available as a library on GitHub.

Destructors

Foreign libraries often hand off ownership of resources to the calling code. When this occurs, we must use Rust’s destructors to provide safety and guarantee the release of these resources (especially in the case of panic).

For more about destructors, see the Drop trait.

Callbacks from C code to Rust functions

Some external libraries require the usage of callbacks to report back their current state or intermediate data to the caller. It is possible to pass functions defined in Rust to an external library. The requirement for this is that the callback function is marked as extern with the correct calling convention to make it callable from C code.

The callback function can then be sent through a registration call to the C library and afterwards be invoked from there.

A basic example is:

Rust code:

extern fn callback(a: i32) {
      println!("I'm called from C with value {0}", a);
  }
  
  #[link(name = "extlib")]
  extern {
     fn register_callback(cb: extern fn(i32)) -> i32;
     fn trigger_callback();
  }
  
  fn main() {
      unsafe {
          register_callback(callback);
          trigger_callback(); // Triggers the callback
      }
  }

C code:

typedef void (*rust_callback)(int32_t);
  rust_callback cb;
  
  int32_t register_callback(rust_callback callback) {
      cb = callback;
      return 1;
  }
  
  void trigger_callback() {
    cb(7); // Will call callback(7) in Rust
  }

In this example Rust’s main() will call trigger_callback() in C, which would, in turn, call back to callback() in Rust.

Targeting callbacks to Rust objects

The former example showed how a global function can be called from C code. However it is often desired that the callback is targeted to a special Rust object. This could be the object that represents the wrapper for the respective C object.

This can be achieved by passing an raw pointer to the object down to the C library. The C library can then include the pointer to the Rust object in the notification. This will allow the callback to unsafely access the referenced Rust object.

Rust code:

#[repr(C)]
  struct RustObject {
      a: i32,
      // other members
  }
  
  extern "C" fn callback(target: *mut RustObject, a: i32) {
      println!("I'm called from C with value {0}", a);
      unsafe {
          // Update the value in RustObject with the value received from the callback
          (*target).a = a;
      }
  }
  
  #[link(name = "extlib")]
  extern {
     fn register_callback(target: *mut RustObject,
                          cb: extern fn(*mut RustObject, i32)) -> i32;
     fn trigger_callback();
  }
  
  fn main() {
      // Create the object that will be referenced in the callback
      let mut rust_object = Box::new(RustObject { a: 5 });
  
      unsafe {
          register_callback(&mut *rust_object, callback);
          trigger_callback();
      }
  }

C code:

typedef void (*rust_callback)(void*, int32_t);
  void* cb_target;
  rust_callback cb;
  
  int32_t register_callback(void* callback_target, rust_callback callback) {
      cb_target = callback_target;
      cb = callback;
      return 1;
  }
  
  void trigger_callback() {
    cb(cb_target, 7); // Will call callback(&rustObject, 7) in Rust
  }

Asynchronous callbacks

In the previously given examples the callbacks are invoked as a direct reaction to a function call to the external C library. The control over the current thread is switched from Rust to C to Rust for the execution of the callback, but in the end the callback is executed on the same thread that called the function which triggered the callback.

Things get more complicated when the external library spawns its own threads and invokes callbacks from there. In these cases access to Rust data structures inside the callbacks is especially unsafe and proper synchronization mechanisms must be used. Besides classical synchronization mechanisms like mutexes, one possibility in Rust is to use channels (in std::sync::mpsc) to forward data from the C thread that invoked the callback into a Rust thread.

If an asynchronous callback targets a special object in the Rust address space it is also absolutely necessary that no more callbacks are performed by the C library after the respective Rust object gets destroyed. This can be achieved by unregistering the callback in the object’s destructor and designing the library in a way that guarantees that no callback will be performed after deregistration.

Linking

The link attribute on extern blocks provides the basic building block for instructing rustc how it will link to native libraries. There are two accepted forms of the link attribute today:

  • #[link(name = "foo")]
  • #[link(name = "foo", kind = "bar")]

In both of these cases, foo is the name of the native library that we’re linking to, and in the second case bar is the type of native library that the compiler is linking to. There are currently three known types of native libraries:

  • Dynamic - #[link(name = "readline")]
  • Static - #[link(name = "my_build_dependency", kind = "static")]
  • Frameworks - #[link(name = "CoreFoundation", kind = "framework")]

Note that frameworks are only available on OSX targets.

The different kind values are meant to differentiate how the native library participates in linkage. From a linkage perspective, the Rust compiler creates two flavors of artifacts: partial (rlib/staticlib) and final (dylib/binary). Native dynamic library and framework dependencies are propagated to the final artifact boundary, while static library dependencies are not propagated at all, because the static libraries are integrated directly into the subsequent artifact.

A few examples of how this model can be used are:

  • A native build dependency. Sometimes some C/C++ glue is needed when writing some Rust code, but distribution of the C/C++ code in a library format is just a burden. In this case, the code will be archived into libfoo.a and then the Rust crate would declare a dependency via #[link(name = "foo", kind = "static")].

Regardless of the flavor of output for the crate, the native static library will be included in the output, meaning that distribution of the native static library is not necessary.

  • A normal dynamic dependency. Common system libraries (like readline) are available on a large number of systems, and often a static copy of these libraries cannot be found. When this dependency is included in a Rust crate, partial targets (like rlibs) will not link to the library, but when the rlib is included in a final target (like a binary), the native library will be linked in.

On OSX, frameworks behave with the same semantics as a dynamic library.

Unsafe blocks

Some operations, like dereferencing raw pointers or calling functions that have been marked unsafe are only allowed inside unsafe blocks. Unsafe blocks isolate unsafety and are a promise to the compiler that the unsafety does not leak out of the block.

Unsafe functions, on the other hand, advertise it to the world. An unsafe function is written like this:

unsafe fn kaboom(ptr: *const i32) -> i32 { *ptr }

This function can only be called from an unsafe block or another unsafe function.

Accessing foreign globals

Foreign APIs often export a global variable which could do something like track global state. In order to access these variables, you declare them in extern blocks with the static keyword:

# #![feature(libc)]
  extern crate libc;
  
  #[link(name = "readline")]
  extern {
      static rl_readline_version: libc::c_int;
  }
  
  fn main() {
      println!("You have readline version {} installed.",
               rl_readline_version as i32);
  }

Alternatively, you may need to alter global state provided by a foreign interface. To do this, statics can be declared with mut so we can mutate them.

# #![feature(libc)]
  extern crate libc;
  
  use std::ffi::CString;
  use std::ptr;
  
  #[link(name = "readline")]
  extern {
      static mut rl_prompt: *const libc::c_char;
  }
  
  fn main() {
      let prompt = CString::new("[my-awesome-shell] $").unwrap();
      unsafe {
          rl_prompt = prompt.as_ptr();
  
          println!("{:?}", rl_prompt);
  
          rl_prompt = ptr::null();
      }
  }

Note that all interaction with a static mut is unsafe, both reading and writing. Dealing with global mutable state requires a great deal of care.

Foreign calling conventions

Most foreign code exposes a C ABI, and Rust uses the platform’s C calling convention by default when calling foreign functions. Some foreign functions, most notably the Windows API, use other calling conventions. Rust provides a way to tell the compiler which convention to use:

extern crate libc;
  
  #[cfg(all(target_os = "win32", target_arch = "x86"))]
  #[link(name = "kernel32")]
  #[allow(non_snake_case)]
  extern "stdcall" {
      fn SetEnvironmentVariableA(n: *const u8, v: *const u8) -> libc::c_int;
  }

This applies to the entire extern block. The list of supported ABI constraints are:

  • stdcall
  • aapcs
  • cdecl
  • fastcall
  • Rust
  • rust-intrinsic
  • system
  • C
  • win64

Most of the abis in this list are self-explanatory, but the system abi may seem a little odd. This constraint selects whatever the appropriate ABI is for interoperating with the target’s libraries. For example, on win32 with a x86 architecture, this means that the abi used would be stdcall. On x86_64, however, windows uses the C calling convention, so C would be used. This means that in our previous example, we could have used extern "system" { ... } to define a block for all windows systems, not just x86 ones.

Interoperability with foreign code

Rust guarantees that the layout of a struct is compatible with the platform’s representation in C only if the #[repr(C)] attribute is applied to it. #[repr(C, packed)] can be used to lay out struct members without padding. #[repr(C)] can also be applied to an enum.

Rust’s owned boxes (Box<T>) use non-nullable pointers as handles which point to the contained object. However, they should not be manually created because they are managed by internal allocators. References can safely be assumed to be non-nullable pointers directly to the type. However, breaking the borrow checking or mutability rules is not guaranteed to be safe, so prefer using raw pointers (*) if that’s needed because the compiler can’t make as many assumptions about them.

Vectors and strings share the same basic memory layout, and utilities are available in the vec and str modules for working with C APIs. However, strings are not terminated with \0. If you need a NUL-terminated string for interoperability with C, you should use the CString type in the std::ffi module.

The libc crate on crates.io includes type aliases and function definitions for the C standard library in the libc module, and Rust links against libc and libm by default.

The “nullable pointer optimization”

Certain types are defined to not be null. This includes references (&T, &mut T), boxes (Box<T>), and function pointers (extern "abi" fn()). When interfacing with C, pointers that might be null are often used. As a special case, a generic enum that contains exactly two variants, one of which contains no data and the other containing a single field, is eligible for the “nullable pointer optimization”. When such an enum is instantiated with one of the non-nullable types, it is represented as a single pointer, and the non-data variant is represented as the null pointer. So Option<extern "C" fn(c_int) -> c_int> is how one represents a nullable function pointer using the C ABI.

Calling Rust code from C

You may wish to compile Rust code in a way so that it can be called from C. This is fairly easy, but requires a few things:

#[no_mangle]
  pub extern fn hello_rust() -> *const u8 {
      "Hello, world!\0".as_ptr()
  }

The extern makes this function adhere to the C calling convention, as discussed above in “Foreign Calling Conventions”. The no_mangle attribute turns off Rust’s name mangling, so that it is easier to link to.

FFI and panics

It’s important to be mindful of panic!s when working with FFI. A panic! across an FFI boundary is undefined behavior. If you’re writing code that may panic, you should run it in another thread, so that the panic doesn’t bubble up to C:

use std::thread;
  
  #[no_mangle]
  pub extern fn oh_no() -> i32 {
      let h = thread::spawn(|| {
          panic!("Oops!");
      });
  
      match h.join() {
          Ok(_) => 1,
          Err(_) => 0,
      }
  }

Representing opaque structs

Sometimes, a C library wants to provide a pointer to something, but not let you know the internal details of the thing it wants. The simplest way is to use a void * argument:

void foo(void *arg);
  void bar(void *arg);

We can represent this in Rust with the c_void type:

extern crate libc;
  
  extern "C" {
      pub fn foo(arg: *mut libc::c_void);
      pub fn bar(arg: *mut libc::c_void);
  }

This is a perfectly valid way of handling the situation. However, we can do a bit better. To solve this, some C libraries will instead create a struct, where the details and memory layout of the struct are private. This gives some amount of type safety. These structures are called ‘opaque’. Here’s an example, in C:

struct Foo; /* Foo is a structure, but its contents are not part of the public interfa
  ↳ ce */
  struct Bar;
  void foo(struct Foo *arg);
  void bar(struct Bar *arg);

To do this in Rust, let’s create our own opaque types with enum:

pub enum Foo {}
  pub enum Bar {}
  
  extern "C" {
      pub fn foo(arg: *mut Foo);
      pub fn bar(arg: *mut Bar);
  }

By using an enum with no variants, we create an opaque type that we can’t instantiate, as it has no variants. But because our Foo and Bar types are different, we’ll get type safety between the two of them, so we cannot accidentally pass a pointer to Foo to bar().

Borrow and AsRef

The Borrow and AsRef traits are very similar, but different. Here’s a quick refresher on what these two traits mean.

Borrow

The Borrow trait is used when you’re writing a datastructure, and you want to use either an owned or borrowed type as synonymous for some purpose.

For example, HashMap has a get method which uses Borrow:

fn get<Q: ?Sized>(&self, k: &Q) -> Option<&V>
      where K: Borrow<Q>,
            Q: Hash + Eq

This signature is pretty complicated. The K parameter is what we’re interested in here. It refers to a parameter of the HashMap itself:

struct HashMap<K, V, S = RandomState> {

The K parameter is the type of key the HashMap uses. So, looking at the signature of get() again, we can use get() when the key implements Borrow<Q>. That way, we can make a HashMap which uses String keys, but use &strs when we’re searching:

use std::collections::HashMap;
  
  let mut map = HashMap::new();
  map.insert("Foo".to_string(), 42);
  
  assert_eq!(map.get("Foo"), Some(&42));

This is because the standard library has impl Borrow<str> for String.

For most types, when you want to take an owned or borrowed type, a &T is enough. But one area where Borrow is effective is when there’s more than one kind of borrowed value. This is especially true of references and slices: you can have both an &T or a &mut T. If we wanted to accept both of these types, Borrow is up for it:

use std::borrow::Borrow;
  use std::fmt::Display;
  
  fn foo<T: Borrow<i32> + Display>(a: T) {
      println!("a is borrowed: {}", a);
  }
  
  let mut i = 5;
  
  foo(&i);
  foo(&mut i);

This will print out a is borrowed: 5 twice.

AsRef

The AsRef trait is a conversion trait. It’s used for converting some value to a reference in generic code. Like this:

let s = "Hello".to_string();
  
  fn foo<T: AsRef<str>>(s: T) {
      let slice = s.as_ref();
  }

Which should I use?

We can see how they’re kind of the same: they both deal with owned and borrowed versions of some type. However, they’re a bit different.

Choose Borrow when you want to abstract over different kinds of borrowing, or when you’re building a datastructure that treats owned and borrowed values in equivalent ways, such as hashing and comparison.

Choose AsRef when you want to convert something to a reference directly, and you’re writing generic code.

Release Channels

The Rust project uses a concept called ‘release channels’ to manage releases. It’s important to understand this process to choose which version of Rust your project should use.

Overview

There are three channels for Rust releases:

  • Nightly
  • Beta
  • Stable

New nightly releases are created once a day. Every six weeks, the latest nightly release is promoted to ‘Beta’. At that point, it will only receive patches to fix serious errors. Six weeks later, the beta is promoted to ‘Stable’, and becomes the next release of 1.x.

This process happens in parallel. So every six weeks, on the same day, nightly goes to beta, beta goes to stable. When 1.x is released, at the same time, 1.(x + 1)-beta is released, and the nightly becomes the first version of 1.(x + 2)-nightly.

Choosing a version

Generally speaking, unless you have a specific reason, you should be using the stable release channel. These releases are intended for a general audience.

However, depending on your interest in Rust, you may choose to use nightly instead. The basic tradeoff is this: in the nightly channel, you can use unstable, new Rust features. However, unstable features are subject to change, and so any new nightly release may break your code. If you use the stable release, you cannot use experimental features, but the next release of Rust will not cause significant issues through breaking changes.

Helping the ecosystem through CI

What about beta? We encourage all Rust users who use the stable release channel to also test against the beta channel in their continuous integration systems. This will help alert the team in case there’s an accidental regression.

Additionally, testing against nightly can catch regressions even sooner, and so if you don’t mind a third build, we’d appreciate testing against all channels.

As an example, many Rust programmers use Travis to test their crates, which is free for open source projects. Travis supports Rust directly, and you can use a .travis.yml file like this to test on all channels:

language: rust
  rust:
    - nightly
    - beta
    - stable
  
  matrix:
    allow_failures:
      - rust: nightly

With this configuration, Travis will test all three channels, but if something breaks on nightly, it won’t fail your build. A similar configuration is recommended for any CI system, check the documentation of the one you’re using for more details.

Syntax and Semantics

This section breaks Rust down into small chunks, one for each concept.

If you’d like to learn Rust from the bottom up, reading this in order is a great way to do that.

These sections also form a reference for each concept, so if you’re reading another tutorial and find something confusing, you can find it explained somewhere in here.

Variable Bindings

Virtually every non-‘Hello World’ Rust program uses variable bindings. They look like this:

fn main() {
      let x = 5;
  }

Putting fn main() { in each example is a bit tedious, so we’ll leave that out in the future. If you’re following along, make sure to edit your main() function, rather than leaving it off. Otherwise, you’ll get an error.

In many languages, this is called a variable, but Rust’s variable bindings have a few tricks up their sleeves. For example the left-hand side of a let expression is a ‘pattern’, not just a variable name. This means we can do things like:

let (x, y) = (1, 2);

After this expression is evaluated, x will be one, and y will be two. Patterns are really powerful, and have their own section in the book. We don’t need those features for now, so we’ll just keep this in the back of our minds as we go forward.

Rust is a statically typed language, which means that we specify our types up front, and they’re checked at compile time. So why does our first example compile? Well, Rust has this thing called ‘type inference’. If it can figure out what the type of something is, Rust doesn’t require you to actually type it out.

We can add the type if we want to, though. Types come after a colon (:):

let x: i32 = 5;

If I asked you to read this out loud to the rest of the class, you’d say “x is a binding with the type i32 and the value five.”

In this case we chose to represent x as a 32-bit signed integer. Rust has many different primitive integer types. They begin with i for signed integers and u for unsigned integers. The possible integer sizes are 8, 16, 32, and 64 bits.

In future examples, we may annotate the type in a comment. The examples will look like this:

fn main() {
      let x = 5; // x: i32
  }

Note the similarities between this annotation and the syntax you use with let. Including these kinds of comments is not idiomatic Rust, but we’ll occasionally include them to help you understand what the types that Rust infers are.

By default, bindings are immutable. This code will not compile:

let x = 5;
  x = 10;

It will give you this error:

error: re-assignment of immutable variable `x`
       x = 10;
       ^~~~~~~

If you want a binding to be mutable, you can use mut:

let mut x = 5; // mut x: i32
  x = 10;

There is no single reason that bindings are immutable by default, but we can think about it through one of Rust’s primary focuses: safety. If you forget to say mut, the compiler will catch it, and let you know that you have mutated something you may not have intended to mutate. If bindings were mutable by default, the compiler would not be able to tell you this. If you did intend mutation, then the solution is quite easy: add mut.

There are other good reasons to avoid mutable state when possible, but they’re out of the scope of this guide. In general, you can often avoid explicit mutation, and so it is preferable in Rust. That said, sometimes, mutation is what you need, so it’s not verboten.

Let’s get back to bindings. Rust variable bindings have one more aspect that differs from other languages: bindings are required to be initialized with a value before you’re allowed to use them.

Let’s try it out. Change your src/main.rs file to look like this:

fn main() {
      let x: i32;
  
      println!("Hello world!");
  }

You can use cargo build on the command line to build it. You’ll get a warning, but it will still print “Hello, world!”:

   Compiling hello_world v0.0.1 (file:///home/you/projects/hello_world)
  src/main.rs:2:9: 2:10 warning: unused variable: `x`, #[warn(unused_variable)]
     on by default
  src/main.rs:2     let x: i32;
                        ^

Rust warns us that we never use the variable binding, but since we never use it, no harm, no foul. Things change if we try to actually use this x, however. Let’s do that. Change your program to look like this:

fn main() {
      let x: i32;
  
      println!("The value of x is: {}", x);
  }

And try to build it. You’ll get an error:

$ cargo build
     Compiling hello_world v0.0.1 (file:///home/you/projects/hello_world)
  src/main.rs:4:39: 4:40 error: use of possibly uninitialized variable: `x`
  src/main.rs:4     println!("The value of x is: {}", x);
                                                      ^
  note: in expansion of format_args!
  <std macros>:2:23: 2:77 note: expansion site
  <std macros>:1:1: 3:2 note: in expansion of println!
  src/main.rs:4:5: 4:42 note: expansion site
  error: aborting due to previous error
  Could not compile `hello_world`.

Rust will not let us use a value that has not been initialized. Next, let’s talk about this stuff we’ve added to println!.

If you include two curly braces ({}, some call them moustaches…) in your string to print, Rust will interpret this as a request to interpolate some sort of value. String interpolation is a computer science term that means “stick in the middle of a string.” We add a comma, and then x, to indicate that we want x to be the value we’re interpolating. The comma is used to separate arguments we pass to functions and macros, if you’re passing more than one.

When you just use the curly braces, Rust will attempt to display the value in a meaningful way by checking out its type. If you want to specify the format in a more detailed manner, there are a wide number of options available. For now, we’ll just stick to the default: integers aren’t very complicated to print.

Functions

Every Rust program has at least one function, the main function:

fn main() {
  }

This is the simplest possible function declaration. As we mentioned before, fn says ‘this is a function’, followed by the name, some parentheses because this function takes no arguments, and then some curly braces to indicate the body. Here’s a function named foo:

fn foo() {
  }

So, what about taking arguments? Here’s a function that prints a number:

fn print_number(x: i32) {
      println!("x is: {}", x);
  }

Here’s a complete program that uses print_number:

fn main() {
      print_number(5);
  }
  
  fn print_number(x: i32) {
      println!("x is: {}", x);
  }

As you can see, function arguments work very similar to let declarations: you add a type to the argument name, after a colon.

Here’s a complete program that adds two numbers together and prints them:

fn main() {
      print_sum(5, 6);
  }
  
  fn print_sum(x: i32, y: i32) {
      println!("sum is: {}", x + y);
  }

You separate arguments with a comma, both when you call the function, as well as when you declare it.

Unlike let, you must declare the types of function arguments. This does not work:

fn print_sum(x, y) {
      println!("sum is: {}", x + y);
  }

You get this error:

expected one of `!`, `:`, or `@`, found `)`
  fn print_number(x, y) {

This is a deliberate design decision. While full-program inference is possible, languages which have it, like Haskell, often suggest that documenting your types explicitly is a best-practice. We agree that forcing functions to declare types while allowing for inference inside of function bodies is a wonderful sweet spot between full inference and no inference.

What about returning a value? Here’s a function that adds one to an integer:

fn add_one(x: i32) -> i32 {
      x + 1
  }

Rust functions return exactly one value, and you declare the type after an ‘arrow’, which is a dash (-) followed by a greater-than sign (>). The last line of a function determines what it returns. You’ll note the lack of a semicolon here. If we added it in:

fn add_one(x: i32) -> i32 {
      x + 1;
  }

We would get an error:

error: not all control paths return a value
  fn add_one(x: i32) -> i32 {
       x + 1;
  }
  
  help: consider removing this semicolon:
       x + 1;
            ^

This reveals two interesting things about Rust: it is an expression-based language, and semicolons are different from semicolons in other ‘curly brace and semicolon’-based languages. These two things are related.

Expressions vs. Statements

Rust is primarily an expression-based language. There are only two kinds of statements, and everything else is an expression.

So what’s the difference? Expressions return a value, and statements do not. That’s why we end up with ‘not all control paths return a value’ here: the statement x + 1; doesn’t return a value. There are two kinds of statements in Rust: ‘declaration statements’ and ‘expression statements’. Everything else is an expression. Let’s talk about declaration statements first.

In some languages, variable bindings can be written as expressions, not just statements. Like Ruby:

x = y = 5

In Rust, however, using let to introduce a binding is not an expression. The following will produce a compile-time error:

let x = (let y = 5); // expected identifier, found keyword `let`

The compiler is telling us here that it was expecting to see the beginning of an expression, and a let can only begin a statement, not an expression.

Note that assigning to an already-bound variable (e.g. y = 5) is still an expression, although its value is not particularly useful. Unlike other languages where an assignment evaluates to the assigned value (e.g. 5 in the previous example), in Rust the value of an assignment is an empty tuple () because the assigned value can have just one owner, and any other returned value would be too surprising:

let mut y = 5;
  
  let x = (y = 6);  // x has the value `()`, not `6`

The second kind of statement in Rust is the expression statement. Its purpose is to turn any expression into a statement. In practical terms, Rust’s grammar expects statements to follow other statements. This means that you use semicolons to separate expressions from each other. This means that Rust looks a lot like most other languages that require you to use semicolons at the end of every line, and you will see semicolons at the end of almost every line of Rust code you see.

What is this exception that makes us say “almost”? You saw it already, in this code:

fn add_one(x: i32) -> i32 {
      x + 1
  }

Our function claims to return an i32, but with a semicolon, it would return () instead. Rust realizes this probably isn’t what we want, and suggests removing the semicolon in the error we saw before.

Early returns

But what about early returns? Rust does have a keyword for that, return:

fn foo(x: i32) -> i32 {
      return x;
  
      // we never run this code!
      x + 1
  }

Using a return as the last line of a function works, but is considered poor style:

fn foo(x: i32) -> i32 {
      return x + 1;
  }

The previous definition without return may look a bit strange if you haven’t worked in an expression-based language before, but it becomes intuitive over time.

Diverging functions

Rust has some special syntax for ‘diverging functions’, which are functions that do not return:

fn diverges() -> ! {
      panic!("This function never returns!");
  }

panic! is a macro, similar to println!() that we’ve already seen. Unlike println!(), panic!() causes the current thread of execution to crash with the given message. Because this function will cause a crash, it will never return, and so it has the type ‘!’, which is read ‘diverges’.

If you add a main function that calls diverges() and run it, you’ll get some output that looks like this:

thread ‘<main>’ panicked at ‘This function never returns!’, hello.rs:2

If you want more information, you can get a backtrace by setting the RUST_BACKTRACE environment variable:

$ RUST_BACKTRACE=1 ./diverges
  thread '<main>' panicked at 'This function never returns!', hello.rs:2
  stack backtrace:
     1:     0x7f402773a829 - sys::backtrace::write::h0942de78b6c02817K8r
     2:     0x7f402773d7fc - panicking::on_panic::h3f23f9d0b5f4c91bu9w
     3:     0x7f402773960e - rt::unwind::begin_unwind_inner::h2844b8c5e81e79558Bw
     4:     0x7f4027738893 - rt::unwind::begin_unwind::h4375279447423903650
     5:     0x7f4027738809 - diverges::h2266b4c4b850236beaa
     6:     0x7f40277389e5 - main::h19bb1149c2f00ecfBaa
     7:     0x7f402773f514 - rt::unwind::try::try_fn::h13186883479104382231
     8:     0x7f402773d1d8 - __rust_try
     9:     0x7f402773f201 - rt::lang_start::ha172a3ce74bb453aK5w
    10:     0x7f4027738a19 - main
    11:     0x7f402694ab44 - __libc_start_main
    12:     0x7f40277386c8 - <unknown>
    13:                0x0 - <unknown>

RUST_BACKTRACE also works with Cargo’s run command:

$ RUST_BACKTRACE=1 cargo run
       Running `target/debug/diverges`
  thread '<main>' panicked at 'This function never returns!', hello.rs:2
  stack backtrace:
     1:     0x7f402773a829 - sys::backtrace::write::h0942de78b6c02817K8r
     2:     0x7f402773d7fc - panicking::on_panic::h3f23f9d0b5f4c91bu9w
     3:     0x7f402773960e - rt::unwind::begin_unwind_inner::h2844b8c5e81e79558Bw
     4:     0x7f4027738893 - rt::unwind::begin_unwind::h4375279447423903650
     5:     0x7f4027738809 - diverges::h2266b4c4b850236beaa
     6:     0x7f40277389e5 - main::h19bb1149c2f00ecfBaa
     7:     0x7f402773f514 - rt::unwind::try::try_fn::h13186883479104382231
     8:     0x7f402773d1d8 - __rust_try
     9:     0x7f402773f201 - rt::lang_start::ha172a3ce74bb453aK5w
    10:     0x7f4027738a19 - main
    11:     0x7f402694ab44 - __libc_start_main
    12:     0x7f40277386c8 - <unknown>
    13:                0x0 - <unknown>

A diverging function can be used as any type:

# fn diverges() -> ! {
  #    panic!("This function never returns!");
  # }
  let x: i32 = diverges();
  let x: String = diverges();

Function pointers

We can also create variable bindings which point to functions:

let f: fn(i32) -> i32;

f is a variable binding which points to a function that takes an i32 as an argument and returns an i32. For example:

fn plus_one(i: i32) -> i32 {
      i + 1
  }
  
  // without type inference
  let f: fn(i32) -> i32 = plus_one;
  
  // with type inference
  let f = plus_one;

We can then use f to call the function:

let six = f(5);

Primitive Types

The Rust language has a number of types that are considered ‘primitive’. This means that they’re built-in to the language. Rust is structured in such a way that the standard library also provides a number of useful types built on top of these ones, as well, but these are the most primitive.

Booleans

Rust has a built in boolean type, named bool. It has two values, true and false:

let x = true;
  
  let y: bool = false;

A common use of booleans is in if conditionals.

You can find more documentation for bools in the standard library documentation.

char

The char type represents a single Unicode scalar value. You can create chars with a single tick: (')

let x = 'x';
  let two_hearts = '💕';

Unlike some other languages, this means that Rust’s char is not a single byte, but four.

You can find more documentation for chars in the standard library documentation.

Numeric types

Rust has a variety of numeric types in a few categories: signed and unsigned, fixed and variable, floating-point and integer.

These types consist of two parts: the category, and the size. For example, u16 is an unsigned type with sixteen bits of size. More bits lets you have bigger numbers.

If a number literal has nothing to cause its type to be inferred, it defaults:

let x = 42; // x has type i32
  
  let y = 1.0; // y has type f64

Here’s a list of the different numeric types, with links to their documentation in the standard library:

Let’s go over them by category:

Signed and Unsigned

Integer types come in two varieties: signed and unsigned. To understand the difference, let’s consider a number with four bits of size. A signed, four-bit number would let you store numbers from -8 to +7. Signed numbers use “two’s complement representation”. An unsigned four bit number, since it does not need to store negatives, can store values from 0 to +15.

Unsigned types use a u for their category, and signed types use i. The i is for ‘integer’. So u8 is an eight-bit unsigned number, and i8 is an eight-bit signed number.

Fixed size types

Fixed size types have a specific number of bits in their representation. Valid bit sizes are 8, 16, 32, and 64. So, u32 is an unsigned, 32-bit integer, and i64 is a signed, 64-bit integer.

Variable sized types

Rust also provides types whose size depends on the size of a pointer of the underlying machine. These types have ‘size’ as the category, and come in signed and unsigned varieties. This makes for two types: isize and usize.

Floating-point types

Rust also has two floating point types: f32 and f64. These correspond to IEEE-754 single and double precision numbers.

Arrays

Like many programming languages, Rust has list types to represent a sequence of things. The most basic is the array, a fixed-size list of elements of the same type. By default, arrays are immutable.

let a = [1, 2, 3]; // a: [i32; 3]
  let mut m = [1, 2, 3]; // m: [i32; 3]

Arrays have type [T; N]. We’ll talk about this T notation in the generics section. The N is a compile-time constant, for the length of the array.

There’s a shorthand for initializing each element of an array to the same value. In this example, each element of a will be initialized to 0:

let a = [0; 20]; // a: [i32; 20]

You can get the number of elements in an array a with a.len():

let a = [1, 2, 3];
  
  println!("a has {} elements", a.len());

You can access a particular element of an array with subscript notation:

let names = ["Graydon", "Brian", "Niko"]; // names: [&str; 3]
  
  println!("The second name is: {}", names[1]);

Subscripts start at zero, like in most programming languages, so the first name is names[0] and the second name is names[1]. The above example prints The second name is: Brian. If you try to use a subscript that is not in the array, you will get an error: array access is bounds-checked at run-time. Such errant access is the source of many bugs in other systems programming languages.

You can find more documentation for arrays in the standard library documentation.

Slices

A ‘slice’ is a reference to (or “view” into) another data structure. They are useful for allowing safe, efficient access to a portion of an array without copying. For example, you might want to reference just one line of a file read into memory. By nature, a slice is not created directly, but from an existing variable. Slices have a length, can be mutable or not, and in many ways behave like arrays:

let a = [0, 1, 2, 3, 4];
  let middle = &a[1..4]; // A slice of a: just the elements 1, 2, and 3
  let complete = &a[..]; // A slice containing all of the elements in a

Slices have type &[T]. We’ll talk about that T when we cover generics.

You can find more documentation for slices in the standard library documentation.

str

Rust’s str type is the most primitive string type. As an unsized type, it’s not very useful by itself, but becomes useful when placed behind a reference, like &str. As such, we’ll just leave it at that.

You can find more documentation for str in the standard library documentation.

Tuples

A tuple is an ordered list of fixed size. Like this:

let x = (1, "hello");

The parentheses and commas form this two-length tuple. Here’s the same code, but with the type annotated:

let x: (i32, &str) = (1, "hello");

As you can see, the type of a tuple looks just like the tuple, but with each position having a type name rather than the value. Careful readers will also note that tuples are heterogeneous: we have an i32 and a &str in this tuple. In systems programming languages, strings are a bit more complex than in other languages. For now, just read &str as a string slice, and we’ll learn more soon.

You can assign one tuple into another, if they have the same contained types and arity. Tuples have the same arity when they have the same length.

let mut x = (1, 2); // x: (i32, i32)
  let y = (2, 3); // y: (i32, i32)
  
  x = y;

You can access the fields in a tuple through a destructuring let. Here’s an example:

let (x, y, z) = (1, 2, 3);
  
  println!("x is {}", x);

Remember before when I said the left-hand side of a let statement was more powerful than just assigning a binding? Here we are. We can put a pattern on the left-hand side of the let, and if it matches up to the right-hand side, we can assign multiple bindings at once. In this case, let “destructures” or “breaks up” the tuple, and assigns the bits to three bindings.

This pattern is very powerful, and we’ll see it repeated more later.

You can disambiguate a single-element tuple from a value in parentheses with a comma:

(0,); // single-element tuple
  (0); // zero in parentheses

Tuple Indexing

You can also access fields of a tuple with indexing syntax:

let tuple = (1, 2, 3);
  
  let x = tuple.0;
  let y = tuple.1;
  let z = tuple.2;
  
  println!("x is {}", x);

Like array indexing, it starts at zero, but unlike array indexing, it uses a ., rather than []s.

You can find more documentation for tuples in the standard library documentation.

Functions

Functions also have a type! They look like this:

fn foo(x: i32) -> i32 { x }
  
  let x: fn(i32) -> i32 = foo;

In this case, x is a ‘function pointer’ to a function that takes an i32 and returns an i32.

Comments

Now that we have some functions, it’s a good idea to learn about comments. Comments are notes that you leave to other programmers to help explain things about your code. The compiler mostly ignores them.

Rust has two kinds of comments that you should care about: line comments and doc comments.

// Line comments are anything after ‘//’ and extend to the end of the line.
  
  let x = 5; // this is also a line comment.
  
  // If you have a long explanation for something, you can put line comments next
  // to each other. Put a space between the // and your comment so that it’s
  // more readable.

The other kind of comment is a doc comment. Doc comments use /// instead of //, and support Markdown notation inside:

/// Adds one to the number given.
  ///
  /// # Examples
  ///
  /// ```
  /// let five = 5;
  ///
  /// assert_eq!(6, add_one(5));
  /// # fn add_one(x: i32) -> i32 {
  /// #     x + 1
  /// # }
  /// ```
  fn add_one(x: i32) -> i32 {
      x + 1
  }

There is another style of doc comment, //!, to comment containing items (e.g. crates, modules or functions), instead of the items following it. Commonly used inside crates root (lib.rs) or modules root (mod.rs):

//! # The Rust Standard Library
  //!
  //! The Rust Standard Library provides the essential runtime
  //! functionality for building portable Rust software.

When writing doc comments, providing some examples of usage is very, very helpful. You’ll notice we’ve used a new macro here: assert_eq!. This compares two values, and panic!s if they’re not equal to each other. It’s very helpful in documentation. There’s another macro, assert!, which panic!s if the value passed to it is false.

You can use the rustdoc tool to generate HTML documentation from these doc comments, and also to run the code examples as tests!

if

Rust’s take on if is not particularly complex, but it’s much more like the if you’ll find in a dynamically typed language than in a more traditional systems language. So let’s talk about it, to make sure you grasp the nuances.

if is a specific form of a more general concept, the ‘branch’. The name comes from a branch in a tree: a decision point, where depending on a choice, multiple paths can be taken.

In the case of if, there is one choice that leads down two paths:

let x = 5;
  
  if x == 5 {
      println!("x is five!");
  }

If we changed the value of x to something else, this line would not print. More specifically, if the expression after the if evaluates to true, then the block is executed. If it’s false, then it is not.

If you want something to happen in the false case, use an else:

let x = 5;
  
  if x == 5 {
      println!("x is five!");
  } else {
      println!("x is not five :(");
  }

If there is more than one case, use an else if:

let x = 5;
  
  if x == 5 {
      println!("x is five!");
  } else if x == 6 {
      println!("x is six!");
  } else {
      println!("x is not five or six :(");
  }

This is all pretty standard. However, you can also do this:

let x = 5;
  
  let y = if x == 5 {
      10
  } else {
      15
  }; // y: i32

Which we can (and probably should) write like this:

let x = 5;
  
  let y = if x == 5 { 10 } else { 15 }; // y: i32

This works because if is an expression. The value of the expression is the value of the last expression in whichever branch was chosen. An if without an else always results in () as the value.

Loops

Rust currently provides three approaches to performing some kind of iterative activity. They are: loop, while and for. Each approach has its own set of uses.

loop

The infinite loop is the simplest form of loop available in Rust. Using the keyword loop, Rust provides a way to loop indefinitely until some terminating statement is reached. Rust’s infinite loops look like this:

loop {
      println!("Loop forever!");
  }

while

Rust also has a while loop. It looks like this:

let mut x = 5; // mut x: i32
  let mut done = false; // mut done: bool
  
  while !done {
      x += x - 3;
  
      println!("{}", x);
  
      if x % 5 == 0 {
          done = true;
      }
  }

while loops are the correct choice when you’re not sure how many times you need to loop.

If you need an infinite loop, you may be tempted to write this:

while true {

However, loop is far better suited to handle this case:

loop {

Rust’s control-flow analysis treats this construct differently than a while true, since we know that it will always loop. In general, the more information we can give to the compiler, the better it can do with safety and code generation, so you should always prefer loop when you plan to loop infinitely.

for

The for loop is used to loop a particular number of times. Rust’s for loops work a bit differently than in other systems languages, however. Rust’s for loop doesn’t look like this “C-style” for loop:

for (x = 0; x < 10; x++) {
      printf( "%d\n", x );
  }

Instead, it looks like this:

for x in 0..10 {
      println!("{}", x); // x: i32
  }

In slightly more abstract terms,

for var in expression {
      code
  }

The expression is an iterator. The iterator gives back a series of elements. Each element is one iteration of the loop. That value is then bound to the name var, which is valid for the loop body. Once the body is over, the next value is fetched from the iterator, and we loop another time. When there are no more values, the for loop is over.

In our example, 0..10 is an expression that takes a start and an end position, and gives an iterator over those values. The upper bound is exclusive, though, so our loop will print 0 through 9, not 10.

Rust does not have the “C-style” for loop on purpose. Manually controlling each element of the loop is complicated and error prone, even for experienced C developers.

Enumerate

When you need to keep track of how many times you already looped, you can use the .enumerate() function.

On ranges:
for (i,j) in (5..10).enumerate() {
      println!("i = {} and j = {}", i, j);
  }

Outputs:

i = 0 and j = 5
  i = 1 and j = 6
  i = 2 and j = 7
  i = 3 and j = 8
  i = 4 and j = 9

Don’t forget to add the parentheses around the range.

On iterators:
for (linenumber, line) in lines.enumerate() {
      println!("{}: {}", linenumber, line);
  }

Outputs:

0: Content of line one
  1: Content of line two
  2: Content of line three
  3: Content of line four

Ending iteration early

Let’s take a look at that while loop we had earlier:

let mut x = 5;
  let mut done = false;
  
  while !done {
      x += x - 3;
  
      println!("{}", x);
  
      if x % 5 == 0 {
          done = true;
      }
  }

We had to keep a dedicated mut boolean variable binding, done, to know when we should exit out of the loop. Rust has two keywords to help us with modifying iteration: break and continue.

In this case, we can write the loop in a better way with break:

let mut x = 5;
  
  loop {
      x += x - 3;
  
      println!("{}", x);
  
      if x % 5 == 0 { break; }
  }

We now loop forever with loop and use break to break out early. Issuing an explicit return statement will also serve to terminate the loop early.

continue is similar, but instead of ending the loop, goes to the next iteration. This will only print the odd numbers:

for x in 0..10 {
      if x % 2 == 0 { continue; }
  
      println!("{}", x);
  }

Loop labels

You may also encounter situations where you have nested loops and need to specify which one your break or continue statement is for. Like most other languages, by default a break or continue will apply to innermost loop. In a situation where you would like to a break or continue for one of the outer loops, you can use labels to specify which loop the break or continue statement applies to. This will only print when both x and y are odd:

'outer: for x in 0..10 {
      'inner: for y in 0..10 {
          if x % 2 == 0 { continue 'outer; } // continues the loop over x
          if y % 2 == 0 { continue 'inner; } // continues the loop over y
          println!("x: {}, y: {}", x, y);
      }
  }

Ownership

This guide is one of three presenting Rust’s ownership system. This is one of Rust’s most unique and compelling features, with which Rust developers should become quite acquainted. Ownership is how Rust achieves its largest goal, memory safety. There are a few distinct concepts, each with its own chapter:

  • ownership, which you’re reading now
  • borrowing, and their associated feature ‘references’
  • lifetimes, an advanced concept of borrowing

These three chapters are related, and in order. You’ll need all three to fully understand the ownership system.

Meta

Before we get to the details, two important notes about the ownership system.

Rust has a focus on safety and speed. It accomplishes these goals through many ‘zero-cost abstractions’, which means that in Rust, abstractions cost as little as possible in order to make them work. The ownership system is a prime example of a zero-cost abstraction. All of the analysis we’ll talk about in this guide is done at compile time. You do not pay any run-time cost for any of these features.

However, this system does have a certain cost: learning curve. Many new users to Rust experience something we like to call ‘fighting with the borrow checker’, where the Rust compiler refuses to compile a program that the author thinks is valid. This often happens because the programmer’s mental model of how ownership should work doesn’t match the actual rules that Rust implements. You probably will experience similar things at first. There is good news, however: more experienced Rust developers report that once they work with the rules of the ownership system for a period of time, they fight the borrow checker less and less.

With that in mind, let’s learn about ownership.

Ownership

Variable bindings have a property in Rust: they ‘have ownership’ of what they’re bound to. This means that when a binding goes out of scope, Rust will free the bound resources. For example:

fn foo() {
      let v = vec![1, 2, 3];
  }

When v comes into scope, a new Vec<T> is created. In this case, the vector also allocates space on the heap, for the three elements. When v goes out of scope at the end of foo(), Rust will clean up everything related to the vector, even the heap-allocated memory. This happens deterministically, at the end of the scope.

Move semantics

There’s some more subtlety here, though: Rust ensures that there is exactly one binding to any given resource. For example, if we have a vector, we can assign it to another binding:

let v = vec![1, 2, 3];
  
  let v2 = v;

But, if we try to use v afterwards, we get an error:

let v = vec![1, 2, 3];
  
  let v2 = v;
  
  println!("v[0] is: {}", v[0]);

It looks like this:

error: use of moved value: `v`
  println!("v[0] is: {}", v[0]);
                          ^

A similar thing happens if we define a function which takes ownership, and try to use something after we’ve passed it as an argument:

fn take(v: Vec<i32>) {
      // what happens here isn’t important.
  }
  
  let v = vec![1, 2, 3];
  
  take(v);
  
  println!("v[0] is: {}", v[0]);

Same error: ‘use of moved value’. When we transfer ownership to something else, we say that we’ve ‘moved’ the thing we refer to. You don’t need some sort of special annotation here, it’s the default thing that Rust does.

The details

The reason that we cannot use a binding after we’ve moved it is subtle, but important. When we write code like this:

let v = vec![1, 2, 3];
  
  let v2 = v;

The first line allocates memory for the vector object, v, and for the data it contains. The vector object is stored on the stack and contains a pointer to the content ([1, 2, 3]) stored on the heap. When we move v to v2, it creates a copy of that pointer, for v2. Which means that there would be two pointers to the content of the vector on the heap. It would violate Rust’s safety guarantees by introducing a data race. Therefore, Rust forbids using v after we’ve done the move.

It’s also important to note that optimizations may remove the actual copy of the bytes on the stack, depending on circumstances. So it may not be as inefficient as it initially seems.

Copy types

We’ve established that when ownership is transferred to another binding, you cannot use the original binding. However, there’s a trait that changes this behavior, and it’s called Copy. We haven’t discussed traits yet, but for now, you can think of them as an annotation to a particular type that adds extra behavior. For example:

let v = 1;
  
  let v2 = v;
  
  println!("v is: {}", v);

In this case, v is an i32, which implements the Copy trait. This means that, just like a move, when we assign v to v2, a copy of the data is made. But, unlike a move, we can still use v afterward. This is because an i32 has no pointers to data somewhere else, copying it is a full copy.

All primitive types implement the Copy trait and their ownership is therefore not moved like one would assume, following the ´ownership rules´. To give an example, the two following snippets of code only compile because the i32 and bool types implement the Copy trait.

fn main() {
      let a = 5;
  
      let _y = double(a);
      println!("{}", a);
  }
  
  fn double(x: i32) -> i32 {
      x * 2
  }
fn main() {
      let a = true;
  
      let _y = change_truth(a);
      println!("{}", a);
  }
  
  fn change_truth(x: bool) -> bool {
      !x
  }

If we would have used types that do not implement the Copy trait, we would have gotten a compile error because we tried to use a moved value.

error: use of moved value: `a`
  println!("{}", a);
                 ^

We will discuss how to make your own types Copy in the traits section.

More than ownership

Of course, if we had to hand ownership back with every function we wrote:

fn foo(v: Vec<i32>) -> Vec<i32> {
      // do stuff with v
  
      // hand back ownership
      v
  }

This would get very tedious. It gets worse the more things we want to take ownership of:

fn foo(v1: Vec<i32>, v2: Vec<i32>) -> (Vec<i32>, Vec<i32>, i32) {
      // do stuff with v1 and v2
  
      // hand back ownership, and the result of our function
      (v1, v2, 42)
  }
  
  let v1 = vec![1, 2, 3];
  let v2 = vec![1, 2, 3];
  
  let (v1, v2, answer) = foo(v1, v2);

Ugh! The return type, return line, and calling the function gets way more complicated.

Luckily, Rust offers a feature, borrowing, which helps us solve this problem. It’s the topic of the next section!

References and Borrowing

This guide is one of three presenting Rust’s ownership system. This is one of Rust’s most unique and compelling features, with which Rust developers should become quite acquainted. Ownership is how Rust achieves its largest goal, memory safety. There are a few distinct concepts, each with its own chapter:

  • ownership, the key concept
  • borrowing, which you’re reading now
  • lifetimes, an advanced concept of borrowing

These three chapters are related, and in order. You’ll need all three to fully understand the ownership system.

Meta

Before we get to the details, two important notes about the ownership system.

Rust has a focus on safety and speed. It accomplishes these goals through many ‘zero-cost abstractions’, which means that in Rust, abstractions cost as little as possible in order to make them work. The ownership system is a prime example of a zero cost abstraction. All of the analysis we’ll talk about in this guide is done at compile time. You do not pay any run-time cost for any of these features.

However, this system does have a certain cost: learning curve. Many new users to Rust experience something we like to call ‘fighting with the borrow checker’, where the Rust compiler refuses to compile a program that the author thinks is valid. This often happens because the programmer’s mental model of how ownership should work doesn’t match the actual rules that Rust implements. You probably will experience similar things at first. There is good news, however: more experienced Rust developers report that once they work with the rules of the ownership system for a period of time, they fight the borrow checker less and less.

With that in mind, let’s learn about borrowing.

Borrowing

At the end of the ownership section, we had a nasty function that looked like this:

fn foo(v1: Vec<i32>, v2: Vec<i32>) -> (Vec<i32>, Vec<i32>, i32) {
      // do stuff with v1 and v2
  
      // hand back ownership, and the result of our function
      (v1, v2, 42)
  }
  
  let v1 = vec![1, 2, 3];
  let v2 = vec![1, 2, 3];
  
  let (v1, v2, answer) = foo(v1, v2);

This is not idiomatic Rust, however, as it doesn’t take advantage of borrowing. Here’s the first step:

fn foo(v1: &Vec<i32>, v2: &Vec<i32>) -> i32 {
      // do stuff with v1 and v2
  
      // return the answer
      42
  }
  
  let v1 = vec![1, 2, 3];
  let v2 = vec![1, 2, 3];
  
  let answer = foo(&v1, &v2);
  
  // we can use v1 and v2 here!

Instead of taking Vec<i32>s as our arguments, we take a reference: &Vec<i32>. And instead of passing v1 and v2 directly, we pass &v1 and &v2. We call the &T type a ‘reference’, and rather than owning the resource, it borrows ownership. A binding that borrows something does not deallocate the resource when it goes out of scope. This means that after the call to foo(), we can use our original bindings again.

References are immutable, just like bindings. This means that inside of foo(), the vectors can’t be changed at all:

fn foo(v: &Vec<i32>) {
       v.push(5);
  }
  
  let v = vec![];
  
  foo(&v);

errors with:

error: cannot borrow immutable borrowed content `*v` as mutable
  v.push(5);
  ^

Pushing a value mutates the vector, and so we aren’t allowed to do it.

&mut references

There’s a second kind of reference: &mut T. A ‘mutable reference’ allows you to mutate the resource you’re borrowing. For example:

let mut x = 5;
  {
      let y = &mut x;
      *y += 1;
  }
  println!("{}", x);

This will print 6. We make y a mutable reference to x, then add one to the thing y points at. You’ll notice that x had to be marked mut as well, if it wasn’t, we couldn’t take a mutable borrow to an immutable value.

You’ll also notice we added an asterisk (*) in front of y, making it *y, this is because y is an &mut reference. You’ll also need to use them for accessing the contents of a reference as well.

Otherwise, &mut references are just like references. There is a large difference between the two, and how they interact, though. You can tell something is fishy in the above example, because we need that extra scope, with the { and }. If we remove them, we get an error:

error: cannot borrow `x` as immutable because it is also borrowed as mutable
      println!("{}", x);
                     ^
  note: previous borrow of `x` occurs here; the mutable borrow prevents
  subsequent moves, borrows, or modification of `x` until the borrow ends
          let y = &mut x;
                       ^
  note: previous borrow ends here
  fn main() {
  
  }
  ^

As it turns out, there are rules.

The Rules

Here’s the rules about borrowing in Rust:

First, any borrow must last for a scope no greater than that of the owner. Second, you may have one or the other of these two kinds of borrows, but not both at the same time:

  • one or more references (&T) to a resource,
  • exactly one mutable reference (&mut T).

You may notice that this is very similar, though not exactly the same as, to the definition of a data race:

There is a ‘data race’ when two or more pointers access the same memory location at the same time, where at least one of them is writing, and the operations are not synchronized.

With references, you may have as many as you’d like, since none of them are writing. If you are writing, you need two or more pointers to the same memory, and you can only have one &mut at a time. This is how Rust prevents data races at compile time: we’ll get errors if we break the rules.

With this in mind, let’s consider our example again.

Thinking in scopes

Here’s the code:

let mut x = 5;
  let y = &mut x;
  
  *y += 1;
  
  println!("{}", x);

This code gives us this error:

error: cannot borrow `x` as immutable because it is also borrowed as mutable
      println!("{}", x);
                     ^

This is because we’ve violated the rules: we have a &mut T pointing to x, and so we aren’t allowed to create any &Ts. One or the other. The note hints at how to think about this problem:

note: previous borrow ends here
  fn main() {
  
  }
  ^

In other words, the mutable borrow is held through the rest of our example. What we want is for the mutable borrow to end before we try to call println! and make an immutable borrow. In Rust, borrowing is tied to the scope that the borrow is valid for. And our scopes look like this:

let mut x = 5;
  
  let y = &mut x;    // -+ &mut borrow of x starts here
                     //  |
  *y += 1;           //  |
                     //  |
  println!("{}", x); // -+ - try to borrow x here
                     // -+ &mut borrow of x ends here

The scopes conflict: we can’t make an &x while y is in scope.

So when we add the curly braces:

let mut x = 5;
  
  {                   
      let y = &mut x; // -+ &mut borrow starts here
      *y += 1;        //  |
  }                   // -+ ... and ends here
  
  println!("{}", x);  // <- try to borrow x here

There’s no problem. Our mutable borrow goes out of scope before we create an immutable one. But scope is the key to seeing how long a borrow lasts for.

Issues borrowing prevents

Why have these restrictive rules? Well, as we noted, these rules prevent data races. What kinds of issues do data races cause? Here’s a few.

Iterator invalidation

One example is ‘iterator invalidation’, which happens when you try to mutate a collection that you’re iterating over. Rust’s borrow checker prevents this from happening:

let mut v = vec![1, 2, 3];
  
  for i in &v {
      println!("{}", i);
  }

This prints out one through three. As we iterate through the vectors, we’re only given references to the elements. And v is itself borrowed as immutable, which means we can’t change it while we’re iterating:

let mut v = vec![1, 2, 3];
  
  for i in &v {
      println!("{}", i);
      v.push(34);
  }

Here’s the error:

error: cannot borrow `v` as mutable because it is also borrowed as immutable
      v.push(34);
      ^
  note: previous borrow of `v` occurs here; the immutable borrow prevents
  subsequent moves or mutable borrows of `v` until the borrow ends
  for i in &v {
            ^
  note: previous borrow ends here
  for i in &v {
      println!(“{}”, i);
      v.push(34);
  }
  ^

We can’t modify v because it’s borrowed by the loop.

use after free

References must not live longer than the resource they refer to. Rust will check the scopes of your references to ensure that this is true.

If Rust didn’t check this property, we could accidentally use a reference which was invalid. For example:

let y: &i32;
  { 
      let x = 5;
      y = &x;
  }
  
  println!("{}", y);

We get this error:

error: `x` does not live long enough
      y = &x;
           ^
  note: reference must be valid for the block suffix following statement 0 at
  2:16...
  let y: &i32;
  { 
      let x = 5;
      y = &x;
  }
  
  note: ...but borrowed value is only valid for the block suffix following
  statement 0 at 4:18
      let x = 5;
      y = &x;
  }

In other words, y is only valid for the scope where x exists. As soon as x goes away, it becomes invalid to refer to it. As such, the error says that the borrow ‘doesn’t live long enough’ because it’s not valid for the right amount of time.

The same problem occurs when the reference is declared before the variable it refers to. This is because resources within the same scope are freed in the opposite order they were declared:

let y: &i32;
  let x = 5;
  y = &x;
  
  println!("{}", y);

We get this error:

error: `x` does not live long enough
  y = &x;
       ^
  note: reference must be valid for the block suffix following statement 0 at
  2:16...
      let y: &i32;
      let x = 5;
      y = &x;
      
      println!("{}", y);
  }
  
  note: ...but borrowed value is only valid for the block suffix following
  statement 1 at 3:14
      let x = 5;
      y = &x;
      
      println!("{}", y);
  }

In the above example, y is declared before x, meaning that y lives longer than x, which is not allowed.

Lifetimes

This guide is one of three presenting Rust’s ownership system. This is one of Rust’s most unique and compelling features, with which Rust developers should become quite acquainted. Ownership is how Rust achieves its largest goal, memory safety. There are a few distinct concepts, each with its own chapter:

  • ownership, the key concept
  • borrowing, and their associated feature ‘references’
  • lifetimes, which you’re reading now

These three chapters are related, and in order. You’ll need all three to fully understand the ownership system.

Meta

Before we get to the details, two important notes about the ownership system.

Rust has a focus on safety and speed. It accomplishes these goals through many ‘zero-cost abstractions’, which means that in Rust, abstractions cost as little as possible in order to make them work. The ownership system is a prime example of a zero-cost abstraction. All of the analysis we’ll talk about in this guide is done at compile time. You do not pay any run-time cost for any of these features.

However, this system does have a certain cost: learning curve. Many new users to Rust experience something we like to call ‘fighting with the borrow checker’, where the Rust compiler refuses to compile a program that the author thinks is valid. This often happens because the programmer’s mental model of how ownership should work doesn’t match the actual rules that Rust implements. You probably will experience similar things at first. There is good news, however: more experienced Rust developers report that once they work with the rules of the ownership system for a period of time, they fight the borrow checker less and less.

With that in mind, let’s learn about lifetimes.

Lifetimes

Lending out a reference to a resource that someone else owns can be complicated. For example, imagine this set of operations:

  • I acquire a handle to some kind of resource.
  • I lend you a reference to the resource.
  • I decide I’m done with the resource, and deallocate it, while you still have your reference.
  • You decide to use the resource.

Uh oh! Your reference is pointing to an invalid resource. This is called a dangling pointer or ‘use after free’, when the resource is memory.

To fix this, we have to make sure that step four never happens after step three. The ownership system in Rust does this through a concept called lifetimes, which describe the scope that a reference is valid for.

When we have a function that takes a reference by argument, we can be implicit or explicit about the lifetime of the reference:

// implicit
  fn foo(x: &i32) {
  }
  
  // explicit
  fn bar<'a>(x: &'a i32) {
  }

The 'a reads ‘the lifetime a’. Technically, every reference has some lifetime associated with it, but the compiler lets you elide them in common cases. Before we get to that, though, let’s break the explicit example down:

fn bar<'a>(...)

We previously talked a little about function syntax, but we didn’t discuss the <>s after a function’s name. A function can have ‘generic parameters’ between the <>s, of which lifetimes are one kind. We’ll discuss other kinds of generics later in the book, but for now, let’s just focus on the lifetimes aspect.

We use <> to declare our lifetimes. This says that bar has one lifetime, 'a. If we had two reference parameters, it would look like this:

fn bar<'a, 'b>(...)

Then in our parameter list, we use the lifetimes we’ve named:

...(x: &'a i32)

If we wanted an &mut reference, we’d do this:

...(x: &'a mut i32)

If you compare &mut i32 to &'a mut i32, they’re the same, it’s just that the lifetime 'a has snuck in between the & and the mut i32. We read &mut i32 as ‘a mutable reference to an i32’ and &'a mut i32 as ‘a mutable reference to an i32 with the lifetime 'a’.

In structs

You’ll also need explicit lifetimes when working with structs:

struct Foo<'a> {
      x: &'a i32,
  }
  
  fn main() {
      let y = &5; // this is the same as `let _y = 5; let y = &_y;`
      let f = Foo { x: y };
  
      println!("{}", f.x);
  }

As you can see, structs can also have lifetimes. In a similar way to functions,

struct Foo<'a> {

declares a lifetime, and

x: &'a i32,

uses it. So why do we need a lifetime here? We need to ensure that any reference to a Foo cannot outlive the reference to an i32 it contains.

impl blocks

Let’s implement a method on Foo:

struct Foo<'a> {
      x: &'a i32,
  }
  
  impl<'a> Foo<'a> {
      fn x(&self) -> &'a i32 { self.x }
  }
  
  fn main() {
      let y = &5; // this is the same as `let _y = 5; let y = &_y;`
      let f = Foo { x: y };
  
      println!("x is: {}", f.x());
  }

As you can see, we need to declare a lifetime for Foo in the impl line. We repeat 'a twice, just like on functions: impl<'a> defines a lifetime 'a, and Foo<'a> uses it.

Multiple lifetimes

If you have multiple references, you can use the same lifetime multiple times:

fn x_or_y<'a>(x: &'a str, y: &'a str) -> &'a str {

This says that x and y both are alive for the same scope, and that the return value is also alive for that scope. If you wanted x and y to have different lifetimes, you can use multiple lifetime parameters:

fn x_or_y<'a, 'b>(x: &'a str, y: &'b str) -> &'a str {

In this example, x and y have different valid scopes, but the return value has the same lifetime as x.

Thinking in scopes

A way to think about lifetimes is to visualize the scope that a reference is valid for. For example:

fn main() {
      let y = &5;     // -+ y goes into scope
                      //  |
      // stuff        //  |
                      //  |
  }                   // -+ y goes out of scope

Adding in our Foo:

struct Foo<'a> {
      x: &'a i32,
  }
  
  fn main() {
      let y = &5;           // -+ y goes into scope
      let f = Foo { x: y }; // -+ f goes into scope
      // stuff              //  |
                            //  |
  }                         // -+ f and y go out of scope

Our f lives within the scope of y, so everything works. What if it didn’t? This code won’t work:

struct Foo<'a> {
      x: &'a i32,
  }
  
  fn main() {
      let x;                    // -+ x goes into scope
                                //  |
      {                         //  |
          let y = &5;           // ---+ y goes into scope
          let f = Foo { x: y }; // ---+ f goes into scope
          x = &f.x;             //  | | error here
      }                         // ---+ f and y go out of scope
                                //  |
      println!("{}", x);        //  |
  }                             // -+ x goes out of scope

Whew! As you can see here, the scopes of f and y are smaller than the scope of x. But when we do x = &f.x, we make x a reference to something that’s about to go out of scope.

Named lifetimes are a way of giving these scopes a name. Giving something a name is the first step towards being able to talk about it.

’static

The lifetime named ‘static’ is a special lifetime. It signals that something has the lifetime of the entire program. Most Rust programmers first come across 'static when dealing with strings:

let x: &'static str = "Hello, world.";

String literals have the type &'static str because the reference is always alive: they are baked into the data segment of the final binary. Another example are globals:

static FOO: i32 = 5;
  let x: &'static i32 = &FOO;

This adds an i32 to the data segment of the binary, and x is a reference to it.

Lifetime Elision

Rust supports powerful local type inference in function bodies, but it’s forbidden in item signatures to allow reasoning about the types based on the item signature alone. However, for ergonomic reasons a very restricted secondary inference algorithm called “lifetime elision” applies in function signatures. It infers only based on the signature components themselves and not based on the body of the function, only infers lifetime parameters, and does this with only three easily memorizable and unambiguous rules. This makes lifetime elision a shorthand for writing an item signature, while not hiding away the actual types involved as full local inference would if applied to it.

When talking about lifetime elision, we use the term input lifetime and output lifetime. An input lifetime is a lifetime associated with a parameter of a function, and an output lifetime is a lifetime associated with the return value of a function. For example, this function has an input lifetime:

fn foo<'a>(bar: &'a str)

This one has an output lifetime:

fn foo<'a>() -> &'a str

This one has a lifetime in both positions:

fn foo<'a>(bar: &'a str) -> &'a str

Here are the three rules:

  • Each elided lifetime in a function’s arguments becomes a distinct lifetime parameter.

  • If there is exactly one input lifetime, elided or not, that lifetime is assigned to all elided lifetimes in the return values of that function.

  • If there are multiple input lifetimes, but one of them is &self or &mut self, the lifetime of self is assigned to all elided output lifetimes.

Otherwise, it is an error to elide an output lifetime.

Examples

Here are some examples of functions with elided lifetimes. We’ve paired each example of an elided lifetime with its expanded form.

fn print(s: &str); // elided
  fn print<'a>(s: &'a str); // expanded
  
  fn debug(lvl: u32, s: &str); // elided
  fn debug<'a>(lvl: u32, s: &'a str); // expanded
  
  // In the preceding example, `lvl` doesn’t need a lifetime because it’s not a
  // reference (`&`). Only things relating to references (such as a `struct`
  // which contains a reference) need lifetimes.
  
  fn substr(s: &str, until: u32) -> &str; // elided
  fn substr<'a>(s: &'a str, until: u32) -> &'a str; // expanded
  
  fn get_str() -> &str; // ILLEGAL, no inputs
  
  fn frob(s: &str, t: &str) -> &str; // ILLEGAL, two inputs
  fn frob<'a, 'b>(s: &'a str, t: &'b str) -> &str; // Expanded: Output lifetime is ambig
  ↳ uous
  
  fn get_mut(&mut self) -> &mut T; // elided
  fn get_mut<'a>(&'a mut self) -> &'a mut T; // expanded
  
  fn args<T:ToCStr>(&mut self, args: &[T]) -> &mut Command // elided
  fn args<'a, 'b, T:ToCStr>(&'a mut self, args: &'b [T]) -> &'a mut Command // expanded
  
  fn new(buf: &mut [u8]) -> BufWriter; // elided
  fn new<'a>(buf: &'a mut [u8]) -> BufWriter<'a> // expanded

Mutability

Mutability, the ability to change something, works a bit differently in Rust than in other languages. The first aspect of mutability is its non-default status:

let x = 5;
  x = 6; // error!

We can introduce mutability with the mut keyword:

let mut x = 5;
  
  x = 6; // no problem!

This is a mutable variable binding. When a binding is mutable, it means you’re allowed to change what the binding points to. So in the above example, it’s not so much that the value at x is changing, but that the binding changed from one i32 to another.

If you want to change what the binding points to, you’ll need a mutable reference:

let mut x = 5;
  let y = &mut x;

y is an immutable binding to a mutable reference, which means that you can’t bind y to something else (y = &mut z), but you can mutate the thing that’s bound to y (*y = 5). A subtle distinction.

Of course, if you need both:

let mut x = 5;
  let mut y = &mut x;

Now y can be bound to another value, and the value it’s referencing can be changed.

It’s important to note that mut is part of a pattern, so you can do things like this:

let (mut x, y) = (5, 6);
  
  fn foo(mut x: i32) {

Interior vs. Exterior Mutability

However, when we say something is ‘immutable’ in Rust, that doesn’t mean that it’s not able to be changed: we mean something has ‘exterior mutability’. Consider, for example, Arc<T>:

use std::sync::Arc;
  
  let x = Arc::new(5);
  let y = x.clone();

When we call clone(), the Arc<T> needs to update the reference count. Yet we’ve not used any muts here, x is an immutable binding, and we didn’t take &mut 5 or anything. So what gives?

To understand this, we have to go back to the core of Rust’s guiding philosophy, memory safety, and the mechanism by which Rust guarantees it, the ownership system, and more specifically, borrowing:

You may have one or the other of these two kinds of borrows, but not both at the same time:

  • one or more references (&T) to a resource,
  • exactly one mutable reference (&mut T).

So, that’s the real definition of ‘immutability’: is this safe to have two pointers to? In Arc<T>’s case, yes: the mutation is entirely contained inside the structure itself. It’s not user facing. For this reason, it hands out &T with clone(). If it handed out &mut Ts, though, that would be a problem.

Other types, like the ones in the std::cell module, have the opposite: interior mutability. For example:

use std::cell::RefCell;
  
  let x = RefCell::new(42);
  
  let y = x.borrow_mut();

RefCell hands out &mut references to what’s inside of it with the borrow_mut() method. Isn’t that dangerous? What if we do:

use std::cell::RefCell;
  
  let x = RefCell::new(42);
  
  let y = x.borrow_mut();
  let z = x.borrow_mut();

This will in fact panic, at runtime. This is what RefCell does: it enforces Rust’s borrowing rules at runtime, and panic!s if they’re violated. This allows us to get around another aspect of Rust’s mutability rules. Let’s talk about it first.

Field-level mutability

Mutability is a property of either a borrow (&mut) or a binding (let mut). This means that, for example, you cannot have a struct with some fields mutable and some immutable:

struct Point {
      x: i32,
      mut y: i32, // nope
  }

The mutability of a struct is in its binding:

struct Point {
      x: i32,
      y: i32,
  }
  
  let mut a = Point { x: 5, y: 6 };
  
  a.x = 10;
  
  let b = Point { x: 5, y: 6};
  
  b.x = 10; // error: cannot assign to immutable field `b.x`

However, by using Cell<T>, you can emulate field-level mutability:

use std::cell::Cell;
  
  struct Point {
      x: i32,
      y: Cell<i32>,
  }
  
  let point = Point { x: 5, y: Cell::new(6) };
  
  point.y.set(7);
  
  println!("y: {:?}", point.y);

This will print y: Cell { value: 7 }. We’ve successfully updated y.

Structs

structs are a way of creating more complex data types. For example, if we were doing calculations involving coordinates in 2D space, we would need both an x and a y value:

let origin_x = 0;
  let origin_y = 0;

A struct lets us combine these two into a single, unified datatype:

struct Point {
      x: i32,
      y: i32,
  }
  
  fn main() {
      let origin = Point { x: 0, y: 0 }; // origin: Point
  
      println!("The origin is at ({}, {})", origin.x, origin.y);
  }

There’s a lot going on here, so let’s break it down. We declare a struct with the struct keyword, and then with a name. By convention, structs begin with a capital letter and are camel cased: PointInSpace, not Point_In_Space.

We can create an instance of our struct via let, as usual, but we use a key: value style syntax to set each field. The order doesn’t need to be the same as in the original declaration.

Finally, because fields have names, we can access the field through dot notation: origin.x.

The values in structs are immutable by default, like other bindings in Rust. Use mut to make them mutable:

struct Point {
      x: i32,
      y: i32,
  }
  
  fn main() {
      let mut point = Point { x: 0, y: 0 };
  
      point.x = 5;
  
      println!("The point is at ({}, {})", point.x, point.y);
  }

This will print The point is at (5, 0).

Rust does not support field mutability at the language level, so you cannot write something like this:

struct Point {
      mut x: i32,
      y: i32,
  }

Mutability is a property of the binding, not of the structure itself. If you’re used to field-level mutability, this may seem strange at first, but it significantly simplifies things. It even lets you make things mutable for a short time only:

struct Point {
      x: i32,
      y: i32,
  }
  
  fn main() {
      let mut point = Point { x: 0, y: 0 };
  
      point.x = 5;
  
      let point = point; // this new binding can’t change now
  
      point.y = 6; // this causes an error
  }

Update syntax

A struct can include .. to indicate that you want to use a copy of some other struct for some of the values. For example:

struct Point3d {
      x: i32,
      y: i32,
      z: i32,
  }
  
  let mut point = Point3d { x: 0, y: 0, z: 0 };
  point = Point3d { y: 1, .. point };

This gives point a new y, but keeps the old x and z values. It doesn’t have to be the same struct either, you can use this syntax when making new ones, and it will copy the values you don’t specify:

let origin = Point3d { x: 0, y: 0, z: 0 };
  let point = Point3d { z: 1, x: 2, .. origin };

Tuple structs

Rust has another data type that’s like a hybrid between a tuple and a struct, called a ‘tuple struct’. Tuple structs have a name, but their fields don’t:

struct Color(i32, i32, i32);
  struct Point(i32, i32, i32);

These two will not be equal, even if they have the same values:

let black = Color(0, 0, 0);
  let origin = Point(0, 0, 0);

It is almost always better to use a struct than a tuple struct. We would write Color and Point like this instead:

struct Color {
      red: i32,
      blue: i32,
      green: i32,
  }
  
  struct Point {
      x: i32,
      y: i32,
      z: i32,
  }

Now, we have actual names, rather than positions. Good names are important, and with a struct, we have actual names.

There is one case when a tuple struct is very useful, though, and that’s a tuple struct with only one element. We call this the ‘newtype’ pattern, because it allows you to create a new type, distinct from that of its contained value and expressing its own semantic meaning:

struct Inches(i32);
  
  let length = Inches(10);
  
  let Inches(integer_length) = length;
  println!("length is {} inches", integer_length);

As you can see here, you can extract the inner integer type through a destructuring let, just as with regular tuples. In this case, the let Inches(integer_length) assigns 10 to integer_length.

Unit-like structs

You can define a struct with no members at all:

struct Electron;

Such a struct is called ‘unit-like’ because it resembles the empty tuple, (), sometimes called ‘unit’. Like a tuple struct, it defines a new type.

This is rarely useful on its own (although sometimes it can serve as a marker type), but in combination with other features, it can become useful. For instance, a library may ask you to create a structure that implements a certain trait to handle events. If you don’t have any data you need to store in the structure, you can just create a unit-like struct.

Enums

An enum in Rust is a type that represents data that could be one of several possible variants:

enum Message {
      Quit,
      ChangeColor(i32, i32, i32),
      Move { x: i32, y: i32 },
      Write(String),
  }

Each variant can optionally have data associated with it. The syntax for defining variants resembles the syntaxes used to define structs: you can have variants with no data (like unit-like structs), variants with named data, and variants with unnamed data (like tuple structs). Unlike separate struct definitions, however, an enum is a single type. A value of the enum can match any of the variants. For this reason, an enum is sometimes called a ‘sum type’: the set of possible values of the enum is the sum of the sets of possible values for each variant.

We use the :: syntax to use the name of each variant: they’re scoped by the name of the enum itself. This allows both of these to work:

let x: Message = Message::Move { x: 3, y: 4 };
  
  enum BoardGameTurn {
      Move { squares: i32 },
      Pass,
  }
  
  let y: BoardGameTurn = BoardGameTurn::Move { squares: 1 };

Both variants are named Move, but since they’re scoped to the name of the enum, they can both be used without conflict.

A value of an enum type contains information about which variant it is, in addition to any data associated with that variant. This is sometimes referred to as a ‘tagged union’, since the data includes a ‘tag’ indicating what type it is. The compiler uses this information to enforce that you’re accessing the data in the enum safely. For instance, you can’t simply try to destructure a value as if it were one of the possible variants:

fn process_color_change(msg: Message) {
      let Message::ChangeColor(r, g, b) = msg; // compile-time error
  }

Not supporting these operations may seem rather limiting, but it’s a limitation which we can overcome. There are two ways: by implementing equality ourselves, or by pattern matching variants with match expressions, which you’ll learn in the next section. We don’t know enough about Rust to implement equality yet, but we’ll find out in the traits section.

Constructors as functions

An enum’s constructors can also be used like functions. For example:

let m = Message::Write("Hello, world".to_string());

Is the same as

fn foo(x: String) -> Message {
      Message::Write(x)
  }
  
  let x = foo("Hello, world".to_string());

This is not immediately useful to us, but when we get to closures, we’ll talk about passing functions as arguments to other functions. For example, with iterators, we can do this to convert a vector of Strings into a vector of Message::Writes:


  let v = vec!["Hello".to_string(), "World".to_string()];
  
  let v1: Vec<Message> = v.into_iter().map(Message::Write).collect();

Match

Often, a simple if/else isn’t enough, because you have more than two possible options. Also, conditions can get quite complex. Rust has a keyword, match, that allows you to replace complicated if/else groupings with something more powerful. Check it out:

let x = 5;
  
  match x {
      1 => println!("one"),
      2 => println!("two"),
      3 => println!("three"),
      4 => println!("four"),
      5 => println!("five"),
      _ => println!("something else"),
  }

match takes an expression and then branches based on its value. Each ‘arm’ of the branch is of the form val => expression. When the value matches, that arm’s expression will be evaluated. It’s called match because of the term ‘pattern matching’, which match is an implementation of. There’s an entire section on patterns that covers all the patterns that are possible here.

So what’s the big advantage? Well, there are a few. First of all, match enforces ‘exhaustiveness checking’. Do you see that last arm, the one with the underscore (_)? If we remove that arm, Rust will give us an error:

error: non-exhaustive patterns: `_` not covered

In other words, Rust is trying to tell us we forgot a value. Because x is an integer, Rust knows that it can have a number of different values – for example, 6. Without the _, however, there is no arm that could match, and so Rust refuses to compile the code. _ acts like a ‘catch-all arm’. If none of the other arms match, the arm with _ will, and since we have this catch-all arm, we now have an arm for every possible value of x, and so our program will compile successfully.

match is also an expression, which means we can use it on the right-hand side of a let binding or directly where an expression is used:

let x = 5;
  
  let number = match x {
      1 => "one",
      2 => "two",
      3 => "three",
      4 => "four",
      5 => "five",
      _ => "something else",
  };

Sometimes it’s a nice way of converting something from one type to another.

Matching on enums

Another important use of the match keyword is to process the possible variants of an enum:

enum Message {
      Quit,
      ChangeColor(i32, i32, i32),
      Move { x: i32, y: i32 },
      Write(String),
  }
  
  fn quit() { /* ... */ }
  fn change_color(r: i32, g: i32, b: i32) { /* ... */ }
  fn move_cursor(x: i32, y: i32) { /* ... */ }
  
  fn process_message(msg: Message) {
      match msg {
          Message::Quit => quit(),
          Message::ChangeColor(r, g, b) => change_color(r, g, b),
          Message::Move { x: x, y: y } => move_cursor(x, y),
          Message::Write(s) => println!("{}", s),
      };
  }

Again, the Rust compiler checks exhaustiveness, so it demands that you have a match arm for every variant of the enum. If you leave one off, it will give you a compile-time error unless you use _.

Unlike the previous uses of match, you can’t use the normal if statement to do this. You can use the if let statement, which can be seen as an abbreviated form of match.

Patterns

Patterns are quite common in Rust. We use them in variable bindings, match statements, and other places, too. Let’s go on a whirlwind tour of all of the things patterns can do!

A quick refresher: you can match against literals directly, and _ acts as an ‘any’ case:

let x = 1;
  
  match x {
      1 => println!("one"),
      2 => println!("two"),
      3 => println!("three"),
      _ => println!("anything"),
  }

This prints one.

Multiple patterns

You can match multiple patterns with |:

let x = 1;
  
  match x {
      1 | 2 => println!("one or two"),
      3 => println!("three"),
      _ => println!("anything"),
  }

This prints one or two.

Destructuring

If you have a compound data type, like a struct, you can destructure it inside of a pattern:

struct Point {
      x: i32,
      y: i32,
  }
  
  let origin = Point { x: 0, y: 0 };
  
  match origin {
      Point { x, y } => println!("({},{})", x, y),
  }

We can use : to give a value a different name.

struct Point {
      x: i32,
      y: i32,
  }
  
  let origin = Point { x: 0, y: 0 };
  
  match origin {
      Point { x: x1, y: y1 } => println!("({},{})", x1, y1),
  }

If we only care about some of the values, we don’t have to give them all names:

struct Point {
      x: i32,
      y: i32,
  }
  
  let origin = Point { x: 0, y: 0 };
  
  match origin {
      Point { x, .. } => println!("x is {}", x),
  }

This prints x is 0.

You can do this kind of match on any member, not just the first:

struct Point {
      x: i32,
      y: i32,
  }
  
  let origin = Point { x: 0, y: 0 };
  
  match origin {
      Point { y, .. } => println!("y is {}", y),
  }

This prints y is 0.

This ‘destructuring’ behavior works on any compound data type, like tuples or enums.

Ignoring bindings

You can use _ in a pattern to disregard the type and value. For example, here’s a match against a Result<T, E>:

match some_value {
      Ok(value) => println!("got a value: {}", value),
      Err(_) => println!("an error occurred"),
  }

In the first arm, we bind the value inside the Ok variant to value. But in the Err arm, we use _ to disregard the specific error, and just print a general error message.

_ is valid in any pattern that creates a binding. This can be useful to ignore parts of a larger structure:

fn coordinate() -> (i32, i32, i32) {
      // generate and return some sort of triple tuple
  }
  
  let (x, _, z) = coordinate();

Here, we bind the first and last element of the tuple to x and z, but ignore the middle element.

Similarly, you can use .. in a pattern to disregard multiple values.

enum OptionalTuple {
      Value(i32, i32, i32),
      Missing,
  }
  
  let x = OptionalTuple::Value(5, -2, 3);
  
  match x {
      OptionalTuple::Value(..) => println!("Got a tuple!"),
      OptionalTuple::Missing => println!("No such luck."),
  }

This prints Got a tuple!.

ref and ref mut

If you want to get a reference, use the ref keyword:

let x = 5;
  
  match x {
      ref r => println!("Got a reference to {}", r),
  }

This prints Got a reference to 5.

Here, the r inside the match has the type &i32. In other words, the ref keyword creates a reference, for use in the pattern. If you need a mutable reference, ref mut will work in the same way:

let mut x = 5;
  
  match x {
      ref mut mr => println!("Got a mutable reference to {}", mr),
  }

Ranges

You can match a range of values with ...:

let x = 1;
  
  match x {
      1 ... 5 => println!("one through five"),
      _ => println!("anything"),
  }

This prints one through five.

Ranges are mostly used with integers and chars:

let x = '💅';
  
  match x {
      'a' ... 'j' => println!("early letter"),
      'k' ... 'z' => println!("late letter"),
      _ => println!("something else"),
  }

This prints something else.

Bindings

You can bind values to names with @:

let x = 1;
  
  match x {
      e @ 1 ... 5 => println!("got a range element {}", e),
      _ => println!("anything"),
  }

This prints got a range element 1. This is useful when you want to do a complicated match of part of a data structure:

#[derive(Debug)]
  struct Person {
      name: Option<String>,
  }
  
  let name = "Steve".to_string();
  let mut x: Option<Person> = Some(Person { name: Some(name) });
  match x {
      Some(Person { name: ref a @ Some(_), .. }) => println!("{:?}", a),
      _ => {}
  }

This prints Some("Steve"): we’ve bound the inner name to a.

If you use @ with |, you need to make sure the name is bound in each part of the pattern:

let x = 5;
  
  match x {
      e @ 1 ... 5 | e @ 8 ... 10 => println!("got a range element {}", e),
      _ => println!("anything"),
  }

Guards

You can introduce ‘match guards’ with if:

enum OptionalInt {
      Value(i32),
      Missing,
  }
  
  let x = OptionalInt::Value(5);
  
  match x {
      OptionalInt::Value(i) if i > 5 => println!("Got an int bigger than five!"),
      OptionalInt::Value(..) => println!("Got an int!"),
      OptionalInt::Missing => println!("No such luck."),
  }

This prints Got an int!.

If you’re using if with multiple patterns, the if applies to both sides:

let x = 4;
  let y = false;
  
  match x {
      4 | 5 if y => println!("yes"),
      _ => println!("no"),
  }

This prints no, because the if applies to the whole of 4 | 5, and not to just the 5, In other words, the the precedence of if behaves like this:

(4 | 5) if y => ...

not this:

4 | (5 if y) => ...

Mix and Match

Whew! That’s a lot of different ways to match things, and they can all be mixed and matched, depending on what you’re doing:

match x {
      Foo { x: Some(ref name), y: None } => ...
  }

Patterns are very powerful. Make good use of them.

Method Syntax

Functions are great, but if you want to call a bunch of them on some data, it can be awkward. Consider this code:

baz(bar(foo));

We would read this left-to-right, and so we see ‘baz bar foo’. But this isn’t the order that the functions would get called in, that’s inside-out: ‘foo bar baz’. Wouldn’t it be nice if we could do this instead?

foo.bar().baz();

Luckily, as you may have guessed with the leading question, you can! Rust provides the ability to use this ‘method call syntax’ via the impl keyword.

Method calls

Here’s how it works:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  }
  
  fn main() {
      let c = Circle { x: 0.0, y: 0.0, radius: 2.0 };
      println!("{}", c.area());
  }

This will print 12.566371.

We’ve made a struct that represents a circle. We then write an impl block, and inside it, define a method, area.

Methods take a special first parameter, of which there are three variants: self, &self, and &mut self. You can think of this first parameter as being the foo in foo.bar(). The three variants correspond to the three kinds of things foo could be: self if it’s just a value on the stack, &self if it’s a reference, and &mut self if it’s a mutable reference. Because we took the &self parameter to area, we can use it just like any other parameter. Because we know it’s a Circle, we can access the radius just like we would with any other struct.

We should default to using &self, as you should prefer borrowing over taking ownership, as well as taking immutable references over mutable ones. Here’s an example of all three variants:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn reference(&self) {
         println!("taking self by reference!");
      }
  
      fn mutable_reference(&mut self) {
         println!("taking self by mutable reference!");
      }
  
      fn takes_ownership(self) {
         println!("taking ownership of self!");
      }
  }

Chaining method calls

So, now we know how to call a method, such as foo.bar(). But what about our original example, foo.bar().baz()? This is called ‘method chaining’. Let’s look at an example:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  
      fn grow(&self, increment: f64) -> Circle {
          Circle { x: self.x, y: self.y, radius: self.radius + increment }
      }
  }
  
  fn main() {
      let c = Circle { x: 0.0, y: 0.0, radius: 2.0 };
      println!("{}", c.area());
  
      let d = c.grow(2.0).area();
      println!("{}", d);
  }

Check the return type:

fn grow(&self, increment: f64) -> Circle {

We just say we’re returning a Circle. With this method, we can grow a new Circle to any arbitrary size.

Associated functions

You can also define associated functions that do not take a self parameter. Here’s a pattern that’s very common in Rust code:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn new(x: f64, y: f64, radius: f64) -> Circle {
          Circle {
              x: x,
              y: y,
              radius: radius,
          }
      }
  }
  
  fn main() {
      let c = Circle::new(0.0, 0.0, 2.0);
  }

This ‘associated function’ builds a new Circle for us. Note that associated functions are called with the Struct::function() syntax, rather than the ref.method() syntax. Some other languages call associated functions ‘static methods’.

Builder Pattern

Let’s say that we want our users to be able to create Circles, but we will allow them to only set the properties they care about. Otherwise, the x and y attributes will be 0.0, and the radius will be 1.0. Rust doesn’t have method overloading, named arguments, or variable arguments. We employ the builder pattern instead. It looks like this:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  }
  
  struct CircleBuilder {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl CircleBuilder {
      fn new() -> CircleBuilder {
          CircleBuilder { x: 0.0, y: 0.0, radius: 1.0, }
      }
  
      fn x(&mut self, coordinate: f64) -> &mut CircleBuilder {
          self.x = coordinate;
          self
      }
  
      fn y(&mut self, coordinate: f64) -> &mut CircleBuilder {
          self.y = coordinate;
          self
      }
  
      fn radius(&mut self, radius: f64) -> &mut CircleBuilder {
          self.radius = radius;
          self
      }
  
      fn finalize(&self) -> Circle {
          Circle { x: self.x, y: self.y, radius: self.radius }
      }
  }
  
  fn main() {
      let c = CircleBuilder::new()
                  .x(1.0)
                  .y(2.0)
                  .radius(2.0)
                  .finalize();
  
      println!("area: {}", c.area());
      println!("x: {}", c.x);
      println!("y: {}", c.y);
  }

What we’ve done here is make another struct, CircleBuilder. We’ve defined our builder methods on it. We’ve also defined our area() method on Circle. We also made one more method on CircleBuilder: finalize(). This method creates our final Circle from the builder. Now, we’ve used the type system to enforce our concerns: we can use the methods on CircleBuilder to constrain making Circles in any way we choose.

Vectors

A ‘vector’ is a dynamic or ‘growable’ array, implemented as the standard library type Vec<T>. The T means that we can have vectors of any type (see the chapter on generics for more). Vectors always allocate their data on the heap. You can create them with the vec! macro:

let v = vec![1, 2, 3, 4, 5]; // v: Vec<i32>

(Notice that unlike the println! macro we’ve used in the past, we use square brackets [] with vec! macro. Rust allows you to use either in either situation, this is just convention.)

There’s an alternate form of vec! for repeating an initial value:

let v = vec![0; 10]; // ten zeroes

Accessing elements

To get the value at a particular index in the vector, we use []s:

let v = vec![1, 2, 3, 4, 5];
  
  println!("The third element of v is {}", v[2]);

The indices count from 0, so the third element is v[2].

Iterating

Once you have a vector, you can iterate through its elements with for. There are three versions:

let mut v = vec![1, 2, 3, 4, 5];
  
  for i in &v {
      println!("A reference to {}", i);
  }
  
  for i in &mut v {
      println!("A mutable reference to {}", i);
  }
  
  for i in v {
      println!("Take ownership of the vector and its element {}", i);
  }

Vectors have many more useful methods, which you can read about in their API documentation.

Strings

Strings are an important concept for any programmer to master. Rust’s string handling system is a bit different from other languages, due to its systems focus. Any time you have a data structure of variable size, things can get tricky, and strings are a re-sizable data structure. That being said, Rust’s strings also work differently than in some other systems languages, such as C.

Let’s dig into the details. A ‘string’ is a sequence of Unicode scalar values encoded as a stream of UTF-8 bytes. All strings are guaranteed to be a valid encoding of UTF-8 sequences. Additionally, unlike some systems languages, strings are not null-terminated and can contain null bytes.

Rust has two main types of strings: &str and String. Let’s talk about &str first. These are called ‘string slices’. String literals are of the type &'static str:

let greeting = "Hello there."; // greeting: &'static str

This string is statically allocated, meaning that it’s saved inside our compiled program, and exists for the entire duration it runs. The greeting binding is a reference to this statically allocated string. String slices have a fixed size, and cannot be mutated.

A String, on the other hand, is a heap-allocated string. This string is growable, and is also guaranteed to be UTF-8. Strings are commonly created by converting from a string slice using the to_string method.

let mut s = "Hello".to_string(); // mut s: String
  println!("{}", s);
  
  s.push_str(", world.");
  println!("{}", s);

Strings will coerce into &str with an &:

fn takes_slice(slice: &str) {
      println!("Got: {}", slice);
  }
  
  fn main() {
      let s = "Hello".to_string();
      takes_slice(&s);
  }

This coercion does not happen for functions that accept one of &str’s traits instead of &str. For example, TcpStream::connect has a parameter of type ToSocketAddrs. A &str is okay but a String must be explicitly converted using &*.

use std::net::TcpStream;
  
  TcpStream::connect("192.168.0.1:3000"); // &str parameter
  
  let addr_string = "192.168.0.1:3000".to_string();
  TcpStream::connect(&*addr_string); // convert addr_string to &str

Viewing a String as a &str is cheap, but converting the &str to a String involves allocating memory. No reason to do that unless you have to!

Indexing

Because strings are valid UTF-8, strings do not support indexing:

let s = "hello";
  
  println!("The first letter of s is {}", s[0]); // ERROR!!!

Usually, access to a vector with [] is very fast. But, because each character in a UTF-8 encoded string can be multiple bytes, you have to walk over the string to find the nᵗʰ letter of a string. This is a significantly more expensive operation, and we don’t want to be misleading. Furthermore, ‘letter’ isn’t something defined in Unicode, exactly. We can choose to look at a string as individual bytes, or as codepoints:

let hachiko = "忠犬ハチ公";
  
  for b in hachiko.as_bytes() {
      print!("{}, ", b);
  }
  
  println!("");
  
  for c in hachiko.chars() {
      print!("{}, ", c);
  }
  
  println!("");

This prints:

229, 191, 160, 231, 138, 172, 227, 131, 143, 227, 131, 129, 229, 133, 172, 
  忠, 犬, ハ, チ, 公, 

As you can see, there are more bytes than chars.

You can get something similar to an index like this:

let dog = hachiko.chars().nth(1); // kinda like hachiko[1]

This emphasizes that we have to walk from the beginning of the list of chars.

Slicing

You can get a slice of a string with slicing syntax:

let dog = "hachiko";
  let hachi = &dog[0..5];

But note that these are byte offsets, not character offsets. So this will fail at runtime:

let dog = "忠犬ハチ公";
  let hachi = &dog[0..2];

with this error:

thread '<main>' panicked at 'index 0 and/or 2 in `忠犬ハチ公` do not lie on
  character boundary'

Concatenation

If you have a String, you can concatenate a &str to the end of it:

let hello = "Hello ".to_string();
  let world = "world!";
  
  let hello_world = hello + world;

But if you have two Strings, you need an &:

let hello = "Hello ".to_string();
  let world = "world!".to_string();
  
  let hello_world = hello + &world;

This is because &String can automatically coerce to a &str. This is a feature called ‘Deref coercions’.

Generics

Sometimes, when writing a function or data type, we may want it to work for multiple types of arguments. In Rust, we can do this with generics. Generics are called ‘parametric polymorphism’ in type theory, which means that they are types or functions that have multiple forms (‘poly’ is multiple, ‘morph’ is form) over a given parameter (‘parametric’).

Anyway, enough type theory, let’s check out some generic code. Rust’s standard library provides a type, Option<T>, that’s generic:

enum Option<T> {
      Some(T),
      None,
  }

The <T> part, which you’ve seen a few times before, indicates that this is a generic data type. Inside the declaration of our enum, wherever we see a T, we substitute that type for the same type used in the generic. Here’s an example of using Option<T>, with some extra type annotations:

let x: Option<i32> = Some(5);

In the type declaration, we say Option<i32>. Note how similar this looks to Option<T>. So, in this particular Option, T has the value of i32. On the right-hand side of the binding, we make a Some(T), where T is 5. Since that’s an i32, the two sides match, and Rust is happy. If they didn’t match, we’d get an error:

let x: Option<f64> = Some(5);
  // error: mismatched types: expected `core::option::Option<f64>`,
  // found `core::option::Option<_>` (expected f64 but found integral variable)

That doesn’t mean we can’t make Option<T>s that hold an f64! They just have to match up:

let x: Option<i32> = Some(5);
  let y: Option<f64> = Some(5.0f64);

This is just fine. One definition, multiple uses.

Generics don’t have to only be generic over one type. Consider another type from Rust’s standard library that’s similar, Result<T, E>:

enum Result<T, E> {
      Ok(T),
      Err(E),
  }

This type is generic over two types: T and E. By the way, the capital letters can be any letter you’d like. We could define Result<T, E> as:

enum Result<A, Z> {
      Ok(A),
      Err(Z),
  }

if we wanted to. Convention says that the first generic parameter should be T, for ‘type’, and that we use E for ‘error’. Rust doesn’t care, however.

The Result<T, E> type is intended to be used to return the result of a computation, and to have the ability to return an error if it didn’t work out.

Generic functions

We can write functions that take generic types with a similar syntax:

fn takes_anything<T>(x: T) {
      // do something with x
  }

The syntax has two parts: the <T> says “this function is generic over one type, T”, and the x: T says “x has the type T.”

Multiple arguments can have the same generic type:

fn takes_two_of_the_same_things<T>(x: T, y: T) {
      // ...
  }

We could write a version that takes multiple types:

fn takes_two_things<T, U>(x: T, y: U) {
      // ...
  }

Generic structs

You can store a generic type in a struct as well:

struct Point<T> {
      x: T,
      y: T,
  }
  
  let int_origin = Point { x: 0, y: 0 };
  let float_origin = Point { x: 0.0, y: 0.0 };

Similar to functions, the <T> is where we declare the generic parameters, and we then use x: T in the type declaration, too.

When you want to add an implementation for the generic struct, you just declare the type parameter after the impl:

#
  impl<T> Point<T> {
      fn swap(&mut self) {
          std::mem::swap(&mut self.x, &mut self.y);
      }
  }

So far you’ve seen generics that take absolutely any type. These are useful in many cases: you’ve already seen Option<T>, and later you’ll meet universal container types like Vec<T>. On the other hand, often you want to trade that flexibility for increased expressive power. Read about trait bounds to see why and how.

Traits

A trait is a language feature that tells the Rust compiler about functionality a type must provide.

Do you remember the impl keyword, used to call a function with method syntax?

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  }

Traits are similar, except that we define a trait with just the method signature, then implement the trait for that struct. Like this:

struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  trait HasArea {
      fn area(&self) -> f64;
  }
  
  impl HasArea for Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  }

As you can see, the trait block looks very similar to the impl block, but we don’t define a body, just a type signature. When we impl a trait, we use impl Trait for Item, rather than just impl Item.

Traits bounds for generic functions

Traits are useful because they allow a type to make certain promises about its behavior. Generic functions can exploit this to constrain the types they accept. Consider this function, which does not compile:

fn print_area<T>(shape: T) {
      println!("This shape has an area of {}", shape.area());
  }

Rust complains:

error: no method named `area` found for type `T` in the current scope

Because T can be any type, we can’t be sure that it implements the area method. But we can add a ‘trait constraint’ to our generic T, ensuring that it does:

fn print_area<T: HasArea>(shape: T) {
      println!("This shape has an area of {}", shape.area());
  }

The syntax <T: HasArea> means “any type that implements the HasArea trait.” Because traits define function type signatures, we can be sure that any type which implements HasArea will have an .area() method.

Here’s an extended example of how this works:

trait HasArea {
      fn area(&self) -> f64;
  }
  
  struct Circle {
      x: f64,
      y: f64,
      radius: f64,
  }
  
  impl HasArea for Circle {
      fn area(&self) -> f64 {
          std::f64::consts::PI * (self.radius * self.radius)
      }
  }
  
  struct Square {
      x: f64,
      y: f64,
      side: f64,
  }
  
  impl HasArea for Square {
      fn area(&self) -> f64 {
          self.side * self.side
      }
  }
  
  fn print_area<T: HasArea>(shape: T) {
      println!("This shape has an area of {}", shape.area());
  }
  
  fn main() {
      let c = Circle {
          x: 0.0f64,
          y: 0.0f64,
          radius: 1.0f64,
      };
  
      let s = Square {
          x: 0.0f64,
          y: 0.0f64,
          side: 1.0f64,
      };
  
      print_area(c);
      print_area(s);
  }

This program outputs:

This shape has an area of 3.141593
  This shape has an area of 1

As you can see, print_area is now generic, but also ensures that we have passed in the correct types. If we pass in an incorrect type:

print_area(5);

We get a compile-time error:

error: the trait `HasArea` is not implemented for the type `_` [E0277]

Traits bounds for generic structs

Your generic structs can also benefit from trait constraints. All you need to do is append the constraint when you declare type parameters. Here is a new type Rectangle<T> and its operation is_square():

struct Rectangle<T> {
      x: T,
      y: T,
      width: T,
      height: T,
  }
  
  impl<T: PartialEq> Rectangle<T> {
      fn is_square(&self) -> bool {
          self.width == self.height
      }
  }
  
  fn main() {
      let mut r = Rectangle {
          x: 0,
          y: 0,
          width: 47,
          height: 47,
      };
  
      assert!(r.is_square());
  
      r.height = 42;
      assert!(!r.is_square());
  }

is_square() needs to check that the sides are equal, so the sides must be of a type that implements the core::cmp::PartialEq trait:

impl<T: PartialEq> Rectangle<T> { ... }

Now, a rectangle can be defined in terms of any type that can be compared for equality.

Here we defined a new struct Rectangle that accepts numbers of any precision—really, objects of pretty much any type—as long as they can be compared for equality. Could we do the same for our HasArea structs, Square and Circle? Yes, but they need multiplication, and to work with that we need to know more about operator traits.

Rules for implementing traits

So far, we’ve only added trait implementations to structs, but you can implement a trait for any type. So technically, we could implement HasArea for i32:

trait HasArea {
      fn area(&self) -> f64;
  }
  
  impl HasArea for i32 {
      fn area(&self) -> f64 {
          println!("this is silly");
  
          *self as f64
      }
  }
  
  5.area();

It is considered poor style to implement methods on such primitive types, even though it is possible.

This may seem like the Wild West, but there are two restrictions around implementing traits that prevent this from getting out of hand. The first is that if the trait isn’t defined in your scope, it doesn’t apply. Here’s an example: the standard library provides a Write trait which adds extra functionality to Files, for doing file I/O. By default, a File won’t have its methods:

let mut f = std::fs::File::open("foo.txt").ok().expect("Couldn’t open foo.txt");
  let buf = b"whatever"; // byte string literal. buf: &[u8; 8]
  let result = f.write(buf);

Here’s the error:

error: type `std::fs::File` does not implement any method in scope named `write`
  let result = f.write(buf);
                 ^~~~~~~~~~

We need to use the Write trait first:

use std::io::Write;
  
  let mut f = std::fs::File::open("foo.txt").ok().expect("Couldn’t open foo.txt");
  let buf = b"whatever";
  let result = f.write(buf);

This will compile without error.

This means that even if someone does something bad like add methods to i32, it won’t affect you, unless you use that trait.

There’s one more restriction on implementing traits: either the trait, or the type you’re writing the impl for, must be defined by you. So, we could implement the HasArea type for i32, because HasArea is in our code. But if we tried to implement ToString, a trait provided by Rust, for i32, we could not, because neither the trait nor the type are in our code.

One last thing about traits: generic functions with a trait bound use ‘monomorphization’ (mono: one, morph: form), so they are statically dispatched. What’s that mean? Check out the chapter on trait objects for more details.

Multiple trait bounds

You’ve seen that you can bound a generic type parameter with a trait:

fn foo<T: Clone>(x: T) {
      x.clone();
  }

If you need more than one bound, you can use +:

use std::fmt::Debug;
  
  fn foo<T: Clone + Debug>(x: T) {
      x.clone();
      println!("{:?}", x);
  }

T now needs to be both Clone as well as Debug.

Where clause

Writing functions with only a few generic types and a small number of trait bounds isn’t too bad, but as the number increases, the syntax gets increasingly awkward:

use std::fmt::Debug;
  
  fn foo<T: Clone, K: Clone + Debug>(x: T, y: K) {
      x.clone();
      y.clone();
      println!("{:?}", y);
  }

The name of the function is on the far left, and the parameter list is on the far right. The bounds are getting in the way.

Rust has a solution, and it’s called a ‘where clause’:

use std::fmt::Debug;
  
  fn foo<T: Clone, K: Clone + Debug>(x: T, y: K) {
      x.clone();
      y.clone();
      println!("{:?}", y);
  }
  
  fn bar<T, K>(x: T, y: K) where T: Clone, K: Clone + Debug {
      x.clone();
      y.clone();
      println!("{:?}", y);
  }
  
  fn main() {
      foo("Hello", "world");
      bar("Hello", "world");
  }

foo() uses the syntax we showed earlier, and bar() uses a where clause. All you need to do is leave off the bounds when defining your type parameters, and then add where after the parameter list. For longer lists, whitespace can be added:

use std::fmt::Debug;
  
  fn bar<T, K>(x: T, y: K)
      where T: Clone,
            K: Clone + Debug {
  
      x.clone();
      y.clone();
      println!("{:?}", y);
  }

This flexibility can add clarity in complex situations.

where is also more powerful than the simpler syntax. For example:

trait ConvertTo<Output> {
      fn convert(&self) -> Output;
  }
  
  impl ConvertTo<i64> for i32 {
      fn convert(&self) -> i64 { *self as i64 }
  }
  
  // can be called with T == i32
  fn normal<T: ConvertTo<i64>>(x: &T) -> i64 {
      x.convert()
  }
  
  // can be called with T == i64
  fn inverse<T>() -> T
          // this is using ConvertTo as if it were "ConvertTo<i64>"
          where i32: ConvertTo<T> {
      42.convert()
  }

This shows off the additional feature of where clauses: they allow bounds where the left-hand side is an arbitrary type (i32 in this case), not just a plain type parameter (like T).

Default methods

If you already know how a typical implementor will define a method, you can let your trait supply a default:

trait Foo {
      fn is_valid(&self) -> bool;
  
      fn is_invalid(&self) -> bool { !self.is_valid() }
  }

Implementors of the Foo trait need to implement is_valid(), but they don’t need to implement is_invalid(). They’ll get this default behavior. They can override the default if they so choose:

#
  struct UseDefault;
  
  impl Foo for UseDefault {
      fn is_valid(&self) -> bool {
          println!("Called UseDefault.is_valid.");
          true
      }
  }
  
  struct OverrideDefault;
  
  impl Foo for OverrideDefault {
      fn is_valid(&self) -> bool {
          println!("Called OverrideDefault.is_valid.");
          true
      }
  
      fn is_invalid(&self) -> bool {
          println!("Called OverrideDefault.is_invalid!");
          true // this implementation is a self-contradiction!
      }
  }
  
  let default = UseDefault;
  assert!(!default.is_invalid()); // prints "Called UseDefault.is_valid."
  
  let over = OverrideDefault;
  assert!(over.is_invalid()); // prints "Called OverrideDefault.is_invalid!"

Inheritance

Sometimes, implementing a trait requires implementing another trait:

trait Foo {
      fn foo(&self);
  }
  
  trait FooBar : Foo {
      fn foobar(&self);
  }

Implementors of FooBar must also implement Foo, like this:

struct Baz;
  
  impl Foo for Baz {
      fn foo(&self) { println!("foo"); }
  }
  
  impl FooBar for Baz {
      fn foobar(&self) { println!("foobar"); }
  }

If we forget to implement Foo, Rust will tell us:

error: the trait `main::Foo` is not implemented for the type `main::Baz` [E0277]

Drop

Now that we’ve discussed traits, let’s talk about a particular trait provided by the Rust standard library, Drop. The Drop trait provides a way to run some code when a value goes out of scope. For example:

struct HasDrop;
  
  impl Drop for HasDrop {
      fn drop(&mut self) {
          println!("Dropping!");
      }
  }
  
  fn main() {
      let x = HasDrop;
  
      // do stuff
  
  } // x goes out of scope here

When x goes out of scope at the end of main(), the code for Drop will run. Drop has one method, which is also called drop(). It takes a mutable reference to self.

That’s it! The mechanics of Drop are very simple, but there are some subtleties. For example, values are dropped in the opposite order they are declared. Here’s another example:

struct Firework {
      strength: i32,
  }
  
  impl Drop for Firework {
      fn drop(&mut self) {
          println!("BOOM times {}!!!", self.strength);
      }
  }
  
  fn main() {
      let firecracker = Firework { strength: 1 };
      let tnt = Firework { strength: 100 };
  }

This will output:

BOOM times 100!!!
  BOOM times 1!!!

The TNT goes off before the firecracker does, because it was declared afterwards. Last in, first out.

So what is Drop good for? Generally, Drop is used to clean up any resources associated with a struct. For example, the Arc<T> type is a reference-counted type. When Drop is called, it will decrement the reference count, and if the total number of references is zero, will clean up the underlying value.

if let

if let allows you to combine if and let together to reduce the overhead of certain kinds of pattern matches.

For example, let’s say we have some sort of Option<T>. We want to call a function on it if it’s Some<T>, but do nothing if it’s None. That looks like this:

match option {
      Some(x) => { foo(x) },
      None => {},
  }

We don’t have to use match here, for example, we could use if:

if option.is_some() {
      let x = option.unwrap();
      foo(x);
  }

Neither of these options is particularly appealing. We can use if let to do the same thing in a nicer way:

if let Some(x) = option {
      foo(x);
  }

If a pattern matches successfully, it binds any appropriate parts of the value to the identifiers in the pattern, then evaluates the expression. If the pattern doesn’t match, nothing happens.

If you’d rather to do something else when the pattern does not match, you can use else:

if let Some(x) = option {
      foo(x);
  } else {
      bar();
  }

while let

In a similar fashion, while let can be used when you want to conditionally loop as long as a value matches a certain pattern. It turns code like this:

loop {
      match option {
          Some(x) => println!("{}", x),
          _ => break,
      }
  }

Into code like this:

while let Some(x) = option {
      println!("{}", x);
  }

Trait Objects

When code involves polymorphism, there needs to be a mechanism to determine which specific version is actually run. This is called ‘dispatch’. There are two major forms of dispatch: static dispatch and dynamic dispatch. While Rust favors static dispatch, it also supports dynamic dispatch through a mechanism called ‘trait objects’.

Background

For the rest of this chapter, we’ll need a trait and some implementations. Let’s make a simple one, Foo. It has one method that is expected to return a String.

trait Foo {
      fn method(&self) -> String;
  }

We’ll also implement this trait for u8 and String:

impl Foo for u8 {
      fn method(&self) -> String { format!("u8: {}", *self) }
  }
  
  impl Foo for String {
      fn method(&self) -> String { format!("string: {}", *self) }
  }

Static dispatch

We can use this trait to perform static dispatch with trait bounds:

fn do_something<T: Foo>(x: T) {
      x.method();
  }
  
  fn main() {
      let x = 5u8;
      let y = "Hello".to_string();
  
      do_something(x);
      do_something(y);
  }

Rust uses ‘monomorphization’ to perform static dispatch here. This means that Rust will create a special version of do_something() for both u8 and String, and then replace the call sites with calls to these specialized functions. In other words, Rust generates something like this:

fn do_something_u8(x: u8) {
      x.method();
  }
  
  fn do_something_string(x: String) {
      x.method();
  }
  
  fn main() {
      let x = 5u8;
      let y = "Hello".to_string();
  
      do_something_u8(x);
      do_something_string(y);
  }

This has a great upside: static dispatch allows function calls to be inlined because the callee is known at compile time, and inlining is the key to good optimization. Static dispatch is fast, but it comes at a tradeoff: ‘code bloat’, due to many copies of the same function existing in the binary, one for each type.

Furthermore, compilers aren’t perfect and may “optimize” code to become slower. For example, functions inlined too eagerly will bloat the instruction cache (cache rules everything around us). This is part of the reason that #[inline] and #[inline(always)] should be used carefully, and one reason why using a dynamic dispatch is sometimes more efficient.

However, the common case is that it is more efficient to use static dispatch, and one can always have a thin statically-dispatched wrapper function that does a dynamic dispatch, but not vice versa, meaning static calls are more flexible. The standard library tries to be statically dispatched where possible for this reason.

Dynamic dispatch

Rust provides dynamic dispatch through a feature called ‘trait objects’. Trait objects, like &Foo or Box<Foo>, are normal values that store a value of any type that implements the given trait, where the precise type can only be known at runtime.

A trait object can be obtained from a pointer to a concrete type that implements the trait by casting it (e.g. &x as &Foo) or coercing it (e.g. using &x as an argument to a function that takes &Foo).

These trait object coercions and casts also work for pointers like &mut T to &mut Foo and Box<T> to Box<Foo>, but that’s all at the moment. Coercions and casts are identical.

This operation can be seen as ‘erasing’ the compiler’s knowledge about the specific type of the pointer, and hence trait objects are sometimes referred to as ‘type erasure’.

Coming back to the example above, we can use the same trait to perform dynamic dispatch with trait objects by casting:


  fn do_something(x: &Foo) {
      x.method();
  }
  
  fn main() {
      let x = 5u8;
      do_something(&x as &Foo);
  }

or by coercing:


  fn do_something(x: &Foo) {
      x.method();
  }
  
  fn main() {
      let x = "Hello".to_string();
      do_something(&x);
  }

A function that takes a trait object is not specialized to each of the types that implements Foo: only one copy is generated, often (but not always) resulting in less code bloat. However, this comes at the cost of requiring slower virtual function calls, and effectively inhibiting any chance of inlining and related optimizations from occurring.

Why pointers?

Rust does not put things behind a pointer by default, unlike many managed languages, so types can have different sizes. Knowing the size of the value at compile time is important for things like passing it as an argument to a function, moving it about on the stack and allocating (and deallocating) space on the heap to store it.

For Foo, we would need to have a value that could be at least either a String (24 bytes) or a u8 (1 byte), as well as any other type for which dependent crates may implement Foo (any number of bytes at all). There’s no way to guarantee that this last point can work if the values are stored without a pointer, because those other types can be arbitrarily large.

Putting the value behind a pointer means the size of the value is not relevant when we are tossing a trait object around, only the size of the pointer itself.

Representation

The methods of the trait can be called on a trait object via a special record of function pointers traditionally called a ‘vtable’ (created and managed by the compiler).

Trait objects are both simple and complicated: their core representation and layout is quite straight-forward, but there are some curly error messages and surprising behaviors to discover.

Let’s start simple, with the runtime representation of a trait object. The std::raw module contains structs with layouts that are the same as the complicated built-in types, including trait objects:

pub struct TraitObject {
      pub data: *mut (),
      pub vtable: *mut (),
  }

That is, a trait object like &Foo consists of a ‘data’ pointer and a ‘vtable’ pointer.

The data pointer addresses the data (of some unknown type T) that the trait object is storing, and the vtable pointer points to the vtable (‘virtual method table’) corresponding to the implementation of Foo for T.

A vtable is essentially a struct of function pointers, pointing to the concrete piece of machine code for each method in the implementation. A method call like trait_object.method() will retrieve the correct pointer out of the vtable and then do a dynamic call of it. For example:

struct FooVtable {
      destructor: fn(*mut ()),
      size: usize,
      align: usize,
      method: fn(*const ()) -> String,
  }
  
  // u8:
  
  fn call_method_on_u8(x: *const ()) -> String {
      // the compiler guarantees that this function is only called
      // with `x` pointing to a u8
      let byte: &u8 = unsafe { &*(x as *const u8) };
  
      byte.method()
  }
  
  static Foo_for_u8_vtable: FooVtable = FooVtable {
      destructor: /* compiler magic */,
      size: 1,
      align: 1,
  
      // cast to a function pointer
      method: call_method_on_u8 as fn(*const ()) -> String,
  };
  
  
  // String:
  
  fn call_method_on_String(x: *const ()) -> String {
      // the compiler guarantees that this function is only called
      // with `x` pointing to a String
      let string: &String = unsafe { &*(x as *const String) };
  
      string.method()
  }
  
  static Foo_for_String_vtable: FooVtable = FooVtable {
      destructor: /* compiler magic */,
      // values for a 64-bit computer, halve them for 32-bit ones
      size: 24,
      align: 8,
  
      method: call_method_on_String as fn(*const ()) -> String,
  };

The destructor field in each vtable points to a function that will clean up any resources of the vtable’s type: for u8 it is trivial, but for String it will free the memory. This is necessary for owning trait objects like Box<Foo>, which need to clean-up both the Box allocation as well as the internal type when they go out of scope. The size and align fields store the size of the erased type, and its alignment requirements; these are essentially unused at the moment since the information is embedded in the destructor, but will be used in the future, as trait objects are progressively made more flexible.

Suppose we’ve got some values that implement Foo. The explicit form of construction and use of Foo trait objects might look a bit like (ignoring the type mismatches: they’re all just pointers anyway):

let a: String = "foo".to_string();
  let x: u8 = 1;
  
  // let b: &Foo = &a;
  let b = TraitObject {
      // store the data
      data: &a,
      // store the methods
      vtable: &Foo_for_String_vtable
  };
  
  // let y: &Foo = x;
  let y = TraitObject {
      // store the data
      data: &x,
      // store the methods
      vtable: &Foo_for_u8_vtable
  };
  
  // b.method();
  (b.vtable.method)(b.data);
  
  // y.method();
  (y.vtable.method)(y.data);

Object Safety

Not every trait can be used to make a trait object. For example, vectors implement Clone, but if we try to make a trait object:

let v = vec![1, 2, 3];
  let o = &v as &Clone;

We get an error:

error: cannot convert to a trait object because trait `core::clone::Clone` is not obje
  ↳ ct-safe [E0038]
  let o = &v as &Clone;
          ^~
  note: the trait cannot require that `Self : Sized`
  let o = &v as &Clone;
          ^~

The error says that Clone is not ‘object-safe’. Only traits that are object-safe can be made into trait objects. A trait is object-safe if both of these are true:

  • the trait does not require that Self: Sized
  • all of its methods are object-safe

So what makes a method object-safe? Each method must require that Self: Sized or all of the following:

  • must not have any type parameters
  • must not use Self

Whew! As we can see, almost all of these rules talk about Self. A good intuition is “except in special circumstances, if your trait’s method uses Self, it is not object-safe.”

Closures

Rust not only has named functions, but anonymous functions as well. Anonymous functions that have an associated environment are called ‘closures’, because they close over an environment. Rust has a really great implementation of them, as we’ll see.

Syntax

Closures look like this:

let plus_one = |x: i32| x + 1;
  
  assert_eq!(2, plus_one(1));

We create a binding, plus_one, and assign it to a closure. The closure’s arguments go between the pipes (|), and the body is an expression, in this case, x + 1. Remember that { } is an expression, so we can have multi-line closures too:

let plus_two = |x| {
      let mut result: i32 = x;
  
      result += 1;
      result += 1;
  
      result
  };
  
  assert_eq!(4, plus_two(2));

You’ll notice a few things about closures that are a bit different from regular functions defined with fn. The first is that we did not need to annotate the types of arguments the closure takes or the values it returns. We can:

let plus_one = |x: i32| -> i32 { x + 1 };
  
  assert_eq!(2, plus_one(1));

But we don’t have to. Why is this? Basically, it was chosen for ergonomic reasons. While specifying the full type for named functions is helpful with things like documentation and type inference, the types of closures are rarely documented since they’re anonymous, and they don’t cause the kinds of error-at-a-distance problems that inferring named function types can.

The second is that the syntax is similar, but a bit different. I’ve added spaces here for easier comparison:

fn  plus_one_v1   (x: i32) -> i32 { x + 1 }
  let plus_one_v2 = |x: i32| -> i32 { x + 1 };
  let plus_one_v3 = |x: i32|          x + 1  ;

Small differences, but they’re similar.

Closures and their environment

Closures are called such because they ‘close over their environment’. It looks like this:

let num = 5;
  let plus_num = |x: i32| x + num;
  
  assert_eq!(10, plus_num(5));

This closure, plus_num, refers to a let binding in its scope: num. More specifically, it borrows the binding. If we do something that would conflict with that binding, we get an error. Like this one:

let mut num = 5;
  let plus_num = |x: i32| x + num;
  
  let y = &mut num;

Which errors with:

error: cannot borrow `num` as mutable because it is also borrowed as immutable
      let y = &mut num;
                   ^~~
  note: previous borrow of `num` occurs here due to use in closure; the immutable
    borrow prevents subsequent moves or mutable borrows of `num` until the borrow
    ends
      let plus_num = |x| x + num;
                     ^~~~~~~~~~~
  note: previous borrow ends here
  fn main() {
      let mut num = 5;
      let plus_num = |x| x + num;
  
      let y = &mut num;
  }
  ^

A verbose yet helpful error message! As it says, we can’t take a mutable borrow on num because the closure is already borrowing it. If we let the closure go out of scope, we can:

let mut num = 5;
  {
      let plus_num = |x: i32| x + num;
  
  } // plus_num goes out of scope, borrow of num ends
  
  let y = &mut num;

If your closure requires it, however, Rust will take ownership and move the environment instead. This doesn’t work:

let nums = vec![1, 2, 3];
  
  let takes_nums = || nums;
  
  println!("{:?}", nums);

We get this error:

note: `nums` moved into closure environment here because it has type
    `[closure(()) -> collections::vec::Vec<i32>]`, which is non-copyable
  let takes_nums = || nums;
                   ^~~~~~~

Vec<T> has ownership over its contents, and therefore, when we refer to it in our closure, we have to take ownership of nums. It’s the same as if we’d passed nums to a function that took ownership of it.

move closures

We can force our closure to take ownership of its environment with the move keyword:

let num = 5;
  
  let owns_num = move |x: i32| x + num;

Now, even though the keyword is move, the variables follow normal move semantics. In this case, 5 implements Copy, and so owns_num takes ownership of a copy of num. So what’s the difference?

let mut num = 5;
  
  {
      let mut add_num = |x: i32| num += x;
  
      add_num(5);
  }
  
  assert_eq!(10, num);

So in this case, our closure took a mutable reference to num, and then when we called add_num, it mutated the underlying value, as we’d expect. We also needed to declare add_num as mut too, because we’re mutating its environment.

If we change to a move closure, it’s different:

let mut num = 5;
  
  {
      let mut add_num = move |x: i32| num += x;
  
      add_num(5);
  }
  
  assert_eq!(5, num);

We only get 5. Rather than taking a mutable borrow out on our num, we took ownership of a copy.

Another way to think about move closures: they give a closure its own stack frame. Without move, a closure may be tied to the stack frame that created it, while a move closure is self-contained. This means that you cannot generally return a non-move closure from a function, for example.

But before we talk about taking and returning closures, we should talk some more about the way that closures are implemented. As a systems language, Rust gives you tons of control over what your code does, and closures are no different.

Closure implementation

Rust’s implementation of closures is a bit different than other languages. They are effectively syntax sugar for traits. You’ll want to make sure to have read the traits chapter before this one, as well as the chapter on trait objects.

Got all that? Good.

The key to understanding how closures work under the hood is something a bit strange: Using () to call a function, like foo(), is an overloadable operator. From this, everything else clicks into place. In Rust, we use the trait system to overload operators. Calling functions is no different. We have three separate traits to overload with:

pub trait Fn<Args> : FnMut<Args> {
      extern "rust-call" fn call(&self, args: Args) -> Self::Output;
  }
  
  pub trait FnMut<Args> : FnOnce<Args> {
      extern "rust-call" fn call_mut(&mut self, args: Args) -> Self::Output;
  }
  
  pub trait FnOnce<Args> {
      type Output;
  
      extern "rust-call" fn call_once(self, args: Args) -> Self::Output;
  }

You’ll notice a few differences between these traits, but a big one is self: Fn takes &self, FnMut takes &mut self, and FnOnce takes self. This covers all three kinds of self via the usual method call syntax. But we’ve split them up into three traits, rather than having a single one. This gives us a large amount of control over what kind of closures we can take.

The || {} syntax for closures is sugar for these three traits. Rust will generate a struct for the environment, impl the appropriate trait, and then use it.

Taking closures as arguments

Now that we know that closures are traits, we already know how to accept and return closures: just like any other trait!

This also means that we can choose static vs dynamic dispatch as well. First, let’s write a function which takes something callable, calls it, and returns the result:

fn call_with_one<F>(some_closure: F) -> i32
      where F : Fn(i32) -> i32 {
  
      some_closure(1)
  }
  
  let answer = call_with_one(|x| x + 2);
  
  assert_eq!(3, answer);

We pass our closure, |x| x + 2, to call_with_one. It just does what it suggests: it calls the closure, giving it 1 as an argument.

Let’s examine the signature of call_with_one in more depth:

fn call_with_one<F>(some_closure: F) -> i32

We take one parameter, and it has the type F. We also return a i32. This part isn’t interesting. The next part is:

    where F : Fn(i32) -> i32 {

Because Fn is a trait, we can bound our generic with it. In this case, our closure takes a i32 as an argument and returns an i32, and so the generic bound we use is Fn(i32) -> i32.

There’s one other key point here: because we’re bounding a generic with a trait, this will get monomorphized, and therefore, we’ll be doing static dispatch into the closure. That’s pretty neat. In many languages, closures are inherently heap allocated, and will always involve dynamic dispatch. In Rust, we can stack allocate our closure environment, and statically dispatch the call. This happens quite often with iterators and their adapters, which often take closures as arguments.

Of course, if we want dynamic dispatch, we can get that too. A trait object handles this case, as usual:

fn call_with_one(some_closure: &Fn(i32) -> i32) -> i32 {
      some_closure(1)
  }
  
  let answer = call_with_one(&|x| x + 2);
  
  assert_eq!(3, answer);

Now we take a trait object, a &Fn. And we have to make a reference to our closure when we pass it to call_with_one, so we use &||.

Function pointers and closures

A function pointer is kind of like a closure that has no environment. As such, you can pass a function pointer to any function expecting a closure argument, and it will work:

fn call_with_one(some_closure: &Fn(i32) -> i32) -> i32 {
      some_closure(1)
  }
  
  fn add_one(i: i32) -> i32 {
      i + 1
  }
  
  let f = add_one;
  
  let answer = call_with_one(&f);
  
  assert_eq!(2, answer);

In this example, we don’t strictly need the intermediate variable f, the name of the function works just fine too:

let answer = call_with_one(&add_one);

Returning closures

It’s very common for functional-style code to return closures in various situations. If you try to return a closure, you may run into an error. At first, it may seem strange, but we’ll figure it out. Here’s how you’d probably try to return a closure from a function:

fn factory() -> (Fn(i32) -> i32) {
      let num = 5;
  
      |x| x + num
  }
  
  let f = factory();
  
  let answer = f(1);
  assert_eq!(6, answer);

This gives us these long, related errors:

error: the trait `core::marker::Sized` is not implemented for the type
  `core::ops::Fn(i32) -> i32` [E0277]
  fn factory() -> (Fn(i32) -> i32) {
                  ^~~~~~~~~~~~~~~~
  note: `core::ops::Fn(i32) -> i32` does not have a constant size known at compile-time
  fn factory() -> (Fn(i32) -> i32) {
                  ^~~~~~~~~~~~~~~~
  error: the trait `core::marker::Sized` is not implemented for the type `core::ops::Fn(
  ↳ i32) -> i32` [E0277]
  let f = factory();
      ^
  note: `core::ops::Fn(i32) -> i32` does not have a constant size known at compile-time
  let f = factory();
      ^

In order to return something from a function, Rust needs to know what size the return type is. But since Fn is a trait, it could be various things of various sizes: many different types can implement Fn. An easy way to give something a size is to take a reference to it, as references have a known size. So we’d write this:

fn factory() -> &(Fn(i32) -> i32) {
      let num = 5;
  
      |x| x + num
  }
  
  let f = factory();
  
  let answer = f(1);
  assert_eq!(6, answer);

But we get another error:

error: missing lifetime specifier [E0106]
  fn factory() -> &(Fn(i32) -> i32) {
                  ^~~~~~~~~~~~~~~~~

Right. Because we have a reference, we need to give it a lifetime. But our factory() function takes no arguments, so elision doesn’t kick in here. What lifetime can we choose? 'static:

fn factory() -> &'static (Fn(i32) -> i32) {
      let num = 5;
  
      |x| x + num
  }
  
  let f = factory();
  
  let answer = f(1);
  assert_eq!(6, answer);

But we get another error:

error: mismatched types:
   expected `&'static core::ops::Fn(i32) -> i32`,
      found `[closure <anon>:7:9: 7:20]`
  (expected &-ptr,
      found closure) [E0308]
           |x| x + num
           ^~~~~~~~~~~
  

This error is letting us know that we don’t have a &'static Fn(i32) -> i32, we have a [closure <anon>:7:9: 7:20]. Wait, what?

Because each closure generates its own environment struct and implementation of Fn and friends, these types are anonymous. They exist just solely for this closure. So Rust shows them as closure <anon>, rather than some autogenerated name.

But why doesn’t our closure implement &'static Fn? Well, as we discussed before, closures borrow their environment. And in this case, our environment is based on a stack-allocated 5, the num variable binding. So the borrow has a lifetime of the stack frame. So if we returned this closure, the function call would be over, the stack frame would go away, and our closure is capturing an environment of garbage memory!

So what to do? This almost works:

fn factory() -> Box<Fn(i32) -> i32> {
      let num = 5;
  
      Box::new(|x| x + num)
  }
  let f = factory();
  
  let answer = f(1);
  assert_eq!(6, answer);

We use a trait object, by Boxing up the Fn. There’s just one last problem:

error: closure may outlive the current function, but it borrows `num`,
  which is owned by the current function [E0373]
  Box::new(|x| x + num)
           ^~~~~~~~~~~

We still have a reference to the parent stack frame. With one last fix, we can make this work:

fn factory() -> Box<Fn(i32) -> i32> {
      let num = 5;
  
      Box::new(move |x| x + num)
  }
  let f = factory();
  
  let answer = f(1);
  assert_eq!(6, answer);

By making the inner closure a move Fn, we create a new stack frame for our closure. By Boxing it up, we’ve given it a known size, and allowing it to escape our stack frame.

Universal Function Call Syntax

Sometimes, functions can have the same names. Consider this code:

trait Foo {
      fn f(&self);
  }
  
  trait Bar {
      fn f(&self);
  }
  
  struct Baz;
  
  impl Foo for Baz {
      fn f(&self) { println!("Baz’s impl of Foo"); }
  }
  
  impl Bar for Baz {
      fn f(&self) { println!("Baz’s impl of Bar"); }
  }
  
  let b = Baz;

If we were to try to call b.f(), we’d get an error:

error: multiple applicable methods in scope [E0034]
  b.f();
    ^~~
  note: candidate #1 is defined in an impl of the trait `main::Foo` for the type
  `main::Baz`
      fn f(&self) { println!("Baz’s impl of Foo"); }
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  note: candidate #2 is defined in an impl of the trait `main::Bar` for the type
  `main::Baz`
      fn f(&self) { println!("Baz’s impl of Bar"); }
      ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  

We need a way to disambiguate which method we need. This feature is called ‘universal function call syntax’, and it looks like this:

Foo::f(&b);
  Bar::f(&b);

Let’s break it down.

Foo::
  Bar::

These halves of the invocation are the types of the two traits: Foo and Bar. This is what ends up actually doing the disambiguation between the two: Rust calls the one from the trait name you use.

f(&b)

When we call a method like b.f() using method syntax, Rust will automatically borrow b if f() takes &self. In this case, Rust will not, and so we need to pass an explicit &b.

Angle-bracket Form

The form of UFCS we just talked about:

Trait::method(args);

Is a short-hand. There’s an expanded form of this that’s needed in some situations:

<Type as Trait>::method(args);

The <>:: syntax is a means of providing a type hint. The type goes inside the <>s. In this case, the type is Type as Trait, indicating that we want Trait’s version of method to be called here. The as Trait part is optional if it’s not ambiguous. Same with the angle brackets, hence the shorter form.

Here’s an example of using the longer form.

trait Foo {
      fn clone(&self);
  }
  
  #[derive(Clone)]
  struct Bar;
  
  impl Foo for Bar {
      fn clone(&self) {
          println!("Making a clone of Bar");
  
          <Bar as Clone>::clone(self);
      }
  }

This will call the Clone trait’s clone() method, rather than Foo’s.

Crates and Modules

When a project starts getting large, it’s considered good software engineering practice to split it up into a bunch of smaller pieces, and then fit them together. It’s also important to have a well-defined interface, so that some of your functionality is private, and some is public. To facilitate these kinds of things, Rust has a module system.

Basic terminology: Crates and Modules

Rust has two distinct terms that relate to the module system: ‘crate’ and ‘module’. A crate is synonymous with a ‘library’ or ‘package’ in other languages. Hence “Cargo” as the name of Rust’s package management tool: you ship your crates to others with Cargo. Crates can produce an executable or a library, depending on the project.

Each crate has an implicit root module that contains the code for that crate. You can then define a tree of sub-modules under that root module. Modules allow you to partition your code within the crate itself.

As an example, let’s make a phrases crate, which will give us various phrases in different languages. To keep things simple, we’ll stick to ‘greetings’ and ‘farewells’ as two kinds of phrases, and use English and Japanese (日本語) as two languages for those phrases to be in. We’ll use this module layout:

                                    +-----------+
                                  +---| greetings |
                                  |   +-----------+
                    +---------+   |
                +---| english |---+
                |   +---------+   |   +-----------+
                |                 +---| farewells |
  +---------+   |                     +-----------+
  | phrases |---+
  +---------+   |                     +-----------+
                |                 +---| greetings |
                |   +----------+  |   +-----------+
                +---| japanese |--+
                    +----------+  |
                                  |   +-----------+
                                  +---| farewells |
                                      +-----------+

In this example, phrases is the name of our crate. All of the rest are modules. You can see that they form a tree, branching out from the crate root, which is the root of the tree: phrases itself.

Now that we have a plan, let’s define these modules in code. To start, generate a new crate with Cargo:

$ cargo new phrases
  $ cd phrases

If you remember, this generates a simple project for us:

$ tree .
  .
  ├── Cargo.toml
  └── src
      └── lib.rs
  
  1 directory, 2 files

src/lib.rs is our crate root, corresponding to the phrases in our diagram above.

Defining Modules

To define each of our modules, we use the mod keyword. Let’s make our src/lib.rs look like this:

mod english {
      mod greetings {
      }
  
      mod farewells {
      }
  }
  
  mod japanese {
      mod greetings {
      }
  
      mod farewells {
      }
  }

After the mod keyword, you give the name of the module. Module names follow the conventions for other Rust identifiers: lower_snake_case. The contents of each module are within curly braces ({}).

Within a given mod, you can declare sub-mods. We can refer to sub-modules with double-colon (::) notation: our four nested modules are english::greetings, english::farewells, japanese::greetings, and japanese::farewells. Because these sub-modules are namespaced under their parent module, the names don’t conflict: english::greetings and japanese::greetings are distinct, even though their names are both greetings.

Because this crate does not have a main() function, and is called lib.rs, Cargo will build this crate as a library:

$ cargo build
     Compiling phrases v0.0.1 (file:///home/you/projects/phrases)
  $ ls target/debug
  build  deps  examples  libphrases-a7448e02a0468eaa.rlib  native

libphrases-hash.rlib is the compiled crate. Before we see how to use this crate from another crate, let’s break it up into multiple files.

Multiple file crates

If each crate were just one file, these files would get very large. It’s often easier to split up crates into multiple files, and Rust supports this in two ways.

Instead of declaring a module like this:

mod english {
      // contents of our module go here
  }

We can instead declare our module like this:

mod english;

If we do that, Rust will expect to find either a english.rs file, or a english/mod.rs file with the contents of our module.

Note that in these files, you don’t need to re-declare the module: that’s already been done with the initial mod declaration.

Using these two techniques, we can break up our crate into two directories and seven files:

$ tree .
  .
  ├── Cargo.lock
  ├── Cargo.toml
  ├── src
  │   ├── english
  │   │   ├── farewells.rs
  │   │   ├── greetings.rs
  │   │   └── mod.rs
  │   ├── japanese
  │   │   ├── farewells.rs
  │   │   ├── greetings.rs
  │   │   └── mod.rs
  │   └── lib.rs
  └── target
      └── debug
          ├── build
          ├── deps
          ├── examples
          ├── libphrases-a7448e02a0468eaa.rlib
          └── native

src/lib.rs is our crate root, and looks like this:

mod english;
  mod japanese;

These two declarations tell Rust to look for either src/english.rs and src/japanese.rs, or src/english/mod.rs and src/japanese/mod.rs, depending on our preference. In this case, because our modules have sub-modules, we’ve chosen the second. Both src/english/mod.rs and src/japanese/mod.rs look like this:

mod greetings;
  mod farewells;

Again, these declarations tell Rust to look for either src/english/greetings.rs and src/japanese/greetings.rs or src/english/farewells/mod.rs and src/japanese/farewells/mod.rs. Because these sub-modules don’t have their own sub-modules, we’ve chosen to make them src/english/greetings.rs and src/japanese/farewells.rs. Whew!

The contents of src/english/greetings.rs and src/japanese/farewells.rs are both empty at the moment. Let’s add some functions.

Put this in src/english/greetings.rs:

fn hello() -> String {
      "Hello!".to_string()
  }

Put this in src/english/farewells.rs:

fn goodbye() -> String {
      "Goodbye.".to_string()
  }

Put this in src/japanese/greetings.rs:

fn hello() -> String {
      "こんにちは".to_string()
  }

Of course, you can copy and paste this from this web page, or just type something else. It’s not important that you actually put ‘konnichiwa’ to learn about the module system.

Put this in src/japanese/farewells.rs:

fn goodbye() -> String {
      "さようなら".to_string()
  }

(This is ‘Sayōnara’, if you’re curious.)

Now that we have some functionality in our crate, let’s try to use it from another crate.

Importing External Crates

We have a library crate. Let’s make an executable crate that imports and uses our library.

Make a src/main.rs and put this in it (it won’t quite compile yet):

extern crate phrases;
  
  fn main() {
      println!("Hello in English: {}", phrases::english::greetings::hello());
      println!("Goodbye in English: {}", phrases::english::farewells::goodbye());
  
      println!("Hello in Japanese: {}", phrases::japanese::greetings::hello());
      println!("Goodbye in Japanese: {}", phrases::japanese::farewells::goodbye());
  }

The extern crate declaration tells Rust that we need to compile and link to the phrases crate. We can then use phrases’ modules in this one. As we mentioned earlier, you can use double colons to refer to sub-modules and the functions inside of them.

(Note: when importing a crate that has dashes in its name “like-this”, which is not a valid Rust identifier, it will be converted by changing the dashes to underscores, so you would write extern crate like_this;.)

Also, Cargo assumes that src/main.rs is the crate root of a binary crate, rather than a library crate. Our package now has two crates: src/lib.rs and src/main.rs. This pattern is quite common for executable crates: most functionality is in a library crate, and the executable crate uses that library. This way, other programs can also use the library crate, and it’s also a nice separation of concerns.

This doesn’t quite work yet, though. We get four errors that look similar to this:

$ cargo build
     Compiling phrases v0.0.1 (file:///home/you/projects/phrases)
  src/main.rs:4:38: 4:72 error: function `hello` is private
  src/main.rs:4     println!("Hello in English: {}", phrases::english::greetings::hello(
  ↳ ));
                                                     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  note: in expansion of format_args!
  <std macros>:2:25: 2:58 note: expansion site
  <std macros>:1:1: 2:62 note: in expansion of print!
  <std macros>:3:1: 3:54 note: expansion site
  <std macros>:1:1: 3:58 note: in expansion of println!
  phrases/src/main.rs:4:5: 4:76 note: expansion site

By default, everything is private in Rust. Let’s talk about this in some more depth.

Exporting a Public Interface

Rust allows you to precisely control which aspects of your interface are public, and so private is the default. To make things public, you use the pub keyword. Let’s focus on the english module first, so let’s reduce our src/main.rs to just this:

extern crate phrases;
  
  fn main() {
      println!("Hello in English: {}", phrases::english::greetings::hello());
      println!("Goodbye in English: {}", phrases::english::farewells::goodbye());
  }

In our src/lib.rs, let’s add pub to the english module declaration:

pub mod english;
  mod japanese;

And in our src/english/mod.rs, let’s make both pub:

pub mod greetings;
  pub mod farewells;

In our src/english/greetings.rs, let’s add pub to our fn declaration:

pub fn hello() -> String {
      "Hello!".to_string()
  }

And also in src/english/farewells.rs:

pub fn goodbye() -> String {
      "Goodbye.".to_string()
  }

Now, our crate compiles, albeit with warnings about not using the japanese functions:

$ cargo run
     Compiling phrases v0.0.1 (file:///home/you/projects/phrases)
  src/japanese/greetings.rs:1:1: 3:2 warning: function is never used: `hello`, #[warn(dead_code)] on by default
  src/japanese/greetings.rs:1 fn hello() -> String {
  src/japanese/greetings.rs:2     "こんにちは".to_string()
  src/japanese/greetings.rs:3 }
  src/japanese/farewells.rs:1:1: 3:2 warning: function is never used: `goodbye`, #[warn(dead_code)] on by default
  src/japanese/farewells.rs:1 fn goodbye() -> String {
  src/japanese/farewells.rs:2     "さようなら".to_string()
  src/japanese/farewells.rs:3 }
       Running `target/debug/phrases`
  Hello in English: Hello!
  Goodbye in English: Goodbye.

pub also applies to structs and their member fields. In keeping with Rust’s tendency toward safety, simply making a struct public won’t automatically make its members public: you must mark the fields individually with pub.

Now that our functions are public, we can use them. Great! However, typing out phrases::english::greetings::hello() is very long and repetitive. Rust has another keyword for importing names into the current scope, so that you can refer to them with shorter names. Let’s talk about use.

Importing Modules with use

Rust has a use keyword, which allows us to import names into our local scope. Let’s change our src/main.rs to look like this:

extern crate phrases;
  
  use phrases::english::greetings;
  use phrases::english::farewells;
  
  fn main() {
      println!("Hello in English: {}", greetings::hello());
      println!("Goodbye in English: {}", farewells::goodbye());
  }

The two use lines import each module into the local scope, so we can refer to the functions by a much shorter name. By convention, when importing functions, it’s considered best practice to import the module, rather than the function directly. In other words, you can do this:

extern crate phrases;
  
  use phrases::english::greetings::hello;
  use phrases::english::farewells::goodbye;
  
  fn main() {
      println!("Hello in English: {}", hello());
      println!("Goodbye in English: {}", goodbye());
  }

But it is not idiomatic. This is significantly more likely to introduce a naming conflict. In our short program, it’s not a big deal, but as it grows, it becomes a problem. If we have conflicting names, Rust will give a compilation error. For example, if we made the japanese functions public, and tried to do this:

extern crate phrases;
  
  use phrases::english::greetings::hello;
  use phrases::japanese::greetings::hello;
  
  fn main() {
      println!("Hello in English: {}", hello());
      println!("Hello in Japanese: {}", hello());
  }

Rust will give us a compile-time error:

   Compiling phrases v0.0.1 (file:///home/you/projects/phrases)
  src/main.rs:4:5: 4:40 error: a value named `hello` has already been imported in this m
  ↳ odule [E0252]
  src/main.rs:4 use phrases::japanese::greetings::hello;
                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  error: aborting due to previous error
  Could not compile `phrases`.

If we’re importing multiple names from the same module, we don’t have to type it out twice. Instead of this:

use phrases::english::greetings;
  use phrases::english::farewells;

We can use this shortcut:

use phrases::english::{greetings, farewells};

Re-exporting with pub use

You don’t just use use to shorten identifiers. You can also use it inside of your crate to re-export a function inside another module. This allows you to present an external interface that may not directly map to your internal code organization.

Let’s look at an example. Modify your src/main.rs to read like this:

extern crate phrases;
  
  use phrases::english::{greetings,farewells};
  use phrases::japanese;
  
  fn main() {
      println!("Hello in English: {}", greetings::hello());
      println!("Goodbye in English: {}", farewells::goodbye());
  
      println!("Hello in Japanese: {}", japanese::hello());
      println!("Goodbye in Japanese: {}", japanese::goodbye());
  }

Then, modify your src/lib.rs to make the japanese mod public:

pub mod english;
  pub mod japanese;

Next, make the two functions public, first in src/japanese/greetings.rs:

pub fn hello() -> String {
      "こんにちは".to_string()
  }

And then in src/japanese/farewells.rs:

pub fn goodbye() -> String {
      "さようなら".to_string()
  }

Finally, modify your src/japanese/mod.rs to read like this:

pub use self::greetings::hello;
  pub use self::farewells::goodbye;
  
  mod greetings;
  mod farewells;

The pub use declaration brings the function into scope at this part of our module hierarchy. Because we’ve pub used this inside of our japanese module, we now have a phrases::japanese::hello() function and a phrases::japanese::goodbye() function, even though the code for them lives in phrases::japanese::greetings::hello() and phrases::japanese::farewells::goodbye(). Our internal organization doesn’t define our external interface.

Here we have a pub use for each function we want to bring into the japanese scope. We could alternatively use the wildcard syntax to include everything from greetings into the current scope: pub use self::greetings::*.

What about the self? Well, by default, use declarations are absolute paths, starting from your crate root. self makes that path relative to your current place in the hierarchy instead. There’s one more special form of use: you can use super:: to reach one level up the tree from your current location. Some people like to think of self as . and super as .., from many shells’ display for the current directory and the parent directory.

Outside of use, paths are relative: foo::bar() refers to a function inside of foo relative to where we are. If that’s prefixed with ::, as in ::foo::bar(), it refers to a different foo, an absolute path from your crate root.

This will build and run:

$ cargo run
     Compiling phrases v0.0.1 (file:///home/you/projects/phrases)
       Running `target/debug/phrases`
  Hello in English: Hello!
  Goodbye in English: Goodbye.
  Hello in Japanese: こんにちは
  Goodbye in Japanese: さようなら

Complex imports

Rust offers several advanced options that can add compactness and convenience to your extern crate and use statements. Here is an example:

extern crate phrases as sayings;
  
  use sayings::japanese::greetings as ja_greetings;
  use sayings::japanese::farewells::*;
  use sayings::english::{self, greetings as en_greetings, farewells as en_farewells};
  
  fn main() {
      println!("Hello in English; {}", en_greetings::hello());
      println!("And in Japanese: {}", ja_greetings::hello());
      println!("Goodbye in English: {}", english::farewells::goodbye());
      println!("Again: {}", en_farewells::goodbye());
      println!("And in Japanese: {}", goodbye());
  }

What’s going on here?

First, both extern crate and use allow renaming the thing that is being imported. So the crate is still called “phrases”, but here we will refer to it as “sayings”. Similarly, the first use statement pulls in the japanese::farewells module from the crate, but makes it available as jp_farewells as opposed to simply farewells. This can help to avoid ambiguity when importing similarly-named items from different places.

The second use statement uses a star glob to bring in all symbols from the sayings::japanese::farewells module. As you can see we can later refer to the Japanese goodbye function with no module qualifiers. This kind of glob should be used sparingly.

The third use statement bears more explanation. It’s using “brace expansion” globbing to compress three use statements into one (this sort of syntax may be familiar if you’ve written Linux shell scripts before). The uncompressed form of this statement would be:

use sayings::english;
  use sayings::english::greetings as en_greetings;
  use sayings::english::farewells as en_farewells;

As you can see, the curly brackets compress use statements for several items under the same path, and in this context self just refers back to that path. Note: The curly brackets cannot be nested or mixed with star globbing.

const and static

Rust has a way of defining constants with the const keyword:

const N: i32 = 5;

Unlike let bindings, you must annotate the type of a const.

Constants live for the entire lifetime of a program. More specifically, constants in Rust have no fixed address in memory. This is because they’re effectively inlined to each place that they’re used. References to the same constant are not necessarily guaranteed to refer to the same memory address for this reason.

static

Rust provides a ‘global variable’ sort of facility in static items. They’re similar to constants, but static items aren’t inlined upon use. This means that there is only one instance for each value, and it’s at a fixed location in memory.

Here’s an example:

static N: i32 = 5;

Unlike let bindings, you must annotate the type of a static.

Statics live for the entire lifetime of a program, and therefore any reference stored in a constant has a 'static lifetime:

static NAME: &'static str = "Steve";

Mutability

You can introduce mutability with the mut keyword:

static mut N: i32 = 5;

Because this is mutable, one thread could be updating N while another is reading it, causing memory unsafety. As such both accessing and mutating a static mut is unsafe, and so must be done in an unsafe block:


  unsafe {
      N += 1;
  
      println!("N: {}", N);
  }

Furthermore, any type stored in a static must be Sync, and may not have a Drop implementation.

Initializing

Both const and static have requirements for giving them a value. They may only be given a value that’s a constant expression. In other words, you cannot use the result of a function call or anything similarly complex or at runtime.

Which construct should I use?

Almost always, if you can choose between the two, choose const. It’s pretty rare that you actually want a memory location associated with your constant, and using a const allows for optimizations like constant propagation not only in your crate but downstream crates.

Attributes

Declarations can be annotated with ‘attributes’ in Rust. They look like this:

#[test]

or like this:

#![test]

The difference between the two is the !, which changes what the attribute applies to:

#[foo]
  struct Foo;
  
  mod bar {
      #![bar]
  }

The #[foo] attribute applies to the next item, which is the struct declaration. The #![bar] attribute applies to the item enclosing it, which is the mod declaration. Otherwise, they’re the same. Both change the meaning of the item they’re attached to somehow.

For example, consider a function like this:

#[test]
  fn check() {
      assert_eq!(2, 1 + 1);
  }

It is marked with #[test]. This means it’s special: when you run tests, this function will execute. When you compile as usual, it won’t even be included. This function is now a test function.

Attributes may also have additional data:

#[inline(always)]
  fn super_fast_fn() {

Or even keys and values:

#[cfg(target_os = "macos")]
  mod macos_only {

Rust attributes are used for a number of different things. There is a full list of attributes in the reference. Currently, you are not allowed to create your own attributes, the Rust compiler defines them.

type aliases

The type keyword lets you declare an alias of another type:

type Name = String;

You can then use this type as if it were a real type:

type Name = String;
  
  let x: Name = "Hello".to_string();

Note, however, that this is an alias, not a new type entirely. In other words, because Rust is strongly typed, you’d expect a comparison between two different types to fail:

let x: i32 = 5;
  let y: i64 = 5;
  
  if x == y {
     // ...
  }

this gives

error: mismatched types:
   expected `i32`,
      found `i64`
  (expected i32,
      found i64) [E0308]
       if x == y {
               ^

But, if we had an alias:

type Num = i32;
  
  let x: i32 = 5;
  let y: Num = 5;
  
  if x == y {
     // ...
  }

This compiles without error. Values of a Num type are the same as a value of type i32, in every way.

You can also use type aliases with generics:

use std::result;
  
  enum ConcreteError {
      Foo,
      Bar,
  }
  
  type Result<T> = result::Result<T, ConcreteError>;

This creates a specialized version of the Result type, which always has a ConcreteError for the E part of Result<T, E>. This is commonly used in the standard library to create custom errors for each subsection. For example, io::Result.

Casting between types

Rust, with its focus on safety, provides two different ways of casting different types between each other. The first, as, is for safe casts. In contrast, transmute allows for arbitrary casting, and is one of the most dangerous features of Rust!

as

The as keyword does basic casting:

let x: i32 = 5;
  
  let y = x as i64;

It only allows certain kinds of casting, however:

let a = [0u8, 0u8, 0u8, 0u8];
  
  let b = a as u32; // four eights makes 32

This errors with:

error: non-scalar cast: `[u8; 4]` as `u32`
  let b = a as u32; // four eights makes 32
          ^~~~~~~~

It’s a ‘non-scalar cast’ because we have multiple values here: the four elements of the array. These kinds of casts are very dangerous, because they make assumptions about the way that multiple underlying structures are implemented. For this, we need something more dangerous.

transmute

The transmute function is provided by a compiler intrinsic, and what it does is very simple, but very scary. It tells Rust to treat a value of one type as though it were another type. It does this regardless of the typechecking system, and just completely trusts you.

In our previous example, we know that an array of four u8s represents a u32 properly, and so we want to do the cast. Using transmute instead of as, Rust lets us:

use std::mem;
  
  unsafe {
      let a = [0u8, 0u8, 0u8, 0u8];
  
      let b = mem::transmute::<[u8; 4], u32>(a);
  }

We have to wrap the operation in an unsafe block for this to compile successfully. Technically, only the mem::transmute call itself needs to be in the block, but it’s nice in this case to enclose everything related, so you know where to look. In this case, the details about a are also important, and so they’re in the block. You’ll see code in either style, sometimes the context is too far away, and wrapping all of the code in unsafe isn’t a great idea.

While transmute does very little checking, it will at least make sure that the types are the same size. This errors:

use std::mem;
  
  unsafe {
      let a = [0u8, 0u8, 0u8, 0u8];
  
      let b = mem::transmute::<[u8; 4], u64>(a);
  }

with:

error: transmute called on types with different sizes: [u8; 4] (32 bits) to u64
  (64 bits)

Other than that, you’re on your own!

Associated Types

Associated types are a powerful part of Rust’s type system. They’re related to the idea of a ‘type family’, in other words, grouping multiple types together. That description is a bit abstract, so let’s dive right into an example. If you want to write a Graph trait, you have two types to be generic over: the node type and the edge type. So you might write a trait, Graph<N, E>, that looks like this:

trait Graph<N, E> {
      fn has_edge(&self, &N, &N) -> bool;
      fn edges(&self, &N) -> Vec<E>;
      // etc
  }

While this sort of works, it ends up being awkward. For example, any function that wants to take a Graph as a parameter now also needs to be generic over the Node and Edge types too:

fn distance<N, E, G: Graph<N, E>>(graph: &G, start: &N, end: &N) -> u32 { ... }

Our distance calculation works regardless of our Edge type, so the E stuff in this signature is just a distraction.

What we really want to say is that a certain Edge and Node type come together to form each kind of Graph. We can do that with associated types:

trait Graph {
      type N;
      type E;
  
      fn has_edge(&self, &Self::N, &Self::N) -> bool;
      fn edges(&self, &Self::N) -> Vec<Self::E>;
      // etc
  }

Now, our clients can be abstract over a given Graph:

fn distance<G: Graph>(graph: &G, start: &G::N, end: &G::N) -> u32 { ... }

No need to deal with the Edge type here!

Let’s go over all this in more detail.

Defining associated types

Let’s build that Graph trait. Here’s the definition:

trait Graph {
      type N;
      type E;
  
      fn has_edge(&self, &Self::N, &Self::N) -> bool;
      fn edges(&self, &Self::N) -> Vec<Self::E>;
  }

Simple enough. Associated types use the type keyword, and go inside the body of the trait, with the functions.

These type declarations can have all the same thing as functions do. For example, if we wanted our N type to implement Display, so we can print the nodes out, we could do this:

use std::fmt;
  
  trait Graph {
      type N: fmt::Display;
      type E;
  
      fn has_edge(&self, &Self::N, &Self::N) -> bool;
      fn edges(&self, &Self::N) -> Vec<Self::E>;
  }

Implementing associated types

Just like any trait, traits that use associated types use the impl keyword to provide implementations. Here’s a simple implementation of Graph:

struct Node;
  
  struct Edge;
  
  struct MyGraph;
  
  impl Graph for MyGraph {
      type N = Node;
      type E = Edge;
  
      fn has_edge(&self, n1: &Node, n2: &Node) -> bool {
          true
      }
  
      fn edges(&self, n: &Node) -> Vec<Edge> {
          Vec::new()
      }
  }

This silly implementation always returns true and an empty Vec<Edge>, but it gives you an idea of how to implement this kind of thing. We first need three structs, one for the graph, one for the node, and one for the edge. If it made more sense to use a different type, that would work as well, we’re just going to use structs for all three here.

Next is the impl line, which is just like implementing any other trait.

From here, we use = to define our associated types. The name the trait uses goes on the left of the =, and the concrete type we’re implementing this for goes on the right. Finally, we use the concrete types in our function declarations.

Trait objects with associated types

There’s one more bit of syntax we should talk about: trait objects. If you try to create a trait object from an associated type, like this:

let graph = MyGraph;
  let obj = Box::new(graph) as Box<Graph>;

You’ll get two errors:

error: the value of the associated type `E` (from the trait `main::Graph`) must
  be specified [E0191]
  let obj = Box::new(graph) as Box<Graph>;
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  24:44 error: the value of the associated type `N` (from the trait
  `main::Graph`) must be specified [E0191]
  let obj = Box::new(graph) as Box<Graph>;
            ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~

We can’t create a trait object like this, because we don’t know the associated types. Instead, we can write this:

let graph = MyGraph;
  let obj = Box::new(graph) as Box<Graph<N=Node, E=Edge>>;

The N=Node syntax allows us to provide a concrete type, Node, for the N type parameter. Same with E=Edge. If we didn’t provide this constraint, we couldn’t be sure which impl to match this trait object to.

Unsized Types

Most types have a particular size, in bytes, that is knowable at compile time. For example, an i32 is thirty-two bits big, or four bytes. However, there are some types which are useful to express, but do not have a defined size. These are called ‘unsized’ or ‘dynamically sized’ types. One example is [T]. This type represents a certain number of T in sequence. But we don’t know how many there are, so the size is not known.

Rust understands a few of these types, but they have some restrictions. There are three:

  1. We can only manipulate an instance of an unsized type via a pointer. An &[T] works just fine, but a [T] does not.
  2. Variables and arguments cannot have dynamically sized types.
  3. Only the last field in a struct may have a dynamically sized type; the other fields must not. Enum variants must not have dynamically sized types as data.

So why bother? Well, because [T] can only be used behind a pointer, if we didn’t have language support for unsized types, it would be impossible to write this:

impl Foo for str {

or

impl<T> Foo for [T] {

Instead, you would have to write:

impl Foo for &str {

Meaning, this implementation would only work for references, and not other types of pointers. With the impl for str, all pointers, including (at some point, there are some bugs to fix first) user-defined custom smart pointers, can use this impl.

?Sized

If you want to write a function that accepts a dynamically sized type, you can use the special bound, ?Sized:

struct Foo<T: ?Sized> {
      f: T,
  }

This ?, read as “T may be Sized”, means that this bound is special: it lets us match more kinds, not less. It’s almost like every T implicitly has T: Sized, and the ? undoes this default.

Operators and Overloading

Rust allows for a limited form of operator overloading. There are certain operators that are able to be overloaded. To support a particular operator between types, there’s a specific trait that you can implement, which then overloads the operator.

For example, the + operator can be overloaded with the Add trait:

use std::ops::Add;
  
  #[derive(Debug)]
  struct Point {
      x: i32,
      y: i32,
  }
  
  impl Add for Point {
      type Output = Point;
  
      fn add(self, other: Point) -> Point {
          Point { x: self.x + other.x, y: self.y + other.y }
      }
  }
  
  fn main() {
      let p1 = Point { x: 1, y: 0 };
      let p2 = Point { x: 2, y: 3 };
  
      let p3 = p1 + p2;
  
      println!("{:?}", p3);
  }

In main, we can use + on our two Points, since we’ve implemented Add<Output=Point> for Point.

There are a number of operators that can be overloaded this way, and all of their associated traits live in the std::ops module. Check out its documentation for the full list.

Implementing these traits follows a pattern. Let’s look at Add in more detail:

pub trait Add<RHS = Self> {
      type Output;
  
      fn add(self, rhs: RHS) -> Self::Output;
  }

There’s three types in total involved here: the type you impl Add for, RHS, which defaults to Self, and Output. For an expression let z = x + y, x is the Self type, y is the RHS, and z is the Self::Output type.

impl Add<i32> for Point {
      type Output = f64;
  
      fn add(self, rhs: i32) -> f64 {
          // add an i32 to a Point and get an f64
      }
  }

will let you do this:

let p: Point = // ...
  let x: f64 = p + 2i32;

Using operator traits in generic structs

Now that we know how operator traits are defined, we can define our HasArea trait and Square struct from the traits chapter more generically:

use std::ops::Mul;
  
  trait HasArea<T> {
      fn area(&self) -> T;
  }
  
  struct Square<T> {
      x: T,
      y: T,
      side: T,
  }
  
  impl<T> HasArea<T> for Square<T>
          where T: Mul<Output=T> + Copy {
      fn area(&self) -> T {
          self.side * self.side
      }
  }
  
  fn main() {
      let s = Square {
          x: 0.0f64,
          y: 0.0f64,
          side: 12.0f64,
      };
  
      println!("Area of s: {}", s.area());
  }

For HasArea and Square, we just declare a type parameter T and replace f64 with it. The impl needs more involved modifications:

impl<T> HasArea<T> for Square<T>
          where T: Mul<Output=T> + Copy { ... }

The area method requires that we can multiply the sides, so we declare that type T must implement std::ops::Mul. Like Add, mentioned above, Mul itself takes an Output parameter: since we know that numbers don’t change type when multiplied, we also set it to T. T must also support copying, so Rust doesn’t try to move self.side into the return value.

Deref coercions

The standard library provides a special trait, Deref. It’s normally used to overload *, the dereference operator:

use std::ops::Deref;
  
  struct DerefExample<T> {
      value: T,
  }
  
  impl<T> Deref for DerefExample<T> {
      type Target = T;
  
      fn deref(&self) -> &T {
          &self.value
      }
  }
  
  fn main() {
      let x = DerefExample { value: 'a' };
      assert_eq!('a', *x);
  }

This is useful for writing custom pointer types. However, there’s a language feature related to Deref: ‘deref coercions’. Here’s the rule: If you have a type U, and it implements Deref<Target=T>, values of &U will automatically coerce to a &T. Here’s an example:

fn foo(s: &str) {
      // borrow a string for a second
  }
  
  // String implements Deref<Target=str>
  let owned = "Hello".to_string();
  
  // therefore, this works:
  foo(&owned);

Using an ampersand in front of a value takes a reference to it. So owned is a String, &owned is an &String, and since impl Deref<Target=str> for String, &String will deref to &str, which foo() takes.

That’s it. This rule is one of the only places in which Rust does an automatic conversion for you, but it adds a lot of flexibility. For example, the Rc<T> type implements Deref<Target=T>, so this works:

use std::rc::Rc;
  
  fn foo(s: &str) {
      // borrow a string for a second
  }
  
  // String implements Deref<Target=str>
  let owned = "Hello".to_string();
  let counted = Rc::new(owned);
  
  // therefore, this works:
  foo(&counted);

All we’ve done is wrap our String in an Rc<T>. But we can now pass the Rc<String> around anywhere we’d have a String. The signature of foo didn’t change, but works just as well with either type. This example has two conversions: Rc<String> to String and then String to &str. Rust will do this as many times as possible until the types match.

Another very common implementation provided by the standard library is:

fn foo(s: &[i32]) {
      // borrow a slice for a second
  }
  
  // Vec<T> implements Deref<Target=[T]>
  let owned = vec![1, 2, 3];
  
  foo(&owned);

Vectors can Deref to a slice.

Deref and method calls

Deref will also kick in when calling a method. Consider the following example.

struct Foo;
  
  impl Foo {
      fn foo(&self) { println!("Foo"); }
  }
  
  let f = &&Foo;
  
  f.foo();

Even though f is a &&Foo and foo takes &self, this works. That’s because these things are the same:

f.foo();
  (&f).foo();
  (&&f).foo();
  (&&&&&&&&f).foo();

A value of type &&&&&&&&&&&&&&&&Foo can still have methods defined on Foo called, because the compiler will insert as many * operations as necessary to get it right. And since it’s inserting *s, that uses Deref.

Macros

By now you’ve learned about many of the tools Rust provides for abstracting and reusing code. These units of code reuse have a rich semantic structure. For example, functions have a type signature, type parameters have trait bounds, and overloaded functions must belong to a particular trait.

This structure means that Rust’s core abstractions have powerful compile-time correctness checking. But this comes at the price of reduced flexibility. If you visually identify a pattern of repeated code, you may find it’s difficult or cumbersome to express that pattern as a generic function, a trait, or anything else within Rust’s semantics.

Macros allow us to abstract at a syntactic level. A macro invocation is shorthand for an “expanded” syntactic form. This expansion happens early in compilation, before any static checking. As a result, macros can capture many patterns of code reuse that Rust’s core abstractions cannot.

The drawback is that macro-based code can be harder to understand, because fewer of the built-in rules apply. Like an ordinary function, a well-behaved macro can be used without understanding its implementation. However, it can be difficult to design a well-behaved macro! Additionally, compiler errors in macro code are harder to interpret, because they describe problems in the expanded code, not the source-level form that developers use.

These drawbacks make macros something of a “feature of last resort”. That’s not to say that macros are bad; they are part of Rust because sometimes they’re needed for truly concise, well-abstracted code. Just keep this tradeoff in mind.

Defining a macro

You may have seen the vec! macro, used to initialize a vector with any number of elements.

let x: Vec<u32> = vec![1, 2, 3];

This can’t be an ordinary function, because it takes any number of arguments. But we can imagine it as syntactic shorthand for

let x: Vec<u32> = {
      let mut temp_vec = Vec::new();
      temp_vec.push(1);
      temp_vec.push(2);
      temp_vec.push(3);
      temp_vec
  };

We can implement this shorthand, using a macro: 4

macro_rules! vec {
      ( $( $x:expr ),* ) => {
          {
              let mut temp_vec = Vec::new();
              $(
                  temp_vec.push($x);
              )*
              temp_vec
          }
      };
  }

Whoa, that’s a lot of new syntax! Let’s break it down.

macro_rules! vec { ... }

This says we’re defining a macro named vec, much as fn vec would define a function named vec. In prose, we informally write a macro’s name with an exclamation point, e.g. vec!. The exclamation point is part of the invocation syntax and serves to distinguish a macro from an ordinary function.

Matching

The macro is defined through a series of rules, which are pattern-matching cases. Above, we had

( $( $x:expr ),* ) => { ... };

This is like a match expression arm, but the matching happens on Rust syntax trees, at compile time. The semicolon is optional on the last (here, only) case. The “pattern” on the left-hand side of => is known as a ‘matcher’. These have [their own little grammar] within the language.

The matcher $x:expr will match any Rust expression, binding that syntax tree to the ‘metavariable’ $x. The identifier expr is a ‘fragment specifier’; the full possibilities are enumerated later in this chapter. Surrounding the matcher with $(...),* will match zero or more expressions, separated by commas.

Aside from the special matcher syntax, any Rust tokens that appear in a matcher must match exactly. For example,

macro_rules! foo {
      (x => $e:expr) => (println!("mode X: {}", $e));
      (y => $e:expr) => (println!("mode Y: {}", $e));
  }
  
  fn main() {
      foo!(y => 3);
  }

will print

mode Y: 3

With

foo!(z => 3);

we get the compiler error

error: no rules expected the token `z`

Expansion

The right-hand side of a macro rule is ordinary Rust syntax, for the most part. But we can splice in bits of syntax captured by the matcher. From the original example:

$(
      temp_vec.push($x);
  )*

Each matched expression $x will produce a single push statement in the macro expansion. The repetition in the expansion proceeds in “lockstep” with repetition in the matcher (more on this in a moment).

Because $x was already declared as matching an expression, we don’t repeat :expr on the right-hand side. Also, we don’t include a separating comma as part of the repetition operator. Instead, we have a terminating semicolon within the repeated block.

Another detail: the vec! macro has two pairs of braces on the right-hand side. They are often combined like so:

macro_rules! foo {
      () => {{
          ...
      }}
  }

The outer braces are part of the syntax of macro_rules!. In fact, you can use () or [] instead. They simply delimit the right-hand side as a whole.

The inner braces are part of the expanded syntax. Remember, the vec! macro is used in an expression context. To write an expression with multiple statements, including let-bindings, we use a block. If your macro expands to a single expression, you don’t need this extra layer of braces.

Note that we never declared that the macro produces an expression. In fact, this is not determined until we use the macro as an expression. With care, you can write a macro whose expansion works in several contexts. For example, shorthand for a data type could be valid as either an expression or a pattern.

Repetition

The repetition operator follows two principal rules:

  1. $(...)* walks through one “layer” of repetitions, for all of the $names it contains, in lockstep, and
  2. each $name must be under at least as many $(...)*s as it was matched against. If it is under more, it’ll be duplicated, as appropriate.

This baroque macro illustrates the duplication of variables from outer repetition levels.

macro_rules! o_O {
      (
          $(
              $x:expr; [ $( $y:expr ),* ]
          );*
      ) => {
          &[ $($( $x + $y ),*),* ]
      }
  }
  
  fn main() {
      let a: &[i32]
          = o_O!(10; [1, 2, 3];
                 20; [4, 5, 6]);
  
      assert_eq!(a, [11, 12, 13, 24, 25, 26]);
  }

That’s most of the matcher syntax. These examples use $(...)*, which is a “zero or more” match. Alternatively you can write $(...)+ for a “one or more” match. Both forms optionally include a separator, which can be any token except + or *.

This system is based on “Macro-by-Example” (PDF link).

Hygiene

Some languages implement macros using simple text substitution, which leads to various problems. For example, this C program prints 13 instead of the expected 25.

#define FIVE_TIMES(x) 5 * x
  
  int main() {
      printf("%d\n", FIVE_TIMES(2 + 3));
      return 0;
  }

After expansion we have 5 * 2 + 3, and multiplication has greater precedence than addition. If you’ve used C macros a lot, you probably know the standard idioms for avoiding this problem, as well as five or six others. In Rust, we don’t have to worry about it.

macro_rules! five_times {
      ($x:expr) => (5 * $x);
  }
  
  fn main() {
      assert_eq!(25, five_times!(2 + 3));
  }

The metavariable $x is parsed as a single expression node, and keeps its place in the syntax tree even after substitution.

Another common problem in macro systems is ‘variable capture’. Here’s a C macro, using [a GNU C extension] to emulate Rust’s expression blocks.

#define LOG(msg) ({ \
      int state = get_log_state(); \
      if (state > 0) { \
          printf("log(%d): %s\n", state, msg); \
      } \
  })

Here’s a simple use case that goes terribly wrong:

const char *state = "reticulating splines";
  LOG(state)

This expands to

const char *state = "reticulating splines";
  int state = get_log_state();
  if (state > 0) {
      printf("log(%d): %s\n", state, state);
  }

The second variable named state shadows the first one. This is a problem because the print statement should refer to both of them.

The equivalent Rust macro has the desired behavior.

macro_rules! log {
      ($msg:expr) => {{
          let state: i32 = get_log_state();
          if state > 0 {
              println!("log({}): {}", state, $msg);
          }
      }};
  }
  
  fn main() {
      let state: &str = "reticulating splines";
      log!(state);
  }

This works because Rust has a [hygienic macro system]. Each macro expansion happens in a distinct ‘syntax context’, and each variable is tagged with the syntax context where it was introduced. It’s as though the variable state inside main is painted a different “color” from the variable state inside the macro, and therefore they don’t conflict.

This also restricts the ability of macros to introduce new bindings at the invocation site. Code such as the following will not work:

macro_rules! foo {
      () => (let x = 3);
  }
  
  fn main() {
      foo!();
      println!("{}", x);
  }

Instead you need to pass the variable name into the invocation, so it’s tagged with the right syntax context.

macro_rules! foo {
      ($v:ident) => (let $v = 3);
  }
  
  fn main() {
      foo!(x);
      println!("{}", x);
  }

This holds for let bindings and loop labels, but not for items. So the following code does compile:

macro_rules! foo {
      () => (fn x() { });
  }
  
  fn main() {
      foo!();
      x();
  }

Recursive macros

A macro’s expansion can include more macro invocations, including invocations of the very same macro being expanded. These recursive macros are useful for processing tree-structured input, as illustrated by this (simplistic) HTML shorthand:

macro_rules! write_html {
      ($w:expr, ) => (());
  
      ($w:expr, $e:tt) => (write!($w, "{}", $e));
  
      ($w:expr, $tag:ident [ $($inner:tt)* ] $($rest:tt)*) => {{
          write!($w, "<{}>", stringify!($tag));
          write_html!($w, $($inner)*);
          write!($w, "</{}>", stringify!($tag));
          write_html!($w, $($rest)*);
      }};
  }
  
  fn main() {
      use std::fmt::Write;
      let mut out = String::new();
  
      write_html!(&mut out,
          html[
              head[title["Macros guide"]]
              body[h1["Macros are the best!"]]
          ]);
  
      assert_eq!(out,
          "<html><head><title>Macros guide</title></head>\
           <body><h1>Macros are the best!</h1></body></html>");
  }

Debugging macro code

To see the results of expanding macros, run rustc --pretty expanded. The output represents a whole crate, so you can also feed it back in to rustc, which will sometimes produce better error messages than the original compilation. Note that the --pretty expanded output may have a different meaning if multiple variables of the same name (but different syntax contexts) are in play in the same scope. In this case --pretty expanded,hygiene will tell you about the syntax contexts.

rustc provides two syntax extensions that help with macro debugging. For now, they are unstable and require feature gates.

  • log_syntax!(...) will print its arguments to standard output, at compile time, and “expand” to nothing.

  • trace_macros!(true) will enable a compiler message every time a macro is expanded. Use trace_macros!(false) later in expansion to turn it off.

Syntactic requirements

Even when Rust code contains un-expanded macros, it can be parsed as a full syntax tree. This property can be very useful for editors and other tools that process code. It also has a few consequences for the design of Rust’s macro system.

One consequence is that Rust must determine, when it parses a macro invocation, whether the macro stands in for

  • zero or more items,
  • zero or more methods,
  • an expression,
  • a statement, or
  • a pattern.

A macro invocation within a block could stand for some items, or for an expression / statement. Rust uses a simple rule to resolve this ambiguity. A macro invocation that stands for items must be either

  • delimited by curly braces, e.g. foo! { ... }, or
  • terminated by a semicolon, e.g. foo!(...);

Another consequence of pre-expansion parsing is that the macro invocation must consist of valid Rust tokens. Furthermore, parentheses, brackets, and braces must be balanced within a macro invocation. For example, foo!([) is forbidden. This allows Rust to know where the macro invocation ends.

More formally, the macro invocation body must be a sequence of ‘token trees’. A token tree is defined recursively as either

  • a sequence of token trees surrounded by matching (), [], or {}, or
  • any other single token.

Within a matcher, each metavariable has a ‘fragment specifier’, identifying which syntactic form it matches.

  • ident: an identifier. Examples: x; foo.
  • path: a qualified name. Example: T::SpecialA.
  • expr: an expression. Examples: 2 + 2; if true then { 1 } else { 2 }; f(42).
  • ty: a type. Examples: i32; Vec<(char, String)>; &T.
  • pat: a pattern. Examples: Some(t); (17, 'a'); _.
  • stmt: a single statement. Example: let x = 3.
  • block: a brace-delimited sequence of statements. Example: { log(error, "hi"); return 12; }.
  • item: an item. Examples: fn foo() { }; struct Bar;.
  • meta: a “meta item”, as found in attributes. Example: cfg(target_os = "windows").
  • tt: a single token tree.

There are additional rules regarding the next token after a metavariable:

  • expr variables may only be followed by one of: => , ;
  • ty and path variables may only be followed by one of: => , : = > as
  • pat variables may only be followed by one of: => , = if in
  • Other variables may be followed by any token.

These rules provide some flexibility for Rust’s syntax to evolve without breaking existing macros.

The macro system does not deal with parse ambiguity at all. For example, the grammar $($t:ty)* $e:expr will always fail to parse, because the parser would be forced to choose between parsing $t and parsing $e. Changing the invocation syntax to put a distinctive token in front can solve the problem. In this case, you can write $(T $t:ty)* E $e:exp.

Scoping and macro import/export

Macros are expanded at an early stage in compilation, before name resolution. One downside is that scoping works differently for macros, compared to other constructs in the language.

Definition and expansion of macros both happen in a single depth-first, lexical-order traversal of a crate’s source. So a macro defined at module scope is visible to any subsequent code in the same module, which includes the body of any subsequent child mod items.

A macro defined within the body of a single fn, or anywhere else not at module scope, is visible only within that item.

If a module has the macro_use attribute, its macros are also visible in its parent module after the child’s mod item. If the parent also has macro_use then the macros will be visible in the grandparent after the parent’s mod item, and so forth.

The macro_use attribute can also appear on extern crate. In this context it controls which macros are loaded from the external crate, e.g.

#[macro_use(foo, bar)]
  extern crate baz;

If the attribute is given simply as #[macro_use], all macros are loaded. If there is no #[macro_use] attribute then no macros are loaded. Only macros defined with the #[macro_export] attribute may be loaded.

To load a crate’s macros without linking it into the output, use #[no_link] as well.

An example:

macro_rules! m1 { () => (()) }
  
  // visible here: m1
  
  mod foo {
      // visible here: m1
  
      #[macro_export]
      macro_rules! m2 { () => (()) }
  
      // visible here: m1, m2
  }
  
  // visible here: m1
  
  macro_rules! m3 { () => (()) }
  
  // visible here: m1, m3
  
  #[macro_use]
  mod bar {
      // visible here: m1, m3
  
      macro_rules! m4 { () => (()) }
  
      // visible here: m1, m3, m4
  }
  
  // visible here: m1, m3, m4

When this library is loaded with #[macro_use] extern crate, only m2 will be imported.

The Rust Reference has a listing of macro-related attributes.

The variable $crate

A further difficulty occurs when a macro is used in multiple crates. Say that mylib defines

pub fn increment(x: u32) -> u32 {
      x + 1
  }
  
  #[macro_export]
  macro_rules! inc_a {
      ($x:expr) => ( ::increment($x) )
  }
  
  #[macro_export]
  macro_rules! inc_b {
      ($x:expr) => ( ::mylib::increment($x) )
  }

inc_a only works within mylib, while inc_b only works outside the library. Furthermore, inc_b will break if the user imports mylib under another name.

Rust does not (yet) have a hygiene system for crate references, but it does provide a simple workaround for this problem. Within a macro imported from a crate named foo, the special macro variable $crate will expand to ::foo. By contrast, when a macro is defined and then used in the same crate, $crate will expand to nothing. This means we can write

#[macro_export]
  macro_rules! inc {
      ($x:expr) => ( $crate::increment($x) )
  }

to define a single macro that works both inside and outside our library. The function name will expand to either ::increment or ::mylib::increment.

To keep this system simple and correct, #[macro_use] extern crate ... may only appear at the root of your crate, not inside mod. This ensures that $crate is a single identifier.

The deep end

The introductory chapter mentioned recursive macros, but it did not give the full story. Recursive macros are useful for another reason: Each recursive invocation gives you another opportunity to pattern-match the macro’s arguments.

As an extreme example, it is possible, though hardly advisable, to implement the Bitwise Cyclic Tag automaton within Rust’s macro system.

macro_rules! bct {
      // cmd 0:  d ... => ...
      (0, $($ps:tt),* ; $_d:tt)
          => (bct!($($ps),*, 0 ; ));
      (0, $($ps:tt),* ; $_d:tt, $($ds:tt),*)
          => (bct!($($ps),*, 0 ; $($ds),*));
  
      // cmd 1p:  1 ... => 1 ... p
      (1, $p:tt, $($ps:tt),* ; 1)
          => (bct!($($ps),*, 1, $p ; 1, $p));
      (1, $p:tt, $($ps:tt),* ; 1, $($ds:tt),*)
          => (bct!($($ps),*, 1, $p ; 1, $($ds),*, $p));
  
      // cmd 1p:  0 ... => 0 ...
      (1, $p:tt, $($ps:tt),* ; $($ds:tt),*)
          => (bct!($($ps),*, 1, $p ; $($ds),*));
  
      // halt on empty data string
      ( $($ps:tt),* ; )
          => (());
  }

Exercise: use macros to reduce duplication in the above definition of the bct! macro.

Common macros

Here are some common macros you’ll see in Rust code.

panic!

This macro causes the current thread to panic. You can give it a message to panic with:

panic!("oh no!");

vec!

The vec! macro is used throughout the book, so you’ve probably seen it already. It creates Vec<T>s with ease:

let v = vec![1, 2, 3, 4, 5];

It also lets you make vectors with repeating values. For example, a hundred zeroes:

let v = vec![0; 100];

assert! and assert_eq!

These two macros are used in tests. assert! takes a boolean. assert_eq! takes two values and checks them for equality. true passes, false panic!s. Like this:

// A-ok!
  
  assert!(true);
  assert_eq!(5, 3 + 2);
  
  // nope :(
  
  assert!(5 < 3);
  assert_eq!(5, 3);

try!

try! is used for error handling. It takes something that can return a Result<T, E>, and gives T if it’s a Ok<T>, and returns with the Err(E) if it’s that. Like this:

use std::fs::File;
  
  fn foo() -> std::io::Result<()> {
      let f = try!(File::create("foo.txt"));
  
      Ok(())
  }

This is cleaner than doing this:

use std::fs::File;
  
  fn foo() -> std::io::Result<()> {
      let f = File::create("foo.txt");
  
      let f = match f {
          Ok(t) => t,
          Err(e) => return Err(e),
      };
  
      Ok(())
  }

unreachable!

This macro is used when you think some code should never execute:

if false {
      unreachable!();
  }

Sometimes, the compiler may make you have a different branch that you know will never, ever run. In these cases, use this macro, so that if you end up wrong, you’ll get a panic! about it.

let x: Option<i32> = None;
  
  match x {
      Some(_) => unreachable!(),
      None => println!("I know x is None!"),
  }

unimplemented!

The unimplemented! macro can be used when you’re trying to get your functions to typecheck, and don’t want to worry about writing out the body of the function. One example of this situation is implementing a trait with multiple required methods, where you want to tackle one at a time. Define the others as unimplemented! until you’re ready to write them.

Procedural macros

If Rust’s macro system can’t do what you need, you may want to write a compiler plugin instead. Compared to macro_rules! macros, this is significantly more work, the interfaces are much less stable, and bugs can be much harder to track down. In exchange you get the flexibility of running arbitrary Rust code within the compiler. Syntax extension plugins are sometimes called ‘procedural macros’ for this reason.

Raw Pointers

Rust has a number of different smart pointer types in its standard library, but there are two types that are extra-special. Much of Rust’s safety comes from compile-time checks, but raw pointers don’t have such guarantees, and are unsafe to use.

*const T and *mut T are called ‘raw pointers’ in Rust. Sometimes, when writing certain kinds of libraries, you’ll need to get around Rust’s safety guarantees for some reason. In this case, you can use raw pointers to implement your library, while exposing a safe interface for your users. For example, * pointers are allowed to alias, allowing them to be used to write shared-ownership types, and even thread-safe shared memory types (the Rc<T> and Arc<T> types are both implemented entirely in Rust).

Here are some things to remember about raw pointers that are different than other pointer types. They:

  • are not guaranteed to point to valid memory and are not even guaranteed to be non-null (unlike both Box and &);
  • do not have any automatic clean-up, unlike Box, and so require manual resource management;
  • are plain-old-data, that is, they don’t move ownership, again unlike Box, hence the Rust compiler cannot protect against bugs like use-after-free;
  • lack any form of lifetimes, unlike &, and so the compiler cannot reason about dangling pointers; and
  • have no guarantees about aliasing or mutability other than mutation not being allowed directly through a *const T.

Basics

Creating a raw pointer is perfectly safe:

let x = 5;
  let raw = &x as *const i32;
  
  let mut y = 10;
  let raw_mut = &mut y as *mut i32;

However, dereferencing one is not. This won’t work:

let x = 5;
  let raw = &x as *const i32;
  
  println!("raw points at {}", *raw);

It gives this error:

error: dereference of raw pointer requires unsafe function or block [E0133]
       println!("raw points at {}", *raw);
                                    ^~~~

When you dereference a raw pointer, you’re taking responsibility that it’s not pointing somewhere that would be incorrect. As such, you need unsafe:

let x = 5;
  let raw = &x as *const i32;
  
  let points_at = unsafe { *raw };
  
  println!("raw points at {}", points_at);

For more operations on raw pointers, see their API documentation.

FFI

Raw pointers are useful for FFI: Rust’s *const T and *mut T are similar to C’s const T* and T*, respectively. For more about this use, consult the FFI chapter.

References and raw pointers

At runtime, a raw pointer * and a reference pointing to the same piece of data have an identical representation. In fact, an &T reference will implicitly coerce to an *const T raw pointer in safe code and similarly for the mut variants (both coercions can be performed explicitly with, respectively, value as *const T and value as *mut T).

Going the opposite direction, from *const to a reference &, is not safe. A &T is always valid, and so, at a minimum, the raw pointer *const T has to point to a valid instance of type T. Furthermore, the resulting pointer must satisfy the aliasing and mutability laws of references. The compiler assumes these properties are true for any references, no matter how they are created, and so any conversion from raw pointers is asserting that they hold. The programmer must guarantee this.

The recommended method for the conversion is

let i: u32 = 1;
  
  // explicit cast
  let p_imm: *const u32 = &i as *const u32;
  let mut m: u32 = 2;
  
  // implicit coercion
  let p_mut: *mut u32 = &mut m;
  
  unsafe {
      let ref_imm: &u32 = &*p_imm;
      let ref_mut: &mut u32 = &mut *p_mut;
  }

The &*x dereferencing style is preferred to using a transmute. The latter is far more powerful than necessary, and the more restricted operation is harder to use incorrectly; for example, it requires that x is a pointer (unlike transmute).

unsafe

Rust’s main draw is its powerful static guarantees about behavior. But safety checks are conservative by nature: there are some programs that are actually safe, but the compiler is not able to verify this is true. To write these kinds of programs, we need to tell the compiler to relax its restrictions a bit. For this, Rust has a keyword, unsafe. Code using unsafe has less restrictions than normal code does.

Let’s go over the syntax, and then we’ll talk semantics. unsafe is used in four contexts. The first one is to mark a function as unsafe:

unsafe fn danger_will_robinson() {
      // scary stuff
  }

All functions called from FFI must be marked as unsafe, for example. The second use of unsafe is an unsafe block:

unsafe {
      // scary stuff
  }

The third is for unsafe traits:

unsafe trait Scary { }

And the fourth is for implementing one of those traits:

unsafe impl Scary for i32 {}

It’s important to be able to explicitly delineate code that may have bugs that cause big problems. If a Rust program segfaults, you can be sure it’s somewhere in the sections marked unsafe.

What does ‘safe’ mean?

Safe, in the context of Rust, means ‘doesn’t do anything unsafe’. It’s also important to know that there are certain behaviors that are probably not desirable in your code, but are expressly not unsafe:

  • Deadlocks
  • Leaks of memory or other resources
  • Exiting without calling destructors
  • Integer overflow

Rust cannot prevent all kinds of software problems. Buggy code can and will be written in Rust. These things aren’t great, but they don’t qualify as unsafe specifically.

In addition, the following are all undefined behaviors in Rust, and must be avoided, even when writing unsafe code:

  • Data races
  • Dereferencing a null/dangling raw pointer
  • Reads of undef (uninitialized) memory
  • Breaking the pointer aliasing rules with raw pointers.
  • &mut T and &T follow LLVM’s scoped noalias model, except if the &T contains an UnsafeCell<U>. Unsafe code must not violate these aliasing guarantees.
  • Mutating an immutable value/reference without UnsafeCell<U>
  • Invoking undefined behavior via compiler intrinsics:
  • Indexing outside of the bounds of an object with std::ptr::offset (offset intrinsic), with the exception of one byte past the end which is permitted.
  • Using std::ptr::copy_nonoverlapping_memory (memcpy32/memcpy64 intrinsics) on overlapping buffers
  • Invalid values in primitive types, even in private fields/locals:
  • Null/dangling references or boxes
  • A value other than false (0) or true (1) in a bool
  • A discriminant in an enum not included in its type definition
  • A value in a char which is a surrogate or above char::MAX
  • Non-UTF-8 byte sequences in a str
  • Unwinding into Rust from foreign code or unwinding from Rust into foreign code.

Unsafe Superpowers

In both unsafe functions and unsafe blocks, Rust will let you do three things that you normally can not do. Just three. Here they are:

  1. Access or update a static mutable variable.
  2. Dereference a raw pointer.
  3. Call unsafe functions. This is the most powerful ability.

That’s it. It’s important that unsafe does not, for example, ‘turn off the borrow checker’. Adding unsafe to some random Rust code doesn’t change its semantics, it won’t just start accepting anything. But it will let you write things that do break some of the rules.

You will also encounter the unsafe keyword when writing bindings to foreign (non-Rust) interfaces. You’re encouraged to write a safe, native Rust interface around the methods provided by the library.

Let’s go over the basic three abilities listed, in order.

Access or update a static mut

Rust has a feature called ‘static mut’ which allows for mutable global state. Doing so can cause a data race, and as such is inherently not safe. For more details, see the static section of the book.

Dereference a raw pointer

Raw pointers let you do arbitrary pointer arithmetic, and can cause a number of different memory safety and security issues. In some senses, the ability to dereference an arbitrary pointer is one of the most dangerous things you can do. For more on raw pointers, see their section of the book.

Call unsafe functions

This last ability works with both aspects of unsafe: you can only call functions marked unsafe from inside an unsafe block.

This ability is powerful and varied. Rust exposes some compiler intrinsics as unsafe functions, and some unsafe functions bypass safety checks, trading safety for speed.

I’ll repeat again: even though you can do arbitrary things in unsafe blocks and functions doesn’t mean you should. The compiler will act as though you’re upholding its invariants, so be careful!

Nightly Rust

Rust provides three distribution channels for Rust: nightly, beta, and stable. Unstable features are only available on nightly Rust. For more details on this process, see ‘Stability as a deliverable’.

To install nightly Rust, you can use rustup.sh:

$ curl -s https://static.rust-lang.org/rustup.sh | sh -s -- --channel=nightly

If you’re concerned about the potential insecurity of using curl | sh, please keep reading and see our disclaimer below. And feel free to use a two-step version of the installation and examine our installation script:

$ curl -f -L https://static.rust-lang.org/rustup.sh -O
  $ sh rustup.sh --channel=nightly

If you’re on Windows, please download either the 32-bit installer or the 64-bit installer and run it.

Uninstalling

If you decide you don’t want Rust anymore, we’ll be a bit sad, but that’s okay. Not every programming language is great for everyone. Just run the uninstall script:

$ sudo /usr/local/lib/rustlib/uninstall.sh

If you used the Windows installer, just re-run the .msi and it will give you an uninstall option.

Some people, and somewhat rightfully so, get very upset when we tell you to curl | sh. Basically, when you do this, you are trusting that the good people who maintain Rust aren’t going to hack your computer and do bad things. That’s a good instinct! If you’re one of those people, please check out the documentation on building Rust from Source, or the official binary downloads.

Oh, we should also mention the officially supported platforms:

  • Windows (7, 8, Server 2008 R2)
  • Linux (2.6.18 or later, various distributions), x86 and x86-64
  • OSX 10.7 (Lion) or greater, x86 and x86-64

We extensively test Rust on these platforms, and a few others, too, like Android. But these are the ones most likely to work, as they have the most testing.

Finally, a comment about Windows. Rust considers Windows to be a first-class platform upon release, but if we’re honest, the Windows experience isn’t as integrated as the Linux/OS X experience is. We’re working on it! If anything does not work, it is a bug. Please let us know if that happens. Each and every commit is tested against Windows just like any other platform.

If you’ve got Rust installed, you can open up a shell, and type this:

$ rustc --version

You should see the version number, commit hash, commit date and build date:

rustc 1.0.0-nightly (f11f3e7ba 2015-01-04) (built 2015-01-06)

If you did, Rust has been installed successfully! Congrats!

This installer also installs a copy of the documentation locally, so you can read it offline. On UNIX systems, /usr/local/share/doc/rust is the location. On Windows, it’s in a share/doc directory, inside wherever you installed Rust to.

If not, there are a number of places where you can get help. The easiest is the #rust IRC channel on irc.mozilla.org, which you can access through Mibbit. Click that link, and you’ll be chatting with other Rustaceans (a silly nickname we call ourselves), and we can help you out. Other great resources include the user’s forum, and Stack Overflow.

Compiler Plugins

Introduction

rustc can load compiler plugins, which are user-provided libraries that extend the compiler’s behavior with new syntax extensions, lint checks, etc.

A plugin is a dynamic library crate with a designated registrar function that registers extensions with rustc. Other crates can load these extensions using the crate attribute #![plugin(...)]. See the rustc::plugin documentation for more about the mechanics of defining and loading a plugin.

If present, arguments passed as #![plugin(foo(... args ...))] are not interpreted by rustc itself. They are provided to the plugin through the Registry’s args method.

In the vast majority of cases, a plugin should only be used through #![plugin] and not through an extern crate item. Linking a plugin would pull in all of libsyntax and librustc as dependencies of your crate. This is generally unwanted unless you are building another plugin. The plugin_as_library lint checks these guidelines.

The usual practice is to put compiler plugins in their own crate, separate from any macro_rules! macros or ordinary Rust code meant to be used by consumers of a library.

Syntax extensions

Plugins can extend Rust’s syntax in various ways. One kind of syntax extension is the procedural macro. These are invoked the same way as ordinary macros, but the expansion is performed by arbitrary Rust code that manipulates syntax trees at compile time.

Let’s write a plugin roman_numerals.rs that implements Roman numeral integer literals.

#![crate_type="dylib"]
  #![feature(plugin_registrar, rustc_private)]
  
  extern crate syntax;
  extern crate rustc;
  
  use syntax::codemap::Span;
  use syntax::parse::token;
  use syntax::ast::{TokenTree, TtToken};
  use syntax::ext::base::{ExtCtxt, MacResult, DummyResult, MacEager};
  use syntax::ext::build::AstBuilder;  // trait for expr_usize
  use rustc::plugin::Registry;
  
  fn expand_rn(cx: &mut ExtCtxt, sp: Span, args: &[TokenTree])
          -> Box<MacResult + 'static> {
  
      static NUMERALS: &'static [(&'static str, u32)] = &[
          ("M", 1000), ("CM", 900), ("D", 500), ("CD", 400),
          ("C",  100), ("XC",  90), ("L",  50), ("XL",  40),
          ("X",   10), ("IX",   9), ("V",   5), ("IV",   4),
          ("I",    1)];
  
      let text = match args {
          [TtToken(_, token::Ident(s, _))] => s.to_string(),
          _ => {
              cx.span_err(sp, "argument should be a single identifier");
              return DummyResult::any(sp);
          }
      };
  
      let mut text = &*text;
      let mut total = 0;
      while !text.is_empty() {
          match NUMERALS.iter().find(|&&(rn, _)| text.starts_with(rn)) {
              Some(&(rn, val)) => {
                  total += val;
                  text = &text[rn.len()..];
              }
              None => {
                  cx.span_err(sp, "invalid Roman numeral");
                  return DummyResult::any(sp);
              }
          }
      }
  
      MacEager::expr(cx.expr_u32(sp, total))
  }
  
  #[plugin_registrar]
  pub fn plugin_registrar(reg: &mut Registry) {
      reg.register_macro("rn", expand_rn);
  }

Then we can use rn!() like any other macro:

#![feature(plugin)]
  #![plugin(roman_numerals)]
  
  fn main() {
      assert_eq!(rn!(MMXV), 2015);
  }

The advantages over a simple fn(&str) -> u32 are:

  • The (arbitrarily complex) conversion is done at compile time.
  • Input validation is also performed at compile time.
  • It can be extended to allow use in patterns, which effectively gives a way to define new literal syntax for any data type.

In addition to procedural macros, you can define new derive-like attributes and other kinds of extensions. See Registry::register_syntax_extension and the SyntaxExtension enum. For a more involved macro example, see regex_macros.

Tips and tricks

Some of the macro debugging tips are applicable.

You can use syntax::parse to turn token trees into higher-level syntax elements like expressions:

fn expand_foo(cx: &mut ExtCtxt, sp: Span, args: &[TokenTree])
          -> Box<MacResult+'static> {
  
      let mut parser = cx.new_parser_from_tts(args);
  
      let expr: P<Expr> = parser.parse_expr();

Looking through libsyntax parser code will give you a feel for how the parsing infrastructure works.

Keep the Spans of everything you parse, for better error reporting. You can wrap Spanned around your custom data structures.

Calling ExtCtxt::span_fatal will immediately abort compilation. It’s better to instead call ExtCtxt::span_err and return DummyResult, so that the compiler can continue and find further errors.

To print syntax fragments for debugging, you can use span_note together with syntax::print::pprust::*_to_string.

The example above produced an integer literal using AstBuilder::expr_usize. As an alternative to the AstBuilder trait, libsyntax provides a set of quasiquote macros. They are undocumented and very rough around the edges. However, the implementation may be a good starting point for an improved quasiquote as an ordinary plugin library.

Lint plugins

Plugins can extend Rust’s lint infrastructure with additional checks for code style, safety, etc. You can see src/test/auxiliary/lint_plugin_test.rs for a full example, the core of which is reproduced here:

declare_lint!(TEST_LINT, Warn,
                "Warn about items named 'lintme'");
  
  struct Pass;
  
  impl LintPass for Pass {
      fn get_lints(&self) -> LintArray {
          lint_array!(TEST_LINT)
      }
  
      fn check_item(&mut self, cx: &Context, it: &ast::Item) {
          if it.ident.name == "lintme" {
              cx.span_lint(TEST_LINT, it.span, "item is named 'lintme'");
          }
      }
  }
  
  #[plugin_registrar]
  pub fn plugin_registrar(reg: &mut Registry) {
      reg.register_lint_pass(box Pass as LintPassObject);
  }

Then code like

#![plugin(lint_plugin_test)]
  
  fn lintme() { }

will produce a compiler warning:

foo.rs:4:1: 4:16 warning: item is named 'lintme', #[warn(test_lint)] on by default
  foo.rs:4 fn lintme() { }
           ^~~~~~~~~~~~~~~

The components of a lint plugin are:

  • one or more declare_lint! invocations, which define static Lint structs;

  • a struct holding any state needed by the lint pass (here, none);

  • a LintPass implementation defining how to check each syntax element. A single LintPass may call span_lint for several different Lints, but should register them all through the get_lints method.

Lint passes are syntax traversals, but they run at a late stage of compilation where type information is available. rustc’s built-in lints mostly use the same infrastructure as lint plugins, and provide examples of how to access type information.

Lints defined by plugins are controlled by the usual attributes and compiler flags, e.g. #[allow(test_lint)] or -A test-lint. These identifiers are derived from the first argument to declare_lint!, with appropriate case and punctuation conversion.

You can run rustc -W help foo.rs to see a list of lints known to rustc, including those provided by plugins loaded by foo.rs.

Inline Assembly

For extremely low-level manipulations and performance reasons, one might wish to control the CPU directly. Rust supports using inline assembly to do this via the asm! macro. The syntax roughly matches that of GCC & Clang:

asm!(assembly template
     : output operands
     : input operands
     : clobbers
     : options
     );

Any use of asm is feature gated (requires #![feature(asm)] on the crate to allow) and of course requires an unsafe block.

Note: the examples here are given in x86/x86-64 assembly, but all platforms are supported.

Assembly template

The assembly template is the only required parameter and must be a literal string (i.e. "")

#![feature(asm)]
  
  #[cfg(any(target_arch = "x86", target_arch = "x86_64"))]
  fn foo() {
      unsafe {
          asm!("NOP");
      }
  }
  
  // other platforms
  #[cfg(not(any(target_arch = "x86", target_arch = "x86_64")))]
  fn foo() { /* ... */ }
  
  fn main() {
      // ...
      foo();
      // ...
  }

(The feature(asm) and #[cfg]s are omitted from now on.)

Output operands, input operands, clobbers and options are all optional but you must add the right number of : if you skip them:

asm!("xor %eax, %eax"
      :
      :
      : "{eax}"
     );

Whitespace also doesn’t matter:

asm!("xor %eax, %eax" ::: "{eax}");

Operands

Input and output operands follow the same format: : "constraints1"(expr1), "constraints2"(expr2), ...". Output operand expressions must be mutable lvalues, or not yet assigned:

fn add(a: i32, b: i32) -> i32 {
      let c: i32;
      unsafe {
          asm!("add $2, $0"
               : "=r"(c)
               : "0"(a), "r"(b)
               );
      }
      c
  }
  
  fn main() {
      assert_eq!(add(3, 14159), 14162)
  }

If you would like to use real operands in this position, however, you are required to put curly braces {} around the register that you want, and you are required to put the specific size of the operand. This is useful for very low level programming, where which register you use is important:

let result: u8;
  asm!("in %dx, %al" : "={al}"(result) : "{dx}"(port));
  result

Clobbers

Some instructions modify registers which might otherwise have held different values so we use the clobbers list to indicate to the compiler not to assume any values loaded into those registers will stay valid.

// Put the value 0x200 in eax
  asm!("mov $$0x200, %eax" : /* no outputs */ : /* no inputs */ : "{eax}");

Input and output registers need not be listed since that information is already communicated by the given constraints. Otherwise, any other registers used either implicitly or explicitly should be listed.

If the assembly changes the condition code register cc should be specified as one of the clobbers. Similarly, if the assembly modifies memory, memory should also be specified.

Options

The last section, options is specific to Rust. The format is comma separated literal strings (i.e. :"foo", "bar", "baz"). It’s used to specify some extra info about the inline assembly:

Current valid options are:

  1. volatile - specifying this is analogous to __asm__ __volatile__ (...) in gcc/clang.
  2. alignstack - certain instructions expect the stack to be aligned a certain way (i.e. SSE) and specifying this indicates to the compiler to insert its usual stack alignment code
  3. intel - use intel syntax instead of the default AT&T.
let result: i32;
  unsafe {
     asm!("mov eax, 2" : "={eax}"(result) : : : "intel")
  }
  println!("eax is currently {}", result);

More Information

The current implementation of the asm! macro is a direct binding to LLVM’s inline assembler expressions, so be sure to check out their documentation as well for more information about clobbers, constraints, etc.

No stdlib

By default, std is linked to every Rust crate. In some contexts, this is undesirable, and can be avoided with the #![no_std] attribute attached to the crate.

// a minimal library
  #![crate_type="lib"]
  #![feature(no_std)]
  #![no_std]
  # // fn main() {} tricked you, rustdoc!

Obviously there’s more to life than just libraries: one can use #[no_std] with an executable, controlling the entry point is possible in two ways: the #[start] attribute, or overriding the default shim for the C main function with your own.

The function marked #[start] is passed the command line parameters in the same format as C:

#![feature(lang_items, start, no_std, libc)]
  #![no_std]
  
  // Pull in the system libc library for what crt0.o likely requires
  extern crate libc;
  
  // Entry point for this program
  #[start]
  fn start(_argc: isize, _argv: *const *const u8) -> isize {
      0
  }
  
  // These functions and traits are used by the compiler, but not
  // for a bare-bones hello world. These are normally
  // provided by libstd.
  #[lang = "eh_personality"] extern fn eh_personality() {}
  #[lang = "panic_fmt"] fn panic_fmt() -> ! { loop {} }

To override the compiler-inserted main shim, one has to disable it with #![no_main] and then create the appropriate symbol with the correct ABI and the correct name, which requires overriding the compiler’s name mangling too:

#![feature(no_std)]
  #![no_std]
  #![no_main]
  #![feature(lang_items, start)]
  
  extern crate libc;
  
  #[no_mangle] // ensure that this symbol is called `main` in the output
  pub extern fn main(argc: i32, argv: *const *const u8) -> i32 {
      0
  }
  
  #[lang = "eh_personality"] extern fn eh_personality() {}
  #[lang = "panic_fmt"] fn panic_fmt() -> ! { loop {} }
  # #[lang = "eh_unwind_resume"] extern fn rust_eh_unwind_resume() {}
  # // fn main() {} tricked you, rustdoc!

The compiler currently makes a few assumptions about symbols which are available in the executable to call. Normally these functions are provided by the standard library, but without it you must define your own.

The first of these two functions, eh_personality, is used by the failure mechanisms of the compiler. This is often mapped to GCC’s personality function (see the libstd implementation for more information), but crates which do not trigger a panic can be assured that this function is never called. The second function, panic_fmt, is also used by the failure mechanisms of the compiler.

Using libcore

Note: the core library’s structure is unstable, and it is recommended to use the standard library instead wherever possible.

With the above techniques, we’ve got a bare-metal executable running some Rust code. There is a good deal of functionality provided by the standard library, however, that is necessary to be productive in Rust. If the standard library is not sufficient, then libcore is designed to be used instead.

The core library has very few dependencies and is much more portable than the standard library itself. Additionally, the core library has most of the necessary functionality for writing idiomatic and effective Rust code.

As an example, here is a program that will calculate the dot product of two vectors provided from C, using idiomatic Rust practices.

#![feature(lang_items, start, no_std, core, libc)]
  #![no_std]
  
  # extern crate libc;
  extern crate core;
  
  use core::prelude::*;
  
  use core::mem;
  
  #[no_mangle]
  pub extern fn dot_product(a: *const u32, a_len: u32,
                            b: *const u32, b_len: u32) -> u32 {
      use core::raw::Slice;
  
      // Convert the provided arrays into Rust slices.
      // The core::raw module guarantees that the Slice
      // structure has the same memory layout as a &[T]
      // slice.
      //
      // This is an unsafe operation because the compiler
      // cannot tell the pointers are valid.
      let (a_slice, b_slice): (&[u32], &[u32]) = unsafe {
          mem::transmute((
              Slice { data: a, len: a_len as usize },
              Slice { data: b, len: b_len as usize },
          ))
      };
  
      // Iterate over the slices, collecting the result
      let mut ret = 0;
      for (i, j) in a_slice.iter().zip(b_slice.iter()) {
          ret += (*i) * (*j);
      }
      return ret;
  }
  
  #[lang = "panic_fmt"]
  extern fn panic_fmt(args: &core::fmt::Arguments,
                      file: &str,
                      line: u32) -> ! {
      loop {}
  }
  
  #[lang = "eh_personality"] extern fn eh_personality() {}
  # #[lang = "eh_unwind_resume"] extern fn rust_eh_unwind_resume() {}
  # #[start] fn start(argc: isize, argv: *const *const u8) -> isize { 0 }
  # fn main() {}

Note that there is one extra lang item here which differs from the examples above, panic_fmt. This must be defined by consumers of libcore because the core library declares panics, but it does not define it. The panic_fmt lang item is this crate’s definition of panic, and it must be guaranteed to never return.

As can be seen in this example, the core library is intended to provide the power of Rust in all circumstances, regardless of platform requirements. Further libraries, such as liballoc, add functionality to libcore which make other platform-specific assumptions, but continue to be more portable than the standard library itself.

Intrinsics

Note: intrinsics will forever have an unstable interface, it is recommended to use the stable interfaces of libcore rather than intrinsics directly.

These are imported as if they were FFI functions, with the special rust-intrinsic ABI. For example, if one was in a freestanding context, but wished to be able to transmute between types, and perform efficient pointer arithmetic, one would import those functions via a declaration like

#![feature(intrinsics)]
  
  extern "rust-intrinsic" {
      fn transmute<T, U>(x: T) -> U;
  
      fn offset<T>(dst: *const T, offset: isize) -> *const T;
  }

As with any other FFI functions, these are always unsafe to call.

Lang items

Note: lang items are often provided by crates in the Rust distribution, and lang items themselves have an unstable interface. It is recommended to use officially distributed crates instead of defining your own lang items.

The rustc compiler has certain pluggable operations, that is, functionality that isn’t hard-coded into the language, but is implemented in libraries, with a special marker to tell the compiler it exists. The marker is the attribute #[lang = "..."] and there are various different values of ..., i.e. various different ‘lang items’.

For example, Box pointers require two lang items, one for allocation and one for deallocation. A freestanding program that uses the Box sugar for dynamic allocations via malloc and free:

#![feature(lang_items, box_syntax, start, no_std, libc)]
  #![no_std]
  
  extern crate libc;
  
  extern {
      fn abort() -> !;
  }
  
  #[lang = "owned_box"]
  pub struct Box<T>(*mut T);
  
  #[lang = "exchange_malloc"]
  unsafe fn allocate(size: usize, _align: usize) -> *mut u8 {
      let p = libc::malloc(size as libc::size_t) as *mut u8;
  
      // malloc failed
      if p as usize == 0 {
          abort();
      }
  
      p
  }
  #[lang = "exchange_free"]
  unsafe fn deallocate(ptr: *mut u8, _size: usize, _align: usize) {
      libc::free(ptr as *mut libc::c_void)
  }
  
  #[start]
  fn main(argc: isize, argv: *const *const u8) -> isize {
      let x = box 1;
  
      0
  }
  
  #[lang = "eh_personality"] extern fn eh_personality() {}
  #[lang = "panic_fmt"] fn panic_fmt() -> ! { loop {} }

Note the use of abort: the exchange_malloc lang item is assumed to return a valid pointer, and so needs to do the check internally.

Other features provided by lang items include:

  • overloadable operators via traits: the traits corresponding to the ==, <, dereferencing (*) and + (etc.) operators are all marked with lang items; those specific four are eq, ord, deref, and add respectively.
  • stack unwinding and general failure; the eh_personality, fail and fail_bounds_checks lang items.
  • the traits in std::marker used to indicate types of various kinds; lang items send, sync and copy.
  • the marker types and variance indicators found in std::marker; lang items covariant_type, contravariant_lifetime, etc.

Lang items are loaded lazily by the compiler; e.g. if one never uses Box then there is no need to define functions for exchange_malloc and exchange_free. rustc will emit an error when an item is needed but not found in the current crate or any that it depends on.

Advanced linking

The common cases of linking with Rust have been covered earlier in this book, but supporting the range of linking possibilities made available by other languages is important for Rust to achieve seamless interaction with native libraries.

Static linking

Static linking refers to the process of creating output that contains all required libraries and so doesn’t need libraries installed on every system where you want to use your compiled project. Pure-Rust dependencies are statically linked by default so you can use created binaries and libraries without installing Rust everywhere. By contrast, native libraries (e.g. libc and libm) are usually dynamically linked, but it is possible to change this and statically link them as well.

Linking is a very platform-dependent topic, and static linking may not even be possible on some platforms! This section assumes some basic familiarity with linking on your platform of choice.

Linux

By default, all Rust programs on Linux will link to the system libc along with a number of other libraries. Let’s look at an example on a 64-bit Linux machine with GCC and glibc (by far the most common libc on Linux):

$ cat example.rs
  fn main() {}
  $ rustc example.rs
  $ ldd example
          linux-vdso.so.1 =>  (0x00007ffd565fd000)
          libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fa81889c000)
          libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fa81867e000)
          librt.so.1 => /lib/x86_64-linux-gnu/librt.so.1 (0x00007fa818475000)
          libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fa81825f000)
          libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fa817e9a000)
          /lib64/ld-linux-x86-64.so.2 (0x00007fa818cf9000)
          libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fa817b93000)

Dynamic linking on Linux can be undesirable if you wish to use new library features on old systems or target systems which do not have the required dependencies for your program to run.

Static linking is supported via an alternative libc, musl. You can compile your own version of Rust with musl enabled and install it into a custom directory with the instructions below:

$ mkdir musldist
  $ PREFIX=$(pwd)/musldist
  $
  $ # Build musl
  $ wget http://www.musl-libc.org/releases/musl-1.1.10.tar.gz
  [...]
  $ tar xf musl-1.1.10.tar.gz
  $ cd musl-1.1.10/
  musl-1.1.10 $ ./configure --disable-shared --prefix=$PREFIX
  [...]
  musl-1.1.10 $ make
  [...]
  musl-1.1.10 $ make install
  [...]
  musl-1.1.10 $ cd ..
  $ du -h musldist/lib/libc.a
  2.2M    musldist/lib/libc.a
  $
  $ # Build libunwind.a
  $ wget http://llvm.org/releases/3.6.1/llvm-3.6.1.src.tar.xz
  $ tar xf llvm-3.6.1.src.tar.xz
  $ cd llvm-3.6.1.src/projects/
  llvm-3.6.1.src/projects $ svn co http://llvm.org/svn/llvm-project/libcxxabi/trunk/ lib
  ↳ cxxabi
  llvm-3.6.1.src/projects $ svn co http://llvm.org/svn/llvm-project/libunwind/trunk/ lib
  ↳ unwind
  llvm-3.6.1.src/projects $ sed -i 's#^\(include_directories\).*$#\0\n\1(../libcxxabi/in
  ↳ clude)#' libunwind/CMakeLists.txt
  llvm-3.6.1.src/projects $ mkdir libunwind/build
  llvm-3.6.1.src/projects $ cd libunwind/build
  llvm-3.6.1.src/projects/libunwind/build $ cmake -DLLVM_PATH=../../.. -DLIBUNWIND_ENABL
  ↳ E_SHARED=0 ..
  llvm-3.6.1.src/projects/libunwind/build $ make
  llvm-3.6.1.src/projects/libunwind/build $ cp lib/libunwind.a $PREFIX/lib/
  llvm-3.6.1.src/projects/libunwind/build $ cd cd ../../../../
  $ du -h musldist/lib/libunwind.a
  164K    musldist/lib/libunwind.a
  $
  $ # Build musl-enabled rust
  $ git clone https://github.com/rust-lang/rust.git muslrust
  $ cd muslrust
  muslrust $ ./configure --target=x86_64-unknown-linux-musl --musl-root=$PREFIX --prefix
  ↳ =$PREFIX
  muslrust $ make
  muslrust $ make install
  muslrust $ cd ..
  $ du -h musldist/bin/rustc
  12K     musldist/bin/rustc

You now have a build of a musl-enabled Rust! Because we’ve installed it to a custom prefix we need to make sure our system can find the binaries and appropriate libraries when we try and run it:

$ export PATH=$PREFIX/bin:$PATH
  $ export LD_LIBRARY_PATH=$PREFIX/lib:$LD_LIBRARY_PATH

Let’s try it out!

$ echo 'fn main() { println!("hi!"); panic!("failed"); }' > example.rs
  $ rustc --target=x86_64-unknown-linux-musl example.rs
  $ ldd example
          not a dynamic executable
  $ ./example
  hi!
  thread '<main>' panicked at 'failed', example.rs:1

Success! This binary can be copied to almost any Linux machine with the same machine architecture and run without issues.

cargo build also permits the --target option so you should be able to build your crates as normal. However, you may need to recompile your native libraries against musl before they can be linked against.

Benchmark Tests

Rust supports benchmark tests, which can test the performance of your code. Let’s make our src/lib.rs look like this (comments elided):

#![feature(test)]
  
  extern crate test;
  
  pub fn add_two(a: i32) -> i32 {
      a + 2
  }
  
  #[cfg(test)]
  mod tests {
      use super::*;
      use test::Bencher;
  
      #[test]
      fn it_works() {
          assert_eq!(4, add_two(2));
      }
  
      #[bench]
      fn bench_add_two(b: &mut Bencher) {
          b.iter(|| add_two(2));
      }
  }

Note the test feature gate, which enables this unstable feature.

We’ve imported the test crate, which contains our benchmarking support. We have a new function as well, with the bench attribute. Unlike regular tests, which take no arguments, benchmark tests take a &mut Bencher. This Bencher provides an iter method, which takes a closure. This closure contains the code we’d like to benchmark.

We can run benchmark tests with cargo bench:

$ cargo bench
     Compiling adder v0.0.1 (file:///home/steve/tmp/adder)
       Running target/release/adder-91b3e234d4ed382a
  
  running 2 tests
  test tests::it_works ... ignored
  test tests::bench_add_two ... bench:         1 ns/iter (+/- 0)
  
  test result: ok. 0 passed; 0 failed; 1 ignored; 1 measured

Our non-benchmark test was ignored. You may have noticed that cargo bench takes a bit longer than cargo test. This is because Rust runs our benchmark a number of times, and then takes the average. Because we’re doing so little work in this example, we have a 1 ns/iter (+/- 0), but this would show the variance if there was one.

Advice on writing benchmarks:

  • Move setup code outside the iter loop; only put the part you want to measure inside
  • Make the code do “the same thing” on each iteration; do not accumulate or change state
  • Make the outer function idempotent too; the benchmark runner is likely to run it many times
  • Make the inner iter loop short and fast so benchmark runs are fast and the calibrator can adjust the run-length at fine resolution
  • Make the code in the iter loop do something simple, to assist in pinpointing performance improvements (or regressions)

Gotcha: optimizations

There’s another tricky part to writing benchmarks: benchmarks compiled with optimizations activated can be dramatically changed by the optimizer so that the benchmark is no longer benchmarking what one expects. For example, the compiler might recognize that some calculation has no external effects and remove it entirely.

#![feature(test)]
  
  extern crate test;
  use test::Bencher;
  
  #[bench]
  fn bench_xor_1000_ints(b: &mut Bencher) {
      b.iter(|| {
          (0..1000).fold(0, |old, new| old ^ new);
      });
  }

gives the following results

running 1 test
  test bench_xor_1000_ints ... bench:         0 ns/iter (+/- 0)
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 1 measured

The benchmarking runner offers two ways to avoid this. Either, the closure that the iter method receives can return an arbitrary value which forces the optimizer to consider the result used and ensures it cannot remove the computation entirely. This could be done for the example above by adjusting the b.iter call to

b.iter(|| {
      // note lack of `;` (could also use an explicit `return`).
      (0..1000).fold(0, |old, new| old ^ new)
  });

Or, the other option is to call the generic test::black_box function, which is an opaque “black box” to the optimizer and so forces it to consider any argument as used.

#![feature(test)]
  
  extern crate test;
  
  b.iter(|| {
      let n = test::black_box(1000);
  
      (0..n).fold(0, |a, b| a ^ b)
  })

Neither of these read or modify the value, and are very cheap for small values. Larger values can be passed indirectly to reduce overhead (e.g. black_box(&huge_struct)).

Performing either of the above changes gives the following benchmarking results

running 1 test
  test bench_xor_1000_ints ... bench:       131 ns/iter (+/- 3)
  
  test result: ok. 0 passed; 0 failed; 0 ignored; 1 measured

However, the optimizer can still modify a testcase in an undesirable manner even when using either of the above.

Box Syntax and Patterns

Currently the only stable way to create a Box is via the Box::new method. Also it is not possible in stable Rust to destructure a Box in a match pattern. The unstable box keyword can be used to both create and destructure a Box. An example usage would be:

#![feature(box_syntax, box_patterns)]
  
  fn main() {
      let b = Some(box 5);
      match b {
          Some(box n) if n < 0 => {
              println!("Box contains negative number {}", n);
          },
          Some(box n) if n >= 0 => {
              println!("Box contains non-negative number {}", n);
          },
          None => {
              println!("No box");
          },
          _ => unreachable!()
      }
  }

Note that these features are currently hidden behind the box_syntax (box creation) and box_patterns (destructuring and pattern matching) gates because the syntax may still change in the future.

Returning Pointers

In many languages with pointers, you’d return a pointer from a function so as to avoid copying a large data structure. For example:

struct BigStruct {
      one: i32,
      two: i32,
      // etc
      one_hundred: i32,
  }
  
  fn foo(x: Box<BigStruct>) -> Box<BigStruct> {
      Box::new(*x)
  }
  
  fn main() {
      let x = Box::new(BigStruct {
          one: 1,
          two: 2,
          one_hundred: 100,
      });
  
      let y = foo(x);
  }

The idea is that by passing around a box, you’re only copying a pointer, rather than the hundred i32s that make up the BigStruct.

This is an antipattern in Rust. Instead, write this:

#![feature(box_syntax)]
  
  struct BigStruct {
      one: i32,
      two: i32,
      // etc
      one_hundred: i32,
  }
  
  fn foo(x: Box<BigStruct>) -> BigStruct {
      *x
  }
  
  fn main() {
      let x = Box::new(BigStruct {
          one: 1,
          two: 2,
          one_hundred: 100,
      });
  
      let y: Box<BigStruct> = box foo(x);
  }

This gives you flexibility without sacrificing performance.

You may think that this gives us terrible performance: return a value and then immediately box it up ?! Isn’t this pattern the worst of both worlds? Rust is smarter than that. There is no copy in this code. main allocates enough room for the box, passes a pointer to that memory into foo as x, and then foo writes the value straight into the Box<T>.

This is important enough that it bears repeating: pointers are not for optimizing returning values from your code. Allow the caller to choose how they want to use your output.

Slice Patterns

If you want to match against a slice or array, you can use & with the slice_patterns feature:

#![feature(slice_patterns)]
  
  fn main() {
      let v = vec!["match_this", "1"];
  
      match &v[..] {
          ["match_this", second] => println!("The second element is {}", second),
          _ => {},
      }
  }

The advanced_slice_patterns gate lets you use .. to indicate any number of elements inside a pattern matching a slice. This wildcard can only be used once for a given array. If there’s an identifier before the .., the result of the slice will be bound to that name. For example:

#![feature(advanced_slice_patterns, slice_patterns)]
  
  fn is_symmetric(list: &[u32]) -> bool {
      match list {
          [] | [_] => true,
          [x, inside.., y] if x == y => is_symmetric(inside),
          _ => false
      }
  }
  
  fn main() {
      let sym = &[0, 1, 4, 2, 4, 1, 0];
      assert!(is_symmetric(sym));
  
      let not_sym = &[0, 1, 7, 2, 4, 1, 0];
      assert!(!is_symmetric(not_sym));
  }

Associated Constants

With the associated_consts feature, you can define constants like this:

#![feature(associated_consts)]
  
  trait Foo {
      const ID: i32;
  }
  
  impl Foo for i32 {
      const ID: i32 = 1;
  }
  
  fn main() {
      assert_eq!(1, i32::ID);
  }

Any implementor of Foo will have to define ID. Without the definition:

#![feature(associated_consts)]
  
  trait Foo {
      const ID: i32;
  }
  
  impl Foo for i32 {
  }

gives

error: not all trait items implemented, missing: `ID` [E0046]
       impl Foo for i32 {
       }

A default value can be implemented as well:

#![feature(associated_consts)]
  
  trait Foo {
      const ID: i32 = 1;
  }
  
  impl Foo for i32 {
  }
  
  impl Foo for i64 {
      const ID: i32 = 5;
  }
  
  fn main() {
      assert_eq!(1, i32::ID);
      assert_eq!(5, i64::ID);
  }

As you can see, when implementing Foo, you can leave it unimplemented, as with i32. It will then use the default value. But, as in i64, we can also add our own definition.

Associated constants don’t have to be associated with a trait. An impl block for a struct works fine too:

#![feature(associated_consts)]
  
  struct Foo;
  
  impl Foo {
      pub const FOO: u32 = 3;
  }

Glossary

Not every Rustacean has a background in systems programming, nor in computer science, so we’ve added explanations of terms that might be unfamiliar.

Abstract Syntax Tree

When a compiler is compiling your program, it does a number of different things. One of the things that it does is turn the text of your program into an ‘abstract syntax tree’, or ‘AST’. This tree is a representation of the structure of your program. For example, 2 + 3 can be turned into a tree:

  +
   / \
  2   3

And 2 + (3 * 4) would look like this:

  +
   / \
  2   *
     / \
    3   4
Arity

Arity refers to the number of arguments a function or operation takes.

let x = (2, 3);
  let y = (4, 6);
  let z = (8, 2, 6);

In the example above x and y have arity 2. z has arity 3.

DST (Dynamically Sized Type)

A type without a statically known size or alignment. (more info)

Expression

In computer programming, an expression is a combination of values, constants, variables, operators and functions that evaluate to a single value. For example, 2 + (3 * 4) is an expression that returns the value 14. It is worth noting that expressions can have side-effects. For example, a function included in an expression might perform actions other than simply returning a value.

Expression-Oriented Language

In early programming languages, expressions and statements were two separate syntactic categories: expressions had a value and statements did things. However, later languages blurred this distinction, allowing expressions to do things and statements to have a value. In an expression-oriented language, (nearly) every statement is an expression and therefore returns a value. Consequently, these expression statements can themselves form part of larger expressions.

Statement

In computer programming, a statement is the smallest standalone element of a programming language that commands a computer to perform an action.

Bibliography

This is a reading list of material relevant to Rust. It includes prior research that has - at one time or another - influenced the design of Rust, as well as publications about Rust.

Type system
Concurrency
Others
Papers about Rust

  1. We can make the memory live longer by transferring ownership, sometimes called ‘moving out of the box’. More complex examples will be covered later.

  2. Arc<UnsafeCell<T>> actually won’t compile since UnsafeCell<T> isn’t Send or Sync, but we can wrap it in a type and implement Send/Sync for it manually to get Arc<Wrapper<T>> where Wrapper is struct Wrapper<T>(UnsafeCell<T>).

  3. &[T] and &mut [T] are slices; they consist of a pointer and a length and can refer to a portion of a vector or array. &mut [T] can have its elements mutated, however its length cannot be touched.

  4. The actual definition of vec! in libcollections differs from the one presented here, for reasons of efficiency and reusability.