The Cairo Book

By the Cairo Community and its contributors. Special thanks to StarkWare through OnlyDust, and Voyager for supporting the creation of this book.

This version of the text assumes you’re using the Cairo Compiler version 2.8.3 and Starknet Foundry version 0.31.0. See the Installation section of Chapter 1 to install or update Cairo and Starknet Foundry.

If you want to play around with Cairo code and see how it compiles into Sierra (Intermediate Representation) and CASM (Cairo Assembly), you can use cairovm.codes playground.

Foreword

In 2020, StarkWare released Cairo 0, a Turing-complete programming language supporting verifiable computation. Cairo started as an assembly language and gradually became more expressive. The learning curve was initially steep, as Cairo 0.x was a low-level language that did not entirely abstract the underlying cryptographic primitives required to build a proof for the execution of a program.

With the release of Cairo 1, the developer experience has considerably improved, abstracting away the underlying immutable memory model of the Cairo architecture where possible. Strongly inspired by Rust, Cairo 1 has been built to help you create provable programs without specific knowledge of its underlying architecture so that you can focus on the program itself, increasing the overall security of Cairo programs. Powered by a Rust VM, the execution of Cairo programs is now blazingly fast, allowing you to build an extensive test suite without compromising on performance.

Blockchain developers that want to deploy contracts on Starknet will use the Cairo programming language to code their smart contracts. This allows the Starknet OS to generate execution traces for transactions to be proved by a prover, which is then verified on Ethereum L1 prior to updating the state root of Starknet.

However, Cairo is not only for blockchain developers. As a general purpose programming language, it can be used for any computation that would benefit from being proved on one computer and verified on other machines with lower hardware requirements.

This book is designed for developers with a basic understanding of programming concepts. It is a friendly and approachable text intended to help you level up your knowledge of Cairo, but also help you develop your programming skills in general. So, dive in and get ready to learn all there is to know about Cairo!

Acknowledgements

This book would not have been possible without the help of the Cairo community. We would like to thank every contributor for their contributions to this book!

We would like to thank the Rust community for the Rust Book, which has been a great source of inspiration for this book. Many examples and explanations have been adapted from the Rust Book to fit the Cairo programming language, as the two languages share many similarities.

Introduction

What is Cairo?

Cairo is a programming language designed for a virtual CPU of the same name. The unique aspect of this processor is that it was not created for the physical constraints of our world but for cryptographic ones, making it capable of efficiently proving the execution of any program running on it. This means that you can perform time consuming operations on a machine you don't trust, and check the result very quickly on a cheaper machine. While Cairo 0 used to be directly compiled to CASM, the Cairo CPU assembly, Cairo 1 is a higher level language. It first compiles to Sierra, an intermediate representation of Cairo which will compile later down to a safe subset of CASM. The point of Sierra is to ensure your CASM will always be provable, even when the computation fails.

What Can you Do with It?

Cairo allows you to compute trustworthy values on untrusted machines. One major usecase is Starknet, a solution to Ethereum scaling. Ethereum is a decentralized blockchain platform that enables the creation of decentralized applications where every single interaction between a user and a d-app is verified by all the participants. Starknet is a Layer 2 built on top of Ethereum. Instead of having all the participants of the network to verify all user interactions, only one node, called the prover, executes the programs and generates proofs that the computations were done correctly. These proofs are then verified by an Ethereum smart contract, requiring significantly less computational power compared to executing the interactions themselves. This approach allows for increased throughput and reduced transaction costs while preserving Ethereum security.

What Are the Differences with Other Programming Languages?

Cairo is quite different from traditional programming languages, especially when it comes to overhead costs and its primary advantages. Your program can be executed in two different ways:

• When executed by the prover, it is similar to any other language. Because Cairo is virtualized, and because the operations were not specifically designed for maximum efficiency, this can lead to some performance overhead but it is not the most relevant part to optimize.

• When the generated proof is verified by a verifier, it is a bit different. This has to be as cheap as possible since it could potentially be verified on many very small machines. Fortunately verifying is faster than computing and Cairo has some unique advantages to improve it even more. A notable one is non-determinism. This is a topic you will cover in more detail later in this book, but the idea is that you can theoretically use a different algorithm for verifying than for computing. Currently, writing custom non-deterministic code is not supported for the developers, but the standard library leverages non-determinism for improved performance. For example sorting an array in Cairo costs the same price as copying it. Because the verifier doesn't sort the array, it just checks that it is sorted, which is cheaper.

Another aspect that sets the language apart is its memory model. In Cairo, memory access is immutable, meaning that once a value is written to memory, it cannot be changed. Cairo 1 provides abstractions that help developers work with these constraints, but it does not fully simulate mutability. Therefore, developers must think carefully about how they manage memory and data structures in their programs to optimize performance.

References

• Cairo CPU Architecture: https://eprint.iacr.org/2021/1063
• Cairo, Sierra and Casm: https://medium.com/nethermind-eth/under-the-hood-of-cairo-1-0-exploring-sierra-7f32808421f5
• State of non determinism: https://twitter.com/PapiniShahar/status/1638203716535713798

Getting Started

Let’s start your Cairo journey! There’s a lot to learn, but every journey starts somewhere. In this chapter, we’ll discuss:

• Installing Scarb, which is Cairo's build toolchain and package manager, on Linux, macOS, and Windows.
• Installing Starknet Foundry, which is the default test runnner when creating a Cairo project.
• Writing a program that prints Hello, world!.
• Using basic Scarb commands to create a project and execute a program.

Getting Help

If you have any questions about Starknet or Cairo, you can ask them in the Starknet Discord server. The community is friendly and always willing to help.

Interacting with the Starknet AI Agent

Starknet proposes its own AI agent designed to assist with Cairo and Starknet-related questions. This AI agent is trained on the Cairo book and the Starknet documentation, using Retrieval-Augmented Generation (RAG) to efficiently retrieve information and provide accurate assistance.

You can find the Starknet Agent on the Starknet Agent website.

Installation

Cairo can be installed by simply downloading Scarb. Scarb bundles the Cairo compiler and the Cairo language server together in an easy-to-install package so that you can start writing Cairo code right away.

Scarb is also Cairo's package manager and is heavily inspired by Cargo, Rust’s build system and package manager.

Scarb handles a lot of tasks for you, such as building your code (either pure Cairo or Starknet contracts), downloading the libraries your code depends on, building those libraries, and provides LSP support for the VSCode Cairo 1 extension.

As you write more complex Cairo programs, you might add dependencies, and if you start a project using Scarb, managing external code and dependencies will be a lot easier to do.

Starknet Foundry is a toolchain for Cairo programs and Starknet smart contract development. It supports many features, including writing and running tests with advanced features, deploying contracts, interacting with the Starknet network, and more.

Let's start by installing Scarb and Starknet Foundry.

Installing Scarb

Requirements

Scarb requires a Git executable to be available in the PATH environment variable.

Installation

To install Scarb, please refer to the installation instructions. We strongly recommend that you install Scarb via asdf, a CLI tool that can manage multiple language runtime versions on a per-project basis. This will ensure that the version of Scarb you use to work on a project always matches the one defined in the project settings, avoiding problems related to version mismatches.

Please refer to the asdf documentation to install all prerequisites.

Once you have asdf installed locally, you can download Scarb plugin with the following command:

asdf plugin add scarb

This will allow you to download specific versions:

asdf install scarb 2.8.3

and set a global version:

asdf global scarb 2.8.3

Otherwise, you can simply run the following command in your terminal, and follow the onscreen instructions. This will install the latest stable release of Scarb.

curl --proto '=https' --tlsv1.2 -sSf https://docs.swmansion.com/scarb/install.sh | sh

$ scarb --version
scarb 2.8.3 (54938ce3b 2024-09-26)
cairo: 2.8.2 (https://crates.io/crates/cairo-lang-compiler/2.8.2)
sierra: 1.6.0

Installing Starknet Foundry

To install Starknet Foundry, please refer to the installation instructions. We also recommend that you install it via asdf.

Once installed, you can run the following command to see the version:

$ snforge --version
snforge 0.31.0

We'll describe Starknet Foundry in more detail in Chapter 10 for Cairo programs testing and in Chapter 17 when discussing Starknet smart contract testing and security in the second part of the book.

Installing the VSCode Extension

Cairo has a VSCode extension that provides syntax highlighting, code completion, and other useful features. You can install it from the VSCode Marketplace. Once installed, go into the extension settings, and make sure to tick the Enable Language Server and Enable Scarb options.

Hello, World

Now that you’ve installed Cairo through Scarb, it’s time to write your first Cairo program. It’s traditional when learning a new language to write a little program that prints the text Hello, world! to the screen, so we’ll do the same here!

Note: This book assumes basic familiarity with the command line. Cairo makes no specific demands about your editing or tooling or where your code lives, so if you prefer to use an integrated development environment (IDE) instead of the command line, feel free to use your favorite IDE. The Cairo team has developed a VSCode extension for the Cairo language that you can use to get the features from the language server and code highlighting. See Appendix F for more details.

Creating a Project Directory

You’ll start by making a directory to store your Cairo code. It doesn’t matter to Cairo where your code lives, but for the exercises and projects in this book, we suggest making a cairo_projects directory in your home directory and keeping all your projects there.

Open a terminal and enter the following commands to make a cairo_projects directory.

For Linux, macOS, and PowerShell on Windows, enter this:

mkdir ~/cairo_projects
cd ~/cairo_projects

For Windows CMD, enter this:

> mkdir "%USERPROFILE%\cairo_projects"
> cd /d "%USERPROFILE%\cairo_projects"

Note: From now on, for each example shown in the book, we assume that you will be working from a Scarb project directory. If you are not using Scarb, and try to run the examples from a different directory, you might need to adjust the commands accordingly or create a Scarb project.

Creating a Project with Scarb

Let’s create a new project using Scarb.

Navigate to your cairo_projects directory (or wherever you decided to store your code). Then run the following:

scarb new hello_world

Scarb will ask you about the dependencies you want to add. You will be given two options :

? Which test runner do you want to set up? ›
❯ Starknet Foundry (default)
  Cairo Test

In general, we'll prefer using the first one ❯ Starknet Foundry (default).

This creates a new directory and project called hello_world. We’ve named our project hello_world, and Scarb creates its files in a directory of the same name.

Go into the hello_world directory with the command cd hello_world. You’ll see that Scarb has generated three files and two directory for us: a Scarb.toml file, a src directory with a lib.cairo file inside and a tests directory containing a test_contract.cairo file. For now, we can remove this tests directory.

It has also initialized a new Git repository along with a .gitignore file

Note: Git is a common version control system. You can stop using version control system by using the --no-vcs flag. Run scarb new --help to see the available options.

Open Scarb.toml in your text editor of choice. It should look similar to the code in Listing 1-1.

Filename: Scarb.toml

[package]
name = "hello_world"
version = "0.1.0"
edition = "2024_07"

# See more keys and their definitions at https://docs.swmansion.com/scarb/docs/reference/manifest.html

[dependencies]
starknet = "2.8.2"

[dev-dependencies]
snforge_std = { git = "https://github.com/foundry-rs/starknet-foundry", tag = "v0.31.0" }

[[target.starknet-contract]]
sierra = true

[scripts]
test = "snforge test"

Listing 1-1: Contents of Scarb.toml generated by scarb new

This file is in the TOML (Tom’s Obvious, Minimal Language) format, which is Scarb’s configuration format.

The first line, [package], is a section heading that indicates that the following statements are configuring a package. As we add more information to this file, we’ll add other sections.

The next three lines set the configuration information Scarb needs to compile your program: the name of the package and the version of Scarb to use, and the edition of the prelude to use. The prelude is the collection of the most commonly used items that are automatically imported into every Cairo program. You can learn more about the prelude in Appendix D.

The [dependencies] section, is the start of a section for you to list any of your project’s dependencies. In Cairo, packages of code are referred to as crates. We won’t need any other crates for this project.

Note: By default, using Starknet Foundry adds the starknet dependency, so that you can also build contracts for Starknet.

The [dev-dependencies] section is about dependencies that are required for development, but are not needed for the actual production build of the project.

The [[target.starknet-contract]] section allows to build Starknet smart contracts. We can remove it for now.

The [script] section allows to define custom scripts. By default, there is one script for running tests using snforge with the scarb test command. We can also remove it for now.

The other file created by Scarb is src/lib.cairo, let's delete all the content and put in the following content, we will explain the reason later.

mod hello_world;

Then create a new file called src/hello_world.cairo and put the following code in it:

Filename: src/hello_world.cairo

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

We have just created a file called lib.cairo, which contains a module declaration referencing another module named hello_world, as well as the file hello_world.cairo, containing the implementation details of the hello_world module.

Scarb requires your source files to be located within the src directory.

The top-level project directory is reserved for README files, license information, configuration files, and any other non-code-related content. Scarb ensures a designated location for all project components, maintaining a structured organization.

If you started a project that doesn’t use Scarb, you can convert it to a project that does use Scarb. Move the project code into the src directory and create an appropriate Scarb.toml file. You can also use scarb init command to generate the src folder and the Scarb.toml it contains.

├── Scarb.toml
├── src
│   ├── lib.cairo
│   └── hello_world.cairo

A sample Scarb project structure

Building a Scarb Project

From your hello_world directory, build your project by entering the following command:

$ scarb build
   Compiling hello_world v0.1.0 (file:///projects/Scarb.toml)
    Finished release target(s) in 0 seconds

This command creates a sierra file in target/dev, let's ignore the sierra file for now.

If you have installed Cairo correctly, you should be able to run the main function of your program with the scarb cairo-run command and see the following output:

$ scarb cairo-run
Running hello_world
Hello, World!
Run completed successfully, returning []

Regardless of your operating system, the string Hello, world! should be printed to the terminal.

If Hello, world! did print, congratulations! You’ve officially written a Cairo program. That makes you a Cairo programmer — welcome!

Anatomy of a Cairo Program

Let’s review this “Hello, world!” program in detail. Here’s the first piece of the puzzle:

fn main() {

}

These lines define a function named main. The main function is special: it is always the first code that runs in every executable Cairo program. Here, the first line declares a function named main that has no parameters and returns nothing. If there were parameters, they would go inside the parentheses ().

The function body is wrapped in {}. Cairo requires curly brackets around all function bodies. It’s good style to place the opening curly bracket on the same line as the function declaration, adding one space in between.

Note: If you want to stick to a standard style across Cairo projects, you can use the automatic formatter tool available with scarb fmt to format your code in a particular style (more on scarb fmt in Appendix F). The Cairo team has included this tool with the standard Cairo distribution, as cairo-run is, so it should already be installed on your computer!

The body of the main function holds the following code:

    println!("Hello, World!");

This line does all the work in this little program: it prints text to the screen. There are four important details to notice here.

First, Cairo style is to indent with four spaces, not a tab.

Second, println! calls a Cairo macro. If it had called a function instead, it would be entered as println (without the !). We’ll discuss Cairo macros in more detail in the "Macros" chapter. For now, you just need to know that using a ! means that you’re calling a macro instead of a normal function and that macros don’t always follow the same rules as functions.

Third, you see the "Hello, world!" string. We pass this string as an argument to println!, and the string is printed to the screen.

Fourth, we end the line with a semicolon (;), which indicates that this expression is over and the next one is ready to begin. Most lines of Cairo code end with a semicolon.

Summary

Let’s recap what we’ve learned so far about Scarb:

• We can install one or multiple Scarb versions, either the latest stable or a specific one, using asdf.
• We can create a project using scarb new.
• We can build a project using scarb build to generate the compiled Sierra code.
• We can execute a Cairo program using the scarb cairo-run command.

An additional advantage of using Scarb is that the commands are the same no matter which operating system you’re working on. So, at this point, we’ll no longer provide specific instructions for Linux and macOS versus Windows.

You’re already off to a great start on your Cairo journey! This is a great time to build a more substantial program to get used to reading and writing Cairo code.

Common Programming Concepts

This chapter covers concepts that appear in almost every programming language and how they work in Cairo. Many programming languages have much in common at their core. None of the concepts presented in this chapter are unique to Cairo, but we’ll discuss them in the context of Cairo and explain the conventions around using these concepts.

Specifically, you’ll learn about variables, basic types, functions, comments, and control flow. These foundations will be in every Cairo program, and learning them early will give you a strong core to start from.

Variables and Mutability

Cairo uses an immutable memory model, meaning that once a memory cell is written to, it can't be overwritten but only read from. To reflect this immutable memory model, variables in Cairo are immutable by default. However, the language abstracts this model and gives you the option to make your variables mutable. Let’s explore how and why Cairo enforces immutability, and how you can make your variables mutable.

When a variable is immutable, once a value is bound to a name, you can’t change that value. To illustrate this, generate a new project called variables in your cairo_projects directory by using scarb new variables.

Then, in your new variables directory, open src/lib.cairo and replace its code with the following code, which won’t compile just yet:

Filename: src/lib.cairo

fn main() {
    let x = 5;
    println!("The value of x is: {}", x);
    x = 6;
    println!("The value of x is: {}", x);
}

Save and run the program using scarb cairo-run. You should receive an error message regarding an immutability error, as shown in this output:

$ scarb cairo-run 
   Compiling no_listing_01_variables_are_immutable v0.1.0 (listings/ch02-common-programming-concepts/no_listing_01_variables_are_immutable/Scarb.toml)
error: Cannot assign to an immutable variable.
 --> listings/ch02-common-programming-concepts/no_listing_01_variables_are_immutable/src/lib.cairo:6:5
    x = 6;
    ^***^

error: could not compile `no_listing_01_variables_are_immutable` due to previous error
error: `scarb metadata` exited with error

This example shows how the compiler helps you find errors in your programs. Compiler errors can be frustrating, but they only mean your program isn’t safely doing what you want it to do yet; they do not mean that you’re not a good programmer! Experienced Caironautes still get compiler errors.

You received the error message Cannot assign to an immutable variable. because you tried to assign a second value to the immutable x variable.

It’s important that we get compile-time errors when we attempt to change a value that’s designated as immutable because this specific situation can lead to bugs. If one part of our code operates on the assumption that a value will never change and another part of our code changes that value, it’s possible that the first part of the code won’t do what it was designed to do. The cause of this kind of bug can be difficult to track down after the fact, especially when the second piece of code changes the value only sometimes.

Cairo, unlike most other languages, has immutable memory. This makes a whole class of bugs impossible, because values will never change unexpectedly. This makes code easier to reason about.

But mutability can be very useful, and can make code more convenient to write. Although variables are immutable by default, you can make them mutable by adding mut in front of the variable name. Adding mut also conveys intent to future readers of the code by indicating that other parts of the code will be changing the value associated to this variable.

However, you might be wondering at this point what exactly happens when a variable is declared as mut, as we previously mentioned that Cairo's memory is immutable. The answer is that the value is immutable, but the variable isn't. The value associated to the variable can be changed. Assigning to a mutable variable in Cairo is essentially equivalent to redeclaring it to refer to another value in another memory cell, but the compiler handles that for you, and the keyword mut makes it explicit. Upon examining the low-level Cairo Assembly code, it becomes clear that variable mutation is implemented as syntactic sugar, which translates mutation operations into a series of steps equivalent to variable shadowing. The only difference is that at the Cairo level, the variable is not redeclared so its type cannot change.

For example, let’s change src/lib.cairo to the following:

fn main() {
    let mut x = 5;
    println!("The value of x is: {}", x);
    x = 6;
    println!("The value of x is: {}", x);
}

When we run the program now, we get this:

$ scarb cairo-run 
   Compiling no_listing_02_adding_mut v0.1.0 (listings/ch02-common-programming-concepts/no_listing_02_adding_mut/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running no_listing_02_adding_mut
The value of x is: 5
The value of x is: 6
Run completed successfully, returning []

We’re allowed to change the value bound to x from 5 to 6 when mut is used. Ultimately, deciding whether to use mutability or not is up to you and depends on what you think is clearest in that particular situation.

Constants

Like immutable variables, constants are values that are bound to a name and are not allowed to change, but there are a few differences between constants and variables.

First, you aren’t allowed to use mut with constants. Constants aren’t just immutable by default—they’re always immutable. You declare constants using the const keyword instead of the let keyword, and the type of the value must be annotated. We’ll cover types and type annotations in the next section, “Data Types”, so don’t worry about the details right now. Just know that you must always annotate the type.

Constant variables can be declared with any usual data type, including structs, enums and fixed-size arrays.

Constants can only be declared in the global scope, which makes them useful for values that many parts of code need to know about.

The last difference is that constants may natively be set only to a constant expression, not the result of a value that could only be computed at runtime.

Here’s an example of constants declaration:

struct AnyStruct {
    a: u256,
    b: u32
}

enum AnyEnum {
    A: felt252,
    B: (usize, u256)
}

const ONE_HOUR_IN_SECONDS: u32 = 3600;
const STRUCT_INSTANCE: AnyStruct = AnyStruct { a: 0, b: 1 };
const ENUM_INSTANCE: AnyEnum = AnyEnum::A('any enum');
const BOOL_FIXED_SIZE_ARRAY: [bool; 2] = [true, false];

Nonetheless, it is possible to use the consteval_int! macro to create a const variable that is the result of some computation:

    const ONE_HOUR_IN_SECONDS: u32 = consteval_int!(60 * 60);

We will dive into more detail about macros in the dedicated section.

Cairo's naming convention for constants is to use all uppercase with underscores between words.

Constants are valid for the entire time a program runs, within the scope in which they were declared. This property makes constants useful for values in your application domain that multiple parts of the program might need to know about, such as the maximum number of points any player of a game is allowed to earn, or the speed of light.

Naming hardcoded values used throughout your program as constants is useful in conveying the meaning of that value to future maintainers of the code. It also helps to have only one place in your code you would need to change if the hardcoded value needed to be updated in the future.

Shadowing

Variable shadowing refers to the declaration of a new variable with the same name as a previous variable. Caironautes say that the first variable is shadowed by the second, which means that the second variable is what the compiler will see when you use the name of the variable. In effect, the second variable overshadows the first, taking any uses of the variable name to itself until either it itself is shadowed or the scope ends. We can shadow a variable by using the same variable’s name and repeating the use of the let keyword as follows:

fn main() {
    let x = 5;
    let x = x + 1;
    {
        let x = x * 2;
        println!("Inner scope x value is: {}", x);
    }
    println!("Outer scope x value is: {}", x);
}

This program first binds x to a value of 5. Then it creates a new variable x by repeating let x =, taking the original value and adding 1 so the value of x is then 6. Then, within an inner scope created with the curly brackets, the third let statement also shadows x and creates a new variable, multiplying the previous value by 2 to give x a value of 12. When that scope is over, the inner shadowing ends and x returns to being 6. When we run this program, it will output the following:

$ scarb cairo-run 
   Compiling no_listing_03_shadowing v0.1.0 (listings/ch02-common-programming-concepts/no_listing_03_shadowing/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_03_shadowing
Inner scope x value is: 12
Outer scope x value is: 6
Run completed successfully, returning []

Shadowing is different from marking a variable as mut because we’ll get a compile-time error if we accidentally try to reassign to this variable without using the let keyword. By using let, we can perform a few transformations on a value but have the variable be immutable after those transformations have been completed.

Another distinction between mut and shadowing is that when we use the let keyword again, we are effectively creating a new variable, which allows us to change the type of the value while reusing the same name. As mentioned before, variable shadowing and mutable variables are equivalent at the lower level. The only difference is that by shadowing a variable, the compiler will not complain if you change its type. For example, say our program performs a type conversion between the u64 and felt252 types.

fn main() {
    let x: u64 = 2;
    println!("The value of x is {} of type u64", x);
    let x: felt252 = x.into(); // converts x to a felt, type annotation is required.
    println!("The value of x is {} of type felt252", x);
}

The first x variable has a u64 type while the second x variable has a felt252 type. Shadowing thus spares us from having to come up with different names, such as x_u64 and x_felt252; instead, we can reuse the simpler x name. However, if we try to use mut for this, as shown here, we’ll get a compile-time error:

fn main() {
    let mut x: u64 = 2;
    println!("The value of x is: {}", x);
    x = 'a short string';
    println!("The value of x is: {}", x);
}

The error says we were expecting a u64 (the original type) but we got a different type:

$ scarb cairo-run 
   Compiling no_listing_05_mut_cant_change_type v0.1.0 (listings/ch02-common-programming-concepts/no_listing_05_mut_cant_change_type/Scarb.toml)
error: The value does not fit within the range of type core::integer::u64.
 --> listings/ch02-common-programming-concepts/no_listing_05_mut_cant_change_type/src/lib.cairo:6:9
    x = 'a short string';
        ^**************^

error: could not compile `no_listing_05_mut_cant_change_type` due to previous error
error: `scarb metadata` exited with error

Now that we’ve explored how variables work, let’s look at more data types they can have.

Data Types

Every value in Cairo is of a certain data type, which tells Cairo what kind of data is being specified so it knows how to work with that data. This section covers two subsets of data types: scalars and compounds.

Keep in mind that Cairo is a statically typed language, which means that it must know the types of all variables at compile time. The compiler can usually infer the desired type based on the value and its usage. In cases when many types are possible, we can use a conversion method where we specify the desired output type.

fn main() {
    let x: felt252 = 3;
    let y: u32 = x.try_into().unwrap();
}

You’ll see different type annotations for other data types.

Scalar Types

A scalar type represents a single value. Cairo has three primary scalar types: felts, integers, and booleans. You may recognize these from other programming languages. Let’s jump into how they work in Cairo.

Felt Type

In Cairo, if you don't specify the type of a variable or argument, its type defaults to a field element, represented by the keyword felt252. In the context of Cairo, when we say “a field element” we mean an integer in the range \( 0 \leq x < P \), where \( P \) is a very large prime number currently equal to \( {2^{251}} + 17 \cdot {2^{192}} + 1 \). When adding, subtracting, or multiplying, if the result falls outside the specified range of the prime number, an overflow (or underflow) occurs, and an appropriate multiple of \( P \) is added or subtracted to bring the result back within the range (i.e., the result is computed \( \mod P \) ).

The most important difference between integers and field elements is division: Division of field elements (and therefore division in Cairo) is unlike regular CPUs division, where integer division \( \frac{x}{y} \) is defined as \( \left\lfloor \frac{x}{y} \right\rfloor \) where the integer part of the quotient is returned (so you get \( \frac{7}{3} = 2 \)) and it may or may not satisfy the equation \( \frac{x}{y} \cdot y == x \), depending on the divisibility of x by y.

In Cairo, the result of \( \frac{x}{y} \) is defined to always satisfy the equation \( \frac{x}{y} \cdot y == x \). If y divides x as integers, you will get the expected result in Cairo (for example \( \frac{6}{2} \) will indeed result in 3). But when y does not divide x, you may get a surprising result: for example, since \( 2 \cdot \frac{P + 1}{2} = P + 1 \equiv 1 \mod P \), the value of \( \frac{1}{2} \) in Cairo is \( \frac{P + 1}{2} \) (and not 0 or 0.5), as it satisfies the above equation.

Integer Types

The felt252 type is a fundamental type that serves as the basis for creating all types in the core library. However, it is highly recommended for programmers to use the integer types instead of the felt252 type whenever possible, as the integer types come with added security features that provide extra protection against potential vulnerabilities in the code, such as overflow and underflow checks. By using these integer types, programmers can ensure that their programs are more secure and less susceptible to attacks or other security threats. An integer is a number without a fractional component. This type declaration indicates the number of bits the programmer can use to store the integer. Table 3-1 shows the built-in integer types in Cairo. We can use any of these variants to declare the type of an integer value.

Each variant has an explicit size. Note that for now, the usize type is just an alias for u32; however, it might be useful when in the future Cairo can be compiled to MLIR. As variables are unsigned, they can't contain a negative number. This code will cause the program to panic:

fn sub_u8s(x: u8, y: u8) -> u8 {
    x - y
}

fn main() {
    sub_u8s(1, 3);
}

All integer types previously mentioned fit into a felt252, except for u256 which needs 4 more bits to be stored. Under the hood, u256 is basically a struct with 2 fields: u256 {low: u128, high: u128}.

Cairo also provides support for signed integers, starting with the prefix i. These integers can represent both positive and negative values, with sizes ranging from i8 to i128. Each signed variant can store numbers from \( -({2^{n - 1}}) \) to \( {2^{n - 1}} - 1 \) inclusive, where n is the number of bits that variant uses. So an i8 can store numbers from \( -({2^7}) \) to \( {2^7} - 1 \), which equals -128 to 127.

You can write integer literals in any of the forms shown in Table 3-2. Note that number literals that can be multiple numeric types allow a type suffix, such as 57_u8, to designate the type. It is also possible to use a visual separator _ for number literals, in order to improve code readability.

So how do you know which type of integer to use? Try to estimate the max value your int can have and choose the good size. The primary situation in which you’d use usize is when indexing some sort of collection.

Numeric Operations

Cairo supports the basic mathematical operations you’d expect for all the integer types: addition, subtraction, multiplication, division, and remainder. Integer division truncates toward zero to the nearest integer. The following code shows how you’d use each numeric operation in a let statement:

fn main() {
    // addition
    let sum = 5_u128 + 10_u128;

    // subtraction
    let difference = 95_u128 - 4_u128;

    // multiplication
    let product = 4_u128 * 30_u128;

    // division
    let quotient = 56_u128 / 32_u128; //result is 1
    let quotient = 64_u128 / 32_u128; //result is 2

    // remainder
    let remainder = 43_u128 % 5_u128; // result is 3
}

Each expression in these statements uses a mathematical operator and evaluates to a single value, which is then bound to a variable.

Appendix B contains a list of all operators that Cairo provides.

The Boolean Type

As in most other programming languages, a Boolean type in Cairo has two possible values: true and false. Booleans are one felt252 in size. The Boolean type in Cairo is specified using bool. For example:

fn main() {
    let t = true;

    let f: bool = false; // with explicit type annotation
}

When declaring a bool variable, it is mandatory to use either true or false literals as value. Hence, it is not allowed to use integer literals (i.e. 0 instead of false) for bool declarations.

The main way to use Boolean values is through conditionals, such as an if expression. We’ll cover how if expressions work in Cairo in the "Control Flow" section.

String Types

Cairo doesn't have a native type for strings but provides two ways to handle them: short strings using simple quotes and ByteArray using double quotes.

Short strings

A short string is an ASCII string where each character is encoded on one byte (see the ASCII table). For example:

• 'a' is equivalent to 0x61
• 'b' is equivalent to 0x62
• 'c' is equivalent to 0x63
• 0x616263 is equivalent to 'abc'.

Cairo uses the felt252 for short strings. As the felt252 is on 251 bits, a short string is limited to 31 characters (31 * 8 = 248 bits, which is the maximum multiple of 8 that fits in 251 bits).

You can choose to represent your short string with an hexadecimal value like 0x616263 or by directly writing the string using simple quotes like 'abc', which is more convenient.

Here are some examples of declaring short strings in Cairo:

fn main() {
    let my_first_char = 'C';
    let my_first_char_in_hex = 0x43;

    let my_first_string = 'Hello world';
    let my_first_string_in_hex = 0x48656C6C6F20776F726C64;

    let long_string: ByteArray = "this is a string which has more than 31 characters";
}

Byte Array Strings

Cairo's Core Library provides a ByteArray type for handling strings and byte sequences longer than short strings. This type is particularly useful for longer strings or when you need to perform operations on the string data.

The ByteArray in Cairo is implemented as a combination of two parts:

1. An array of bytes31 words, where each word contains 31 bytes of data.
2. A pending felt252 word that acts as a buffer for bytes that haven't yet filled a complete bytes31 word.

This design enables efficient handling of byte sequences while aligning with Cairo's memory model and basic types. Developers interact with ByteArray through its provided methods and operators, abstracting away the internal implementation details.

Unlike short strings, ByteArray strings can contain more than 31 characters and are written using double quotes:

fn main() {
    let my_first_char = 'C';
    let my_first_char_in_hex = 0x43;

    let my_first_string = 'Hello world';
    let my_first_string_in_hex = 0x48656C6C6F20776F726C64;

    let long_string: ByteArray = "this is a string which has more than 31 characters";
}

Compound Types

The Tuple Type

A tuple is a general way of grouping together a number of values with a variety of types into one compound type. Tuples have a fixed length: once declared, they cannot grow or shrink in size.

We create a tuple by writing a comma-separated list of values inside parentheses. Each position in the tuple has a type, and the types of the different values in the tuple don’t have to be the same. We’ve added optional type annotations in this example:

fn main() {
    let tup: (u32, u64, bool) = (10, 20, true);
}

The variable tup binds to the entire tuple because a tuple is considered a single compound element. To get the individual values out of a tuple, we can use pattern matching to destructure a tuple value, like this:

fn main() {
    let tup = (500, 6, true);

    let (x, y, z) = tup;

    if y == 6 {
        println!("y is 6!");
    }
}

This program first creates a tuple and binds it to the variable tup. It then uses a pattern with let to take tup and turn it into three separate variables, x, y, and z. This is called destructuring because it breaks the single tuple into three parts. Finally, the program prints y is 6! as the value of y is 6.

We can also declare the tuple with value and types, and destructure it at the same time. For example:

fn main() {
    let (x, y): (felt252, felt252) = (2, 3);
}

The Unit Type ()

A unit type is a type which has only one value (). It is represented by a tuple with no elements. Its size is always zero, and it is guaranteed to not exist in the compiled code.

You might be wondering why you would even need a unit type? In Cairo, everything is an expression, and an expression that returns nothing actually returns () implicitly.

The Fixed Size Array Type

Another way to have a collection of multiple values is with a fixed size array. Unlike a tuple, every element of a fixed size array must have the same type.

We write the values in a fixed-size array as a comma-separated list inside square brackets. The array’s type is written using square brackets with the type of each element, a semicolon, and then the number of elements in the array, like so:

fn main() {
    let arr1: [u64; 5] = [1, 2, 3, 4, 5];
}

In the type annotation [u64; 5], u64 specifies the type of each element, while 5 after the semicolon defines the array's length. This syntax ensures that the array always contains exactly 5 elements of type u64.

Fixed size arrays are useful when you want to hardcode a potentially long sequence of data directly in your program. This type of array must not be confused with the Array<T> type, which is a similar collection type provided by the core library that is allowed to grow in size. If you're unsure whether to use a fixed size array or the Array<T> type, chances are that you are looking for the Array<T> type.

Because their size is known at compile-time, fixed-size arrays don't require runtime memory management, which makes them more efficient than dynamically-sized arrays. Overall, they're more useful when you know the number of elements will not need to change. For example, they can be used to efficiently store lookup tables that won't change during runtime. If you were using the names of the month in a program, you would probably use a fixed size array rather than an Array<T> because you know it will always contain 12 elements:

    let months = [
        'January',
        'February',
        'March',
        'April',
        'May',
        'June',
        'July',
        'August',
        'September',
        'October',
        'November',
        'December'
    ];

You can also initialize an array to contain the same value for each element by specifying the initial value, followed by a semicolon, and then the length of the array in square brackets, as shown here:

    let a = [3; 5];

The array named a will contain 5 elements that will all be set to the value 3 initially. This is the same as writing let a = [3, 3, 3, 3, 3]; but in a more concise way.

Accessing Fixed Size Arrays Elements

As a fixed-size array is a data structure known at compile time, it's content is represented as a sequence of values in the program bytecode. Accessing an element of that array will simply read that value from the program bytecode efficiently.

We have two different ways of accessing fixed size array elements:

• Deconstructing the array into multiple variables, as we did with tuples.

fn main() {
    let my_arr = [1, 2, 3, 4, 5];

    // Accessing elements of a fixed-size array by deconstruction
    let [a, b, c, _, _] = my_arr;
    println!("c: {}", c); // c: 3    
}

• Converting the array to a Span, that supports indexing. This operation is free and doesn't incur any runtime cost.

fn main() {
    let my_arr = [1, 2, 3, 4, 5];

    // Accessing elements of a fixed-size array by index
    let my_span = my_arr.span();
    println!("my_span[2]: {}", my_span[2]); // my_span[2]: 3
}

Note that if we plan to repeatedly access the array, then it makes sense to call .span() only once and keep it available throughout the accesses.

Type Conversion

Cairo addresses conversion between types by using the try_into and into methods provided by the TryInto and Into traits from the core library. There are numerous implementations of these traits within the standard library for conversion between types, and they can be implemented for custom types as well.

Into

The Into trait allows for a type to define how to convert itself into another type. It can be used for type conversion when success is guaranteed, such as when the source type is smaller than the destination type.

To perform the conversion, call var.into() on the source value to convert it to another type. The new variable's type must be explicitly defined, as demonstrated in the example below.

fn main() {
    let my_u8: u8 = 10;
    let my_u16: u16 = my_u8.into();
    let my_u32: u32 = my_u16.into();
    let my_u64: u64 = my_u32.into();
    let my_u128: u128 = my_u64.into();

    let my_felt252 = 10;
    // As a felt252 is smaller than a u256, we can use the into() method
    let my_u256: u256 = my_felt252.into();
    let my_other_felt252: felt252 = my_u8.into();
    let my_third_felt252: felt252 = my_u16.into();
}

TryInto

Similar to Into, TryInto is a generic trait for converting between types. Unlike Into, the TryInto trait is used for fallible conversions, and as such, returns Option<T>. An example of a fallible conversion is when the target type might not fit the source value.

Also similar to Into is the process to perform the conversion; just call var.try_into() on the source value to convert it to another type. The new variable's type also must be explicitly defined, as demonstrated in the example below.

fn main() {
    let my_u256: u256 = 10;

    // Since a u256 might not fit in a felt252, we need to unwrap the Option<T> type
    let my_felt252: felt252 = my_u256.try_into().unwrap();
    let my_u128: u128 = my_felt252.try_into().unwrap();
    let my_u64: u64 = my_u128.try_into().unwrap();
    let my_u32: u32 = my_u64.try_into().unwrap();
    let my_u16: u16 = my_u32.try_into().unwrap();
    let my_u8: u8 = my_u16.try_into().unwrap();

    let my_large_u16: u16 = 2048;
    let my_large_u8: u8 = my_large_u16.try_into().unwrap(); // panics with 'Option::unwrap failed.'
}

Functions

Functions are prevalent in Cairo code. You’ve already seen one of the most important functions in the language: the main function, which is the entry point of many programs. You’ve also seen the fn keyword, which allows you to declare new functions.

Cairo code uses snake case as the conventional style for function and variable names, in which all letters are lowercase and underscores separate words. Here’s a program that contains an example function definition:

fn another_function() {
    println!("Another function.");
}

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

We define a function in Cairo by entering fn followed by a function name and a set of parentheses. The curly brackets tell the compiler where the function body begins and ends.

We can call any function we’ve defined by entering its name followed by a set of parentheses. Because another_function is defined in the program, it can be called from inside the main function. Note that we defined another_function before the main function in the source code; we could have defined it after as well. Cairo doesn’t care where you define your functions, only that they’re defined somewhere in a scope that can be seen by the caller.

Let’s start a new project with Scarb named functions to explore functions further. Place the another_function example in src/lib.cairo and run it. You should see the following output:

$ scarb cairo-run 
   Compiling no_listing_15_functions v0.1.0 (listings/ch02-common-programming-concepts/no_listing_15_functions/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_15_functions
Hello, world!
Another function.
Run completed successfully, returning []

The lines execute in the order in which they appear in the main function. First the Hello, world! message prints, and then another_function is called and its message is printed.

Parameters

We can define functions to have parameters, which are special variables that are part of a function’s signature. When a function has parameters, you can provide it with concrete values for those parameters. Technically, the concrete values are called arguments, but in casual conversation, people tend to use the words parameter and argument interchangeably for either the variables in a function’s definition or the concrete values passed in when you call a function.

In this version of another_function we add a parameter:

fn main() {
    another_function(5);
}

fn another_function(x: felt252) {
    println!("The value of x is: {}", x);
}

Try running this program; you should get the following output:

$ scarb cairo-run 
   Compiling no_listing_16_single_param v0.1.0 (listings/ch02-common-programming-concepts/no_listing_16_single_param/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_16_single_param
The value of x is: 5
Run completed successfully, returning []

The declaration of another_function has one parameter named x. The type of x is specified as felt252. When we pass 5 in to another_function, the println! macro puts 5 where the pair of curly brackets containing x was in the format string.

In function signatures, you must declare the type of each parameter. This is a deliberate decision in Cairo’s design: requiring type annotations in function definitions means the compiler almost never needs you to use them elsewhere in the code to figure out what type you mean. The compiler is also able to give more helpful error messages if it knows what types the function expects.

When defining multiple parameters, separate the parameter declarations with commas, like this:

fn main() {
    print_labeled_measurement(5, "h");
}

fn print_labeled_measurement(value: u128, unit_label: ByteArray) {
    println!("The measurement is: {value}{unit_label}");
}

This example creates a function named print_labeled_measurement with two parameters. The first parameter is named value and is a u128. The second is named unit_label and is of type ByteArray - Cairo's internal type to represent string literals. The function then prints text containing both the value and the unit_label.

Let’s try running this code. Replace the program currently in your functions project’s src/lib.cairo file with the preceding example and run it using scarb cairo-run:

$ scarb cairo-run 
   Compiling no_listing_17_multiple_params v0.1.0 (listings/ch02-common-programming-concepts/no_listing_17_multiple_params/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running no_listing_17_multiple_params
The measurement is: 5h
Run completed successfully, returning []

Because we called the function with 5 as the value for value and "h" as the value for unit_label, the program output contains those values.

Named Parameters

In Cairo, named parameters allow you to specify the names of arguments when you call a function. This makes the function calls more readable and self-descriptive. If you want to use named parameters, you need to specify the name of the parameter and the value you want to pass to it. The syntax is parameter_name: value. If you pass a variable that has the same name as the parameter, you can simply write :parameter_name instead of parameter_name: variable_name.

Here is an example:

fn foo(x: u8, y: u8) {}

fn main() {
    let first_arg = 3;
    let second_arg = 4;
    foo(x: first_arg, y: second_arg);
    let x = 1;
    let y = 2;
    foo(:x, :y)
}

Statements and Expressions

Function bodies are made up of a series of statements optionally ending in an expression. So far, the functions we’ve covered haven’t included an ending expression, but you have seen an expression as part of a statement. Because Cairo is an expression-based language, this is an important distinction to understand. Other languages don’t have the same distinctions, so let’s look at what statements and expressions are and how their differences affect the bodies of functions.

• Statements are instructions that perform some action and do not return a value.
• Expressions evaluate to a resultant value. Let’s look at some examples.

We’ve actually already used statements and expressions. Creating a variable and assigning a value to it with the let keyword is a statement. In Listing 2-1, let y = 6; is a statement.

fn main() {
    let y = 6;
}

Listing 2-1: A main function declaration containing one statement

Function definitions are also statements; the entire preceding example is a statement in itself.

Statements do not return values. Therefore, you can’t assign a let statement to another variable, as the following code tries to do; you’ll get an error:

fn main() {
    let x = (let y = 6);
}

When you run this program, the error you’ll get looks like this:

$ scarb cairo-run 
   Compiling no_listing_18_statements_dont_return_values v0.1.0 (listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/Scarb.toml)
error: Missing token TerminalRParen.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:14
    let x = (let y = 6);
             ^

error: Missing token TerminalSemicolon.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:14
    let x = (let y = 6);
             ^

error: Missing token TerminalSemicolon.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:23
    let x = (let y = 6);
                      ^

error: Skipped tokens. Expected: statement.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:23
    let x = (let y = 6);
                      ^^

warn[E0001]: Unused variable. Consider ignoring by prefixing with `_`.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:9
    let x = (let y = 6);
        ^

warn[E0001]: Unused variable. Consider ignoring by prefixing with `_`.
 --> listings/ch02-common-programming-concepts/no_listing_20_statements_dont_return_values/src/lib.cairo:3:18
    let x = (let y = 6);
                 ^

error: could not compile `no_listing_18_statements_dont_return_values` due to previous error
error: `scarb metadata` exited with error

The let y = 6 statement does not return a value, so there isn’t anything for x to bind to. This is different from what happens in other languages, such as C and Ruby, where the assignment returns the value of the assignment. In those languages, you can write x = y = 6 and have both x and y have the value 6; that is not the case in Cairo.

Expressions evaluate to a value and make up most of the rest of the code that you’ll write in Cairo. Consider a math operation, such as 5 + 6, which is an expression that evaluates to the value 11. Expressions can be part of statements: in Listing 2-1, the 6 in the statement let y = 6; is an expression that evaluates to the value 6.

Calling a function is an expression since it always evaluates to a value: the function's explicit return value, if specified, or the 'unit' type () otherwise.

A new scope block created with curly brackets is an expression, for example:

fn main() {
    let y = {
        let x = 3;
        x + 1
    };

    println!("The value of y is: {}", y);
}

This expression:

    let y = {
        let x = 3;
        x + 1
    };

is a block that, in this case, evaluates to 4. That value gets bound to y as part of the let statement. Note that the x + 1 line doesn’t have a semicolon at the end, which is unlike most of the lines you’ve seen so far. Expressions do not include ending semicolons. If you add a semicolon to the end of an expression, you turn it into a statement, and it will then not return a value. Keep this in mind as you explore function return values and expressions next.

Functions with Return Values

Functions can return values to the code that calls them. We don’t name return values, but we must declare their type after an arrow (->). In Cairo, the return value of the function is synonymous with the value of the final expression in the block of the body of a function. You can return early from a function by using the return keyword and specifying a value, but most functions return the last expression implicitly. Here’s an example of a function that returns a value:

fn five() -> u32 {
    5
}

fn main() {
    let x = five();
    println!("The value of x is: {}", x);
}

There are no function calls, or even let statements in the five function—just the number 5 by itself. That’s a perfectly valid function in Cairo. Note that the function’s return type is specified too, as -> u32. Try running this code; the output should look like this:

$ scarb cairo-run 
   Compiling no_listing_20_function_return_values v0.1.0 (listings/ch02-common-programming-concepts/no_listing_22_function_return_values/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_20_function_return_values
The value of x is: 5
Run completed successfully, returning []

The 5 in five is the function’s return value, which is why the return type is u32. Let’s examine this in more detail. There are two important bits: first, the line let x = five(); shows that we’re using the return value of a function to initialize a variable. Because the function five returns a 5, that line is the same as the following:

let x = 5;

Second, the five function has no parameters and defines the type of the return value, but the body of the function is a lonely 5 with no semicolon because it’s an expression whose value we want to return. Let’s look at another example:

fn main() {
    let x = plus_one(5);

    println!("The value of x is: {}", x);
}

fn plus_one(x: u32) -> u32 {
    x + 1
}

Running this code will print x = 6. But if we place a semicolon at the end of the line containing x + 1, changing it from an expression to a statement, we’ll get an error:

fn main() {
    let x = plus_one(5);

    println!("The value of x is: {}", x);
}

fn plus_one(x: u32) -> u32 {
    x + 1;
}

$ scarb cairo-run 
   Compiling no_listing_22_function_return_invalid v0.1.0 (listings/ch02-common-programming-concepts/no_listing_24_function_return_invalid/Scarb.toml)
error: Unexpected return type. Expected: "core::integer::u32", found: "()".
 --> listings/ch02-common-programming-concepts/no_listing_24_function_return_invalid/src/lib.cairo:9:28
fn plus_one(x: u32) -> u32 {
                           ^

error: could not compile `no_listing_22_function_return_invalid` due to previous error
error: `scarb metadata` exited with error

The main error message, Unexpected return type, reveals the core issue with this code. The definition of the function plus_one says that it will return an u32, but statements don’t evaluate to a value, which is expressed by (), the unit type. Therefore, nothing is returned, which contradicts the function definition and results in an error.

Comments

All programmers strive to make their code easy to understand, but sometimes extra explanation is warranted. In these cases, programmers leave comments in their source code that the compiler will ignore but people reading the source code may find useful.

Here’s a simple comment:

// hello, world

In Cairo, the idiomatic comment style starts a comment with two slashes, and the comment continues until the end of the line. For comments that extend beyond a single line, you’ll need to include // on each line, like this:

// So we’re doing something complicated here, long enough that we need
// multiple lines of comments to do it! Whew! Hopefully, this comment will
// explain what’s going on.

Comments can also be placed at the end of lines containing code:

fn main() -> felt252 {
    1 + 4 // return the sum of 1 and 4
}

But you’ll more often see them used in this format, with the comment on a separate line above the code it’s annotating:

fn main() -> felt252 {
    // this function performs a simple addition
    1 + 4
}

Item-level Documentation

Item-level documentation comments refer to specific items such as functions, implementations, traits, etc. They are prefixed with three slashes (///). These comments provide a detailed description of the item, examples of usage, and any conditions that might cause a panic. In case of functions, the comments may also include separate sections for parameter and return value descriptions.

/// Returns the sum of `arg1` and `arg2`.
/// `arg1` cannot be zero.
///
/// # Panics
///
/// This function will panic if `arg1` is `0`.
///
/// # Examples
///
/// ```
/// let a: felt252 = 2;
/// let b: felt252 = 3;
/// let c: felt252 = add(a, b);
/// assert(c == a + b, "Should equal a + b");
/// ```
fn add(arg1: felt252, arg2: felt252) -> felt252 {
    assert(arg1 != 0, 'Cannot be zero');
    arg1 + arg2
}

Module Documentation

Module documentation comments provide an overview of the entire module, including its purpose and examples of use. These comments are meant to be placed above the module they're describing and are prefixed with //!. This type of documentation gives a broad understanding of what the module does and how it can be used.

//! # my_module and implementation
//!
//! This is an example description of my_module and some of its features.
//!
//! # Examples
//!
//! ```
//! mod my_other_module {
//!   use path::to::my_module;
//!
//!   fn foo() {
//!     my_module.bar();
//!   }
//! }
//! ```
mod my_module { // rest of implementation...
}

Control Flow

The ability to run some code depending on whether a condition is true and to run some code repeatedly while a condition is true are basic building blocks in most programming languages. The most common constructs that let you control the flow of execution of Cairo code are if expressions and loops.

if Expressions

An if expression allows you to branch your code depending on conditions. You provide a condition and then state, “If this condition is met, run this block of code. If the condition is not met, do not run this block of code.”

Create a new project called branches in your cairo_projects directory to explore the if expression. In the src/lib.cairo file, input the following:

fn main() {
    let number = 3;

    if number == 5 {
        println!("condition was true and number = {}", number);
    } else {
        println!("condition was false and number = {}", number);
    }
}

All if expressions start with the keyword if, followed by a condition. In this case, the condition checks whether or not the variable number has a value equal to 5. We place the block of code to execute if the condition is true immediately after the condition inside curly brackets.

Optionally, we can also include an else expression, which we chose to do here, to give the program an alternative block of code to execute should the condition evaluate to false. If you don’t provide an else expression and the condition is false, the program will just skip the if block and move on to the next bit of code.

Try running this code; you should see the following output:

$ scarb cairo-run 
   Compiling no_listing_24_if v0.1.0 (listings/ch02-common-programming-concepts/no_listing_27_if/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_24_if
condition was false and number = 3
Run completed successfully, returning []

Let’s try changing the value of number to a value that makes the condition true to see what happens:

    let number = 5;

$ scarb cairo-run
condition was true and number = 5
Run completed successfully, returning []

It’s also worth noting that the condition in this code must be a bool. If the condition isn’t a bool, we’ll get an error. For example, try running the following code:

fn main() {
    let number = 3;

    if number {
        println!("number was three");
    }
}

The if condition evaluates to a value of 3 this time, and Cairo throws an error:

$ scarb build 
   Compiling no_listing_28_bis_if_not_bool v0.1.0 (listings/ch02-common-programming-concepts/no_listing_28_bis_if_not_bool/Scarb.toml)
error: Mismatched types. The type `core::bool` cannot be created from a numeric literal.
 --> listings/ch02-common-programming-concepts/no_listing_28_bis_if_not_bool/src/lib.cairo:4:18
    let number = 3;
                 ^

error: could not compile `no_listing_28_bis_if_not_bool` due to previous error

The error indicates that Cairo inferred the type of number to be a bool based on its later use as a condition of the if statement. It tries to create a bool from the value 3, but Cairo doesn't support instantiating a bool from a numeric literal anyway - you can only use true or false to create a bool. Unlike languages such as Ruby and JavaScript, Cairo will not automatically try to convert non-Boolean types to a Boolean. If we want the if code block to run only when a number is not equal to 0, for example, we can change the if expression to the following:

fn main() {
    let number = 3;

    if number != 0 {
        println!("number was something other than zero");
    }
}

Running this code will print number was something other than zero.

Handling Multiple Conditions with else if

You can use multiple conditions by combining if and else in an else if expression. For example:

fn main() {
    let number = 3;

    if number == 12 {
        println!("number is 12");
    } else if number == 3 {
        println!("number is 3");
    } else if number - 2 == 1 {
        println!("number minus 2 is 1");
    } else {
        println!("number not found");
    }
}

This program has four possible paths it can take. After running it, you should see the following output:

$ scarb cairo-run 
   Compiling no_listing_25_else_if v0.1.0 (listings/ch02-common-programming-concepts/no_listing_30_else_if/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running no_listing_25_else_if
number is 3
Run completed successfully, returning []

When this program executes, it checks each if expression in turn and executes the first body for which the condition evaluates to true. Note that even though number - 2 == 1 is true, we don’t see the output number minus 2 is 1 nor do we see the number not found text from the else block. That’s because Cairo only executes the block for the first true condition, and once it finds one, it doesn’t even check the rest. Using too many else if expressions can clutter your code, so if you have more than one, you might want to refactor your code. Chapter 6 describes a powerful Cairo branching construct called match for these cases.

Using if in a let Statement

Because if is an expression, we can use it on the right side of a let statement to assign the outcome to a variable.

fn main() {
    let condition = true;
    let number = if condition {
        5
    } else {
        6
    };

    if number == 5 {
        println!("condition was true and number is {}", number);
    }
}

$ scarb cairo-run 
   Compiling no_listing_26_if_let v0.1.0 (listings/ch02-common-programming-concepts/no_listing_31_if_let/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_26_if_let
condition was true and number is 5
Run completed successfully, returning []

The number variable will be bound to a value based on the outcome of the if expression, which will be 5 here.

Repetition with Loops

It’s often useful to execute a block of code more than once. For this task, Cairo provides a simple loop syntax, which will run through the code inside the loop body to the end and then start immediately back at the beginning. To experiment with loops, let’s create a new project called loops.

Cairo has three kinds of loops: loop, while, and for. Let’s try each one.

Repeating Code with loop

The loop keyword tells Cairo to execute a block of code over and over again forever or until you explicitly tell it to stop.

As an example, change the src/lib.cairo file in your loops directory to look like this:

fn main() {
    loop {
        println!("again!");
    }
}

When we run this program, we’ll see again! printed over and over continuously until either the program runs out of gas or we stop the program manually. Most terminals support the keyboard shortcut ctrl-c to interrupt a program that is stuck in a continual loop. Give it a try:

$ scarb cairo-run --available-gas=20000000
   Compiling loops v0.1.0 (file:///projects/loops)
    Finished release target(s) in 0 seconds
     Running loops
again!
again!
again!
^Cagain!

The symbol ^C represents where you pressed ctrl-c. You may or may not see the word again! printed after the ^C, depending on where the code was in the loop when it received the interrupt signal.

Note: Cairo prevents us from running program with infinite loops by including a gas meter. The gas meter is a mechanism that limits the amount of computation that can be done in a program. By setting a value to the --available-gas flag, we can set the maximum amount of gas available to the program. Gas is a unit of measurement that expresses the computation cost of an instruction. When the gas meter runs out, the program will stop. In the previous case, we set the gas limit high enough for the program to run for quite some time.

It is particularly important in the context of smart contracts deployed on Starknet, as it prevents from running infinite loops on the network. If you're writing a program that needs to run a loop, you will need to execute it with the --available-gas flag set to a value that is large enough to run the program.

Now, try running the same program again, but this time with the --available-gas flag set to 200000 instead of 2000000000000. You will see the program only prints again! 3 times before it stops, as it ran out of gas to keep executing the loop.

Fortunately, Cairo also provides a way to break out of a loop using code. You can place the break keyword within the loop to tell the program when to stop executing the loop.

fn main() {
    let mut i: usize = 0;
    loop {
        if i > 10 {
            break;
        }
        println!("i = {}", i);
        i += 1;
    }
}

The continue keyword tells the program to go to the next iteration of the loop and to skip the rest of the code in this iteration. Let's add a continue statement to our loop to skip the println! statement when i is equal to 5.

fn main() {
    let mut i: usize = 0;
    loop {
        if i > 10 {
            break;
        }
        if i == 5 {
            i += 1;
            continue;
        }
        println!("i = {}", i);
        i += 1;
    }
}

Executing this program will not print the value of i when it is equal to 5.

Returning Values from Loops

One of the uses of a loop is to retry an operation you know might fail, such as checking whether an operation has succeeded. You might also need to pass the result of that operation out of the loop to the rest of your code. To do this, you can add the value you want returned after the break expression you use to stop the loop; that value will be returned out of the loop so you can use it, as shown here:

fn main() {
    let mut counter = 0;

    let result = loop {
        if counter == 10 {
            break counter * 2;
        }
        counter += 1;
    };

    println!("The result is {}", result);
}

Before the loop, we declare a variable named counter and initialize it to 0. Then we declare a variable named result to hold the value returned from the loop. On every iteration of the loop, we check whether the counter is equal to 10, and then add 1 to the counter variable. When the condition is met, we use the break keyword with the value counter * 2. After the loop, we use a semicolon to end the statement that assigns the value to result. Finally, we print the value in result, which in this case is 20.

Conditional Loops with while

A program will often need to evaluate a condition within a loop. While the condition is true, the loop runs. When the condition ceases to be true, the program calls break, stopping the loop. It’s possible to implement behavior like this using a combination of loop, if, else, and break; you could try that now in a program, if you’d like. However, this pattern is so common that Cairo has a built-in language construct for it, called a while loop.

In Listing 2-2, we use while to loop the program three times, counting down each time after printing the value of number, and then, after the loop, print a message and exit.

fn main() {
    let mut number = 3;

    while number != 0 {
        println!("{number}!");
        number -= 1;
    };

    println!("LIFTOFF!!!");
}

Listing 2-2: Using a while loop to run code while a condition holds true.

This construct eliminates a lot of nesting that would be necessary if you used loop, if, else, and break, and it’s clearer. While a condition evaluates to true, the code runs; otherwise, it exits the loop.

Looping Through a Collection with for

You can also use the while construct to loop over the elements of a collection, such as an array. For example, the loop in Listing 2-3 prints each element in the array a.

fn main() {
    let a = [10, 20, 30, 40, 50].span();
    let mut index = 0;

    while index < 5 {
        println!("the value is: {}", a[index]);
        index += 1;
    }
}

Listing 2-3: Looping through each element of a collection using a while loop

Here, the code counts up through the elements in the array. It starts at index 0, and then loops until it reaches the final index in the array (that is, when index < 5 is no longer true). Running this code will print every element in the array:

$ scarb cairo-run 
   Compiling no_listing_45_iter_loop_while v0.1.0 (listings/ch02-common-programming-concepts/no_listing_45_iter_loop_while/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_45_iter_loop_while
the value is: 10
the value is: 20
the value is: 30
the value is: 40
the value is: 50
Run completed successfully, returning []

All five array values appear in the terminal, as expected. Even though index will reach a value of 5 at some point, the loop stops executing before trying to fetch a sixth value from the array.

However, this approach is error prone; we could cause the program to panic if the index value or test condition is incorrect. For example, if you changed the definition of the a array to have four elements but forgot to update the condition to while index < 4, the code would panic. It’s also slow, because the compiler adds runtime code to perform the conditional check of whether the index is within the bounds of the array on every iteration through the loop.

As a more concise alternative, you can use a for loop and execute some code for each item in a collection. A for loop looks like the code in Listing 2-4.

fn main() {
    let a = [10, 20, 30, 40, 50].span();

    for element in a {
        println!("the value is: {element}");
    }
}

Listing 2-4: Looping through each element of a collection using a for loop

When we run this code, we’ll see the same output as in Listing 2-3. More importantly, we’ve now increased the safety of the code and eliminated the chance of bugs that might result from going beyond the end of the array or not going far enough and missing some items.

Using the for loop, you wouldn’t need to remember to change any other code if you changed the number of values in the array, as you would with the method used in Listing 2-3.

The safety and conciseness of for loops make them the most commonly used loop construct in Cairo. Even in situations in which you want to run some code a certain number of times, as in the countdown example that used a while loop in Listing 2-2. Another way to run code a certain number of times would be to use a Range, provided by the core library, which generates all numbers in sequence starting from one number and ending before another number.

Here’s how you can use a Range to count from 1 to 3:

fn main() {
    for number in 1..4_u8 {
        println!("{number}!");
    };
    println!("Go!!!");
}

This code is a bit nicer, isn’t it?

Equivalence Between Loops and Recursive Functions

Loops and recursive functions are two common ways to repeat a block of code multiple times. The loop keyword is used to create an infinite loop that can be broken by using the break keyword.

fn main() -> felt252 {
    let mut x: felt252 = 0;
    loop {
        if x == 2 {
            break;
        } else {
            x += 1;
        }
    };
    x
}

Loops can be transformed into recursive functions by calling the function within itself. Here is an example of a recursive function that mimics the behavior of the loop example above.

fn main() -> felt252 {
    recursive_function(0)
}

fn recursive_function(mut x: felt252) -> felt252 {
    if x == 2 {
        x
    } else {
        recursive_function(x + 1)
    }
}

In both cases, the code block will run indefinitely until the condition x == 2 is met, at which point the value of x will be displayed.

In Cairo, loops and recursions are not only conceptually equivalent: they are also compiled down to similar low-level representations. To understand this, we can compile both examples to Sierra, and analyze the Sierra Code generated by the Cairo compiler for both examples. Add the following in your Scarb.toml file:

[lib]
sierra-text = true

Then, run scarb build to compile both examples. You will find the Sierra code generated by for both examples is extremely similar, as the loop is compiled to a recursive function in the Sierra statements.

Note: For our example, our findings came from understanding the statements section in Sierra that shows the execution traces of the two programs. If you are curious to learn more about Sierra, check out Exploring Sierra.

Summary

You made it! This was a sizable chapter: you learned about variables, data types, functions, comments, if expressions and loops! To practice with the concepts discussed in this chapter, try building programs to do the following:

• Generate the n-th Fibonacci number.
• Compute the factorial of a number n.

Now, we’ll review the common collection types in Cairo in the next chapter.

Common Collections

Cairo provides a set of common collection types that can be used to store and manipulate data. These collections are designed to be efficient, flexible, and easy to use. This section introduces the primary collection types available in Cairo: Arrays and Dictionaries.

Arrays

An array is a collection of elements of the same type. You can create and use array methods by using the ArrayTrait trait from the core library.

An important thing to note is that arrays have limited modification options. Arrays are, in fact, queues whose values can't be modified. This has to do with the fact that once a memory slot is written to, it cannot be overwritten, but only read from it. You can only append items to the end of an array and remove items from the front.

Creating an Array

Creating an array is done with the ArrayTrait::new() call. Here's an example of creating an array and appending 3 elements to it:

fn main() {
    let mut a = ArrayTrait::new();
    a.append(0);
    a.append(1);
    a.append(2);
}

When required, you can pass the expected type of items inside the array when instantiating the array like this, or explicitly define the type of the variable.

let mut arr = ArrayTrait::<u128>::new();

let mut arr:Array<u128> = ArrayTrait::new();

Updating an Array

Adding Elements

To add an element to the end of an array, you can use the append() method:

fn main() {
    let mut a = ArrayTrait::new();
    a.append(0);
    a.append(1);
    a.append(2);
}

Removing Elements

You can only remove elements from the front of an array by using the pop_front() method. This method returns an Option that can be unwrapped, containing the removed element, or Option::None if the array is empty.

fn main() {
    let mut a = ArrayTrait::new();
    a.append(10);
    a.append(1);
    a.append(2);

    let first_value = a.pop_front().unwrap();
    println!("The first value is {}", first_value);
}

The above code will print The first value is 10 as we remove the first element that was added.

In Cairo, memory is immutable, which means that it is not possible to modify the elements of an array once they've been added. You can only add elements to the end of an array and remove elements from the front of an array. These operations do not require memory mutation, as they involve updating pointers rather than directly modifying the memory cells.

Reading Elements from an Array

To access array elements, you can use get() or at() array methods that return different types. Using arr.at(index) is equivalent to using the subscripting operator arr[index].

get() Method

The get function returns an Option<Box<@T>>, which means it returns an option to a Box type (Cairo's smart-pointer type) containing a snapshot to the element at the specified index if that element exists in the array. If the element doesn't exist, get returns None. This method is useful when you expect to access indices that may not be within the array's bounds and want to handle such cases gracefully without panics. Snapshots will be explained in more detail in the "References and Snapshots" chapter.

Here is an example with the get() method:

fn main() -> u128 {
    let mut arr = ArrayTrait::<u128>::new();
    arr.append(100);
    let index_to_access =
        1; // Change this value to see different results, what would happen if the index doesn't exist?
    match arr.get(index_to_access) {
        Option::Some(x) => {
            *x
                .unbox() // Don't worry about * for now, if you are curious see Chapter 4.2 #desnap operator
            // It basically means "transform what get(idx) returned into a real value"
        },
        Option::None => { panic!("out of bounds") }
    }
}

at() Method

The at function, and its equivalent the subscripting operator, on the other hand, directly return a snapshot to the element at the specified index using the unbox() operator to extract the value stored in a box. If the index is out of bounds, a panic error occurs. You should only use at when you want the program to panic if the provided index is out of the array's bounds, which can prevent unexpected behavior.

fn main() {
    let mut a = ArrayTrait::new();
    a.append(0);
    a.append(1);

    // using `at()` method
    assert(*a.at(0) == 0, 'item mismatch on index 0');
    // using subscripting operator
    assert(*a[1] == 1, 'item mismatch on index 1');
}

In this example, the variable named first will get the value 0 because that is the value at index 0 in the array. The variable named second will get the value 1 from index 1 in the array.

In summary, use at when you want to panic on out-of-bounds access attempts, and use get when you prefer to handle such cases gracefully without panicking.

Size-related Methods

To determine the number of elements in an array, use the len() method. The return value is of type usize.

If you want to check if an array is empty or not, you can use the is_empty() method, which returns true if the array is empty and false otherwise.

array! Macro

Sometimes, we need to create arrays with values that are already known at compile time. The basic way of doing that is redundant. You would first declare the array and then append each value one by one. array! is a simpler way of doing this task by combining the two steps. At compile-time, the compiler will expand the macro to generate the code that appends the items sequentially.

Without array!:

    let mut arr = ArrayTrait::new();
    arr.append(1);
    arr.append(2);
    arr.append(3);
    arr.append(4);
    arr.append(5);

With array!:

    let arr = array![1, 2, 3, 4, 5];

Storing Multiple Types with Enums

If you want to store elements of different types in an array, you can use an Enum to define a custom data type that can hold multiple types. Enums will be explained in more detail in the "Enums and Pattern Matching" chapter.

#[derive(Copy, Drop)]
enum Data {
    Integer: u128,
    Felt: felt252,
    Tuple: (u32, u32),
}

fn main() {
    let mut messages: Array<Data> = array![];
    messages.append(Data::Integer(100));
    messages.append(Data::Felt('hello world'));
    messages.append(Data::Tuple((10, 30)));
}

Span

Span is a struct that represents a snapshot of an Array. It is designed to provide safe and controlled access to the elements of an array without modifying the original array. Span is particularly useful for ensuring data integrity and avoiding borrowing issues when passing arrays between functions or when performing read-only operations, as introduced in "References and Snapshots".

All methods provided by Array can also be used with Span, except for the append() method.

Turning an Array into Span

To create a Span of an Array, call the span() method:

fn main() {
    let mut array: Array<u8> = ArrayTrait::new();
    array.span();
}

Dictionaries

Cairo provides in its core library a dictionary-like type. The Felt252Dict<T> data type represents a collection of key-value pairs where each key is unique and associated with a corresponding value. This type of data structure is known differently across different programming languages such as maps, hash tables, associative arrays and many others.

The Felt252Dict<T> type is useful when you want to organize your data in a certain way for which using an Array<T> and indexing doesn't suffice. Cairo dictionaries also allow the programmer to easily simulate the existence of mutable memory when there is none.

Basic Use of Dictionaries

It is normal in other languages when creating a new dictionary to define the data types of both key and value. In Cairo, the key type is restricted to felt252, leaving only the possibility to specify the value data type, represented by T in Felt252Dict<T>.

The core functionality of a Felt252Dict<T> is implemented in the trait Felt252DictTrait which includes all basic operations. Among them we can find:

1. insert(felt252, T) -> () to write values to a dictionary instance and
2. get(felt252) -> T to read values from it.

These functions allow us to manipulate dictionaries like in any other language. In the following example, we create a dictionary to represent a mapping between individuals and their balance:

use core::dict::Felt252Dict;

fn main() {
    let mut balances: Felt252Dict<u64> = Default::default();

    balances.insert('Alex', 100);
    balances.insert('Maria', 200);

    let alex_balance = balances.get('Alex');
    assert!(alex_balance == 100, "Balance is not 100");

    let maria_balance = balances.get('Maria');
    assert!(maria_balance == 200, "Balance is not 200");
}

We can create a new instance of Felt252Dict<u64> by using the default method of the Default trait and add two individuals, each one with their own balance, using the insert method. Finally, we check the balance of our users with the get method. These methods are defined in the Felt252DictTrait trait in the core library.

Throughout the book we have talked about how Cairo's memory is immutable, meaning you can only write to a memory cell once but the Felt252Dict<T> type represents a way to overcome this obstacle. We will explain how this is implemented later on in "Dictionaries Underneath".

Building upon our previous example, let us show a code example where the balance of the same user changes:

use core::dict::Felt252Dict;

fn main() {
    let mut balances: Felt252Dict<u64> = Default::default();

    // Insert Alex with 100 balance
    balances.insert('Alex', 100);
    // Check that Alex has indeed 100 associated with him
    let alex_balance = balances.get('Alex');
    assert!(alex_balance == 100, "Alex balance is not 100");

    // Insert Alex again, this time with 200 balance
    balances.insert('Alex', 200);
    // Check the new balance is correct
    let alex_balance_2 = balances.get('Alex');
    assert!(alex_balance_2 == 200, "Alex balance is not 200");
}

Notice how in this example we added the 'Alex' individual twice, each time using a different balance and each time that we checked for its balance it had the last value inserted! Felt252Dict<T> effectively allows us to "rewrite" the stored value for any given key.

Before heading on and explaining how dictionaries are implemented it is worth mentioning that once you instantiate a Felt252Dict<T>, behind the scenes all keys have their associated values initialized as zero. This means that if for example, you tried to get the balance of an inexistent user you will get 0 instead of an error or an undefined value. This also means there is no way to delete data from a dictionary. Something to take into account when incorporating this structure into your code.

Until this point, we have seen all the basic features of Felt252Dict<T> and how it mimics the same behavior as the corresponding data structures in any other language, that is, externally of course. Cairo is at its core a non-deterministic Turing-complete programming language, very different from any other popular language in existence, which as a consequence means that dictionaries are implemented very differently as well!

In the following sections, we are going to give some insights about Felt252Dict<T> inner mechanisms and the compromises that were taken to make them work. After that, we are going to take a look at how to use dictionaries with other data structures as well as use the entry method as another way to interact with them.

Dictionaries Underneath

One of the constraints of Cairo's non-deterministic design is that its memory system is immutable, so in order to simulate mutability, the language implements Felt252Dict<T> as a list of entries. Each of the entries represents a time when a dictionary was accessed for reading/updating/writing purposes. An entry has three fields:

1. A key field that identifies the key for this key-value pair of the dictionary.
2. A previous_value field that indicates which previous value was held at key.
3. A new_value field that indicates the new value that is held at key.

If we try implementing Felt252Dict<T> using high-level structures we would internally define it as Array<Entry<T>> where each Entry<T> has information about what key-value pair it represents and the previous and new values it holds. The definition of Entry<T> would be:

struct Entry<T> {
    key: felt252,
    previous_value: T,
    new_value: T,
}

For each time we interact with a Felt252Dict<T>, a new Entry<T> will be registered:

• A get would register an entry where there is no change in state, and previous and new values are stored with the same value.
• An insert would register a new Entry<T> where the new_value would be the element being inserted, and the previous_value the last element inserted before this. In case it is the first entry for a certain key, then the previous value will be zero.

The use of this entry list shows how there isn't any rewriting, just the creation of new memory cells per Felt252Dict<T> interaction. Let's show an example of this using the balances dictionary from the previous section and inserting the users 'Alex' and 'Maria':

use core::dict::Felt252Dict;

struct Entry<T> {
    key: felt252,
    previous_value: T,
    new_value: T,
}

fn main() {
    let mut balances: Felt252Dict<u64> = Default::default();
    balances.insert('Alex', 100_u64);
    balances.insert('Maria', 50_u64);
    balances.insert('Alex', 200_u64);
    balances.get('Maria');
}

These instructions would then produce the following list of entries:

Notice that since 'Alex' was inserted twice, it appears twice and the previous and current values are set properly. Also reading from 'Maria' registered an entry with no change from previous to current values.

This approach to implementing Felt252Dict<T> means that for each read/write operation, there is a scan for the whole entry list in search of the last entry with the same key. Once the entry has been found, its new_value is extracted and used on the new entry to be added as the previous_value. This means that interacting with Felt252Dict<T> has a worst-case time complexity of O(n) where n is the number of entries in the list.

If you pour some thought into alternate ways of implementing Felt252Dict<T> you'd surely find them, probably even ditching completely the need for a previous_value field, nonetheless, since Cairo is not your normal language this won't work. One of the purposes of Cairo is, with the STARK proof system, to generate proofs of computational integrity. This means that you need to verify that program execution is correct and inside the boundaries of Cairo restrictions. One of those boundary checks consists of "dictionary squashing" and that requires information on both previous and new values for every entry.

Squashing Dictionaries

To verify that the proof generated by a Cairo program execution that used a Felt252Dict<T> is correct, we need to check that there wasn't any illegal tampering with the dictionary. This is done through a method called squash_dict that reviews each entry of the entry list and checks that access to the dictionary remains coherent throughout the execution.

The process of squashing is as follows: given all entries with certain key k, taken in the same order as they were inserted, verify that the ith entry new_value is equal to the ith + 1 entry previous_value.

For example, given the following entry list:

After squashing, the entry list would be reduced to:

In case of a change on any of the values of the first table, squashing would have failed during runtime.

Dictionary Destruction

If you run the examples from "Basic Use of Dictionaries" section, you'd notice that there was never a call to squash dictionary, but the program compiled successfully nonetheless. What happened behind the scene was that squash was called automatically via the Felt252Dict<T> implementation of the Destruct<T> trait. This call occurred just before the balance dictionary went out of scope.

The Destruct<T> trait represents another way of removing instances out of scope apart from Drop<T>. The main difference between these two is that Drop<T> is treated as a no-op operation, meaning it does not generate new CASM while Destruct<T> does not have this restriction. The only type which actively uses the Destruct<T> trait is Felt252Dict<T>, for every other type Destruct<T> and Drop<T> are synonyms. You can read more about these traits in Drop and Destruct section of Appendix C.

Later in "Dictionaries as Struct Members" section, we will have a hands-on example where we implement the Destruct<T> trait for a custom type.

More Dictionaries

Up to this point, we have given a comprehensive overview of the functionality of Felt252Dict<T> as well as how and why it is implemented in a certain way. If you haven't understood all of it, don't worry because in this section we will have some more examples using dictionaries.

We will start by explaining the entry method which is part of a dictionary basic functionality included in Felt252DictTrait<T> which we didn't mention at the beginning. Soon after, we will see examples of how to use Felt252Dict<T> with other complex types such as Array<T>.

Entry and Finalize

In the "Dictionaries Underneath" section, we explained how Felt252Dict<T> internally worked. It was a list of entries for each time the dictionary was accessed in any manner. It would first find the last entry given a certain key and then update it accordingly to whatever operation it was executing. The Cairo language gives us the tools to replicate this ourselves through the entry and finalize methods.

The entry method comes as part of Felt252DictTrait<T> with the purpose of creating a new entry given a certain key. Once called, this method takes ownership of the dictionary and returns the entry to update. The method signature is as follows:

fn entry(self: Felt252Dict<T>, key: felt252) -> (Felt252DictEntry<T>, T) nopanic

The first input parameter takes ownership of the dictionary while the second one is used to create the appropriate entry. It returns a tuple containing a Felt252DictEntry<T>, which is the type used by Cairo to represent dictionary entries, and a T representing the value held previously. The nopanic notation simply indicates that the function is guaranteed to never panic.

The next thing to do is to update the entry with the new value. For this, we use the finalize method which inserts the entry and returns ownership of the dictionary:

fn finalize(self: Felt252DictEntry<T>, new_value: T) -> Felt252Dict<T>

This method receives the entry and the new value as parameters, and returns the updated dictionary.

Let us see an example using entry and finalize. Imagine we would like to implement our own version of the get method from a dictionary. We should then do the following:

1. Create the new entry to add using the entry method.
2. Insert back the entry where the new_value equals the previous_value.
3. Return the value.

Implementing our custom get would look like this:

use core::dict::{Felt252Dict, Felt252DictEntryTrait};

fn custom_get<T, +Felt252DictValue<T>, +Drop<T>, +Copy<T>>(
    ref dict: Felt252Dict<T>, key: felt252
) -> T {
    // Get the new entry and the previous value held at `key`
    let (entry, prev_value) = dict.entry(key);

    // Store the value to return
    let return_value = prev_value;

    // Update the entry with `prev_value` and get back ownership of the dictionary
    dict = entry.finalize(prev_value);

    // Return the read value
    return_value
}

The ref keyword means that the ownership of the variable will be given back at the end of the function. This concept will be explained in more detail in the "References and Snapshots" section.

Implementing the insert method would follow a similar workflow, except for inserting a new value when finalizing. If we were to implement it, it would look like the following:

use core::dict::{Felt252Dict, Felt252DictEntryTrait};

fn custom_insert<T, +Felt252DictValue<T>, +Destruct<T>, +Drop<T>>(
    ref dict: Felt252Dict<T>, key: felt252, value: T
) {
    // Get the last entry associated with `key`
    // Notice that if `key` does not exist, `_prev_value` will
    // be the default value of T.
    let (entry, _prev_value) = dict.entry(key);

    // Insert `entry` back in the dictionary with the updated value,
    // and receive ownership of the dictionary
    dict = entry.finalize(value);
}

As a finalizing note, these two methods are implemented in a similar way to how insert and get are implemented for Felt252Dict<T>. This code shows some example usage:

use core::dict::{Felt252Dict, Felt252DictEntryTrait};

fn custom_get<T, +Felt252DictValue<T>, +Drop<T>, +Copy<T>>(
    ref dict: Felt252Dict<T>, key: felt252
) -> T {
    // Get the new entry and the previous value held at `key`
    let (entry, prev_value) = dict.entry(key);

    // Store the value to return
    let return_value = prev_value;

    // Update the entry with `prev_value` and get back ownership of the dictionary
    dict = entry.finalize(prev_value);

    // Return the read value
    return_value
}

fn custom_insert<T, +Felt252DictValue<T>, +Destruct<T>, +Drop<T>>(
    ref dict: Felt252Dict<T>, key: felt252, value: T
) {
    // Get the last entry associated with `key`
    // Notice that if `key` does not exist, `_prev_value` will
    // be the default value of T.
    let (entry, _prev_value) = dict.entry(key);

    // Insert `entry` back in the dictionary with the updated value,
    // and receive ownership of the dictionary
    dict = entry.finalize(value);
}

fn main() {
    let mut dict: Felt252Dict<u64> = Default::default();

    custom_insert(ref dict, '0', 100);

    let val = custom_get(ref dict, '0');

    assert!(val == 100, "Expecting 100");
}

Dictionaries of Types not Supported Natively

One restriction of Felt252Dict<T> that we haven't talked about is the trait Felt252DictValue<T>. This trait defines the zero_default method which is the one that gets called when a value does not exist in the dictionary. This is implemented by some common data types, such as most unsigned integers, bool and felt252 - but it is not implemented for more complex types such as arrays, structs (including u256), and other types from the core library. This means that making a dictionary of types not natively supported is not a straightforward task, because you would need to write a couple of trait implementations in order to make the data type a valid dictionary value type. To compensate this, you can wrap your type inside a Nullable<T>.

Nullable<T> is a smart pointer type that can either point to a value or be null in the absence of value. It is usually used in Object Oriented Programming Languages when a reference doesn't point anywhere. The difference with Option is that the wrapped value is stored inside a Box<T> data type. The Box<T> type is a smart pointer that allows us to use a dedicated boxed_segment memory segment for our data, and access this segment using a pointer that can only be manipulated in one place at a time. See Smart Pointers Chapter for more information.

Let's show using an example. We will try to store a Span<felt252> inside a dictionary. For that, we will use Nullable<T> and Box<T>. Also, we are storing a Span<T> and not an Array<T> because the latter does not implement the Copy<T> trait which is required for reading from a dictionary.

use core::dict::Felt252Dict;
use core::nullable::{NullableTrait, match_nullable, FromNullableResult};

fn main() {
    // Create the dictionary
    let mut d: Felt252Dict<Nullable<Span<felt252>>> = Default::default();

    // Create the array to insert
    let a = array![8, 9, 10];

    // Insert it as a `Span`
    d.insert(0, NullableTrait::new(a.span()));

//...

In this code snippet, the first thing we did was to create a new dictionary d. We want it to hold a Nullable<Span>. After that, we created an array and filled it with values.

The last step is inserting the array as a span inside the dictionary. Notice that we do this using the new function of the NullableTrait.

Once the element is inside the dictionary, and we want to get it, we follow the same steps but in reverse order. The following code shows how to achieve that:

//...

    // Get value back
    let val = d.get(0);

    // Search the value and assert it is not null
    let span = match match_nullable(val) {
        FromNullableResult::Null => panic!("No value found"),
        FromNullableResult::NotNull(val) => val.unbox(),
    };

    // Verify we are having the right values
    assert!(*span.at(0) == 8, "Expecting 8");
    assert!(*span.at(1) == 9, "Expecting 9");
    assert!(*span.at(2) == 10, "Expecting 10");
}

Here we:

1. Read the value using get.
2. Verified it is non-null using the match_nullable function.
3. Unwrapped the value inside the box and asserted it was correct.

The complete script would look like this:

use core::dict::Felt252Dict;
use core::nullable::{NullableTrait, match_nullable, FromNullableResult};

fn main() {
    // Create the dictionary
    let mut d: Felt252Dict<Nullable<Span<felt252>>> = Default::default();

    // Create the array to insert
    let a = array![8, 9, 10];

    // Insert it as a `Span`
    d.insert(0, NullableTrait::new(a.span()));

    // Get value back
    let val = d.get(0);

    // Search the value and assert it is not null
    let span = match match_nullable(val) {
        FromNullableResult::Null => panic!("No value found"),
        FromNullableResult::NotNull(val) => val.unbox(),
    };

    // Verify we are having the right values
    assert!(*span.at(0) == 8, "Expecting 8");
    assert!(*span.at(1) == 9, "Expecting 9");
    assert!(*span.at(2) == 10, "Expecting 10");
}

Using Arrays inside Dictionaries

In the previous section, we explored how to store and retrieve complex types inside a dictionary using Nullable<T> and Box<T>. Now, let's take a look at how to store an array inside a dictionary and dynamically modify its contents.

Storing arrays in dictionaries in Cairo is slightly different from storing other types. This is because arrays are more complex data structures that require special handling to avoid issues with memory copying and references.

First, let's look at how to create a dictionary and insert an array into it. This process is pretty straightforward and follows a similar pattern to inserting other types of data:

use core::dict::Felt252Dict;

fn main() {
    let arr = array![20, 19, 26];
    let mut dict: Felt252Dict<Nullable<Array<u8>>> = Default::default();
    dict.insert(0, NullableTrait::new(arr));
    println!("Array inserted successfully.");
}

However, attempting to read an array from the dictionary using the get method will result in a compiler error. This is because get tries to copy the array in memory, which is not possible for arrays (as we've already mentioned in the previous section, Array<T> does not implement the Copy<T> trait):

use core::nullable::{match_nullable, FromNullableResult};
use core::dict::Felt252Dict;

fn main() {
    let arr = array![20, 19, 26];
    let mut dict: Felt252Dict<Nullable<Array<u8>>> = Default::default();
    dict.insert(0, NullableTrait::new(arr));
    println!("Array: {:?}", get_array_entry(ref dict, 0));
}

fn get_array_entry(ref dict: Felt252Dict<Nullable<Array<u8>>>, index: felt252) -> Span<u8> {
    let val = dict.get(0); // This will cause a compiler error
    let arr = match match_nullable(val) {
        FromNullableResult::Null => panic!("No value!"),
        FromNullableResult::NotNull(val) => val.unbox()
    };
    arr.span()
}

$ scarb cairo-run 
   Compiling no_listing_15_dict_of_array_attempt_get v0.1.0 (listings/ch03-common-collections/no_listing_15_dict_of_array_attempt_get/Scarb.toml)
error: Trait has no implementation in context: core::traits::Copy::<core::nullable::Nullable::<core::array::Array::<core::integer::u8>>>.
 --> listings/ch03-common-collections/no_listing_15_dict_of_array_attempt_get/src/lib.cairo:13:20
    let val = dict.get(0); // This will cause a compiler error
                   ^*^

error: could not compile `no_listing_15_dict_of_array_attempt_get` due to previous error
error: `scarb metadata` exited with error

To correctly read an array from the dictionary, we need to use dictionary entries. This allows us to get a reference to the array value without copying it:

fn get_array_entry(ref dict: Felt252Dict<Nullable<Array<u8>>>, index: felt252) -> Span<u8> {
    let (entry, _arr) = dict.entry(index);
    let mut arr = _arr.deref_or(array![]);
    let span = arr.span();
    dict = entry.finalize(NullableTrait::new(arr));
    span
}

Note: We must convert the array to a Span before finalizing the entry, because calling NullableTrait::new(arr) moves the array, thus making it impossible to return it from the function.

To modify the stored array, such as appending a new value, we can use a similar approach. The following append_value function demonstrates this:

fn append_value(ref dict: Felt252Dict<Nullable<Array<u8>>>, index: felt252, value: u8) {
    let (entry, arr) = dict.entry(index);
    let mut unboxed_val = arr.deref_or(array![]);
    unboxed_val.append(value);
    dict = entry.finalize(NullableTrait::new(unboxed_val));
}

In the append_value function, we access the dictionary entry, dereference the array, append the new value, and finalize the entry with the updated array.

Note: Removing an item from a stored array can be implemented in a similar manner.

Below is the complete example demonstrating the creation, insertion, reading, and modification of an array in a dictionary:

use core::nullable::NullableTrait;
use core::dict::{Felt252Dict, Felt252DictEntryTrait};

fn append_value(ref dict: Felt252Dict<Nullable<Array<u8>>>, index: felt252, value: u8) {
    let (entry, arr) = dict.entry(index);
    let mut unboxed_val = arr.deref_or(array![]);
    unboxed_val.append(value);
    dict = entry.finalize(NullableTrait::new(unboxed_val));
}

fn get_array_entry(ref dict: Felt252Dict<Nullable<Array<u8>>>, index: felt252) -> Span<u8> {
    let (entry, _arr) = dict.entry(index);
    let mut arr = _arr.deref_or(array![]);
    let span = arr.span();
    dict = entry.finalize(NullableTrait::new(arr));
    span
}

fn main() {
    let arr = array![20, 19, 26];
    let mut dict: Felt252Dict<Nullable<Array<u8>>> = Default::default();
    dict.insert(0, NullableTrait::new(arr));
    println!("Before insertion: {:?}", get_array_entry(ref dict, 0));

    append_value(ref dict, 0, 30);

    println!("After insertion: {:?}", get_array_entry(ref dict, 0));
}

Understanding Cairo's Ownership system

Cairo is a language built around a linear type system that allows us to statically ensure that in every Cairo program, a value is used exactly once. This linear type system helps prevent runtime errors by ensuring that operations that could cause such errors, such as writing twice to a memory cell, are detected at compile time. This is achieved by implementing an ownership system and forbidding copying and dropping values by default. In this chapter, we’ll talk about Cairo's ownership system as well as references and snapshots.

Ownership Using a Linear Type System

Cairo uses a linear type system. In such a type system, any value (a basic type, a struct, an enum) must be used and must only be used once. 'Used' here means that the value is either destroyed or moved.

Destruction can happen in several ways:

• a variable goes out of scope.
• a struct is destructured.
• explicit destruction using destruct().

Moving a value simply means passing that value to another function.

This results in somewhat similar constraints to the Rust ownership model, but there are some differences. In particular, the Rust ownership model exists (in part) to avoid data races and concurrent mutable access to a memory value. This is obviously impossible in Cairo since the memory is immutable. Instead, Cairo leverages its linear type system for two main purposes:

• Ensuring that all code is provable and thus verifiable.
• Abstracting away the immutable memory of the Cairo VM.

Ownership

In Cairo, ownership applies to variables and not to values. A value can safely be referred to by many different variables (even if they are mutable variables), as the value itself is always immutable. Variables however can be mutable, so the compiler must ensure that constant variables aren't accidentally modified by the programmer. This makes it possible to talk about ownership of a variable: the owner is the code that can read (and write if mutable) the variable.

This means that variables (not values) follow similar rules to Rust values:

• Each variable in Cairo has an owner.
• There can only be one owner at a time.
• When the owner goes out of scope, the variable is destroyed.

Now that we’re past basic Cairo syntax, we won’t include all the fn main() { code in examples, so if you’re following along, make sure to put the following examples inside a main function manually. As a result, our examples will be a bit more concise, letting us focus on the actual details rather than boilerplate code.

Variable Scope

As a first example of the linear type system, we’ll look at the scope of some variables. A scope is the range within a program for which an item is valid. Take the following variable:

let s = 'hello';

The variable s refers to a short string. The variable is valid from the point at which it’s declared until the end of the current scope. Listing 4-1 shows a program with comments annotating where the variable s would be valid.

//TAG: ignore_fmt
fn main() {
    { // s is not valid here, it’s not yet declared
        let s = 'hello'; // s is valid from this point forward
        // do stuff with s
    } // this scope is now over, and s is no longer valid
}

Listing 4-1: A variable and the scope in which it is valid

In other words, there are two important points in time here:

• When s comes into scope, it is valid.
• It remains valid until it goes out of scope.

At this point, the relationship between scopes and when variables are valid is similar to that in other programming languages. Now we’ll build on top of this understanding by using the Array type we introduced in the previous "Arrays" section.

Moving values

As said earlier, moving a value simply means passing that value to another function. When that happens, the variable referring to that value in the original scope is destroyed and can no longer be used, and a new variable is created to hold the same value.

Arrays are an example of a complex type that is moved when passing it to another function. Here is a short reminder of what an array looks like:

fn main() {
    let mut arr: Array<u128> = array![];
    arr.append(1);
    arr.append(2);
}

How does the type system ensure that the Cairo program never tries to write to the same memory cell twice? Consider the following code, where we try to remove the front of the array twice:

fn foo(mut arr: Array<u128>) {
    arr.pop_front();
}

fn main() {
    let arr: Array<u128> = array![];
    foo(arr);
    foo(arr);
}

In this case, we try to pass the same value (the array in the arr variable) to both function calls. This means our code tries to remove the first element twice, which would try to write to the same memory cell twice - which is forbidden by the Cairo VM, leading to a runtime error. Thankfully, this code does not actually compile. Once we have passed the array to the foo function, the variable arr is no longer usable. We get this compile-time error, telling us that we would need Array to implement the Copy Trait:

$ scarb cairo-run 
   Compiling no_listing_02_pass_array_by_value v0.1.0 (listings/ch04-understanding-ownership/no_listing_02_pass_array_by_value/Scarb.toml)
warn: Unhandled `#[must_use]` type `core::option::Option::<core::integer::u128>`
 --> listings/ch04-understanding-ownership/no_listing_02_pass_array_by_value/src/lib.cairo:3:5
    arr.pop_front();
    ^*************^

error: Variable was previously moved.
 --> listings/ch04-understanding-ownership/no_listing_02_pass_array_by_value/src/lib.cairo:9:9
    foo(arr);
        ^*^
note: variable was previously used here:
  --> listings/ch04-understanding-ownership/no_listing_02_pass_array_by_value/src/lib.cairo:8:9
    foo(arr);
        ^*^
note: Trait has no implementation in context: core::traits::Copy::<core::array::Array::<core::integer::u128>>.

error: could not compile `no_listing_02_pass_array_by_value` due to previous error
error: `scarb metadata` exited with error

The Copy Trait

The Copy trait allows simple types to be duplicated by copying felts, without allocating new memory segments. This contrasts with Cairo's default "move" semantics, which transfer ownership of values to ensure memory safety and prevent issues like multiple writes to the same memory cell. Copy is implemented for types where duplication is safe and efficient, bypassing the need for move semantics. Types like Array and Felt252Dict cannot implement Copy, as manipulating them in different scopes is forbidden by the type system.

All basic types previously described in "Data Types" implement by default the Copy trait.

While Arrays and Dictionaries can't be copied, custom types that don't contain either of them can be. You can implement the Copy trait on your type by adding the #[derive(Copy)] annotation to your type definition. However, Cairo won't allow a type to be annotated with Copy if the type itself or any of its components doesn't implement the Copy trait.

#[derive(Copy, Drop)]
struct Point {
    x: u128,
    y: u128,
}

fn main() {
    let p1 = Point { x: 5, y: 10 };
    foo(p1);
    foo(p1);
}

fn foo(p: Point) { // do something with p
}

In this example, we can pass p1 twice to the foo function because the Point type implements the Copy trait. This means that when we pass p1 to foo, we are actually passing a copy of p1, so p1 remains valid. In ownership terms, this means that the ownership of p1 remains with the main function. If you remove the Copy trait derivation from the Point type, you will get a compile-time error when trying to compile the code.

Don't worry about the Struct keyword. We will introduce this in Chapter 5.

Destroying Values - Example with FeltDict

The other way linear types can be used is by being destroyed. Destruction must ensure that the 'resource' is now correctly released. In Rust, for example, this could be closing the access to a file, or locking a mutex. In Cairo, one type that has such behaviour is Felt252Dict. For provability, dicts must be 'squashed' when they are destructed. This would be very easy to forget, so it is enforced by the type system and the compiler.

No-op Destruction: the Drop Trait

You may have noticed that the Point type in the previous example also implements the Drop trait. For example, the following code will not compile, because the struct A is not moved or destroyed before it goes out of scope:

struct A {}

fn main() {
    A {}; // error: Variable not dropped.
}

However, types that implement the Drop trait are automatically destroyed when going out of scope. This destruction does nothing, it is a no-op - simply a hint to the compiler that this type can safely be destroyed once it's no longer useful. We call this "dropping" a value.

At the moment, the Drop implementation can be derived for all types, allowing them to be dropped when going out of scope, except for dictionaries (Felt252Dict) and types containing dictionaries. For example, the following code compiles:

#[derive(Drop)]
struct A {}

fn main() {
    A {}; // Now there is no error.
}

Destruction with a Side-effect: the Destruct Trait

When a value is destroyed, the compiler first tries to call the drop method on that type. If it doesn't exist, then the compiler tries to call destruct instead. This method is provided by the Destruct trait.

As said earlier, dictionaries in Cairo are types that must be "squashed" when destructed, so that the sequence of access can be proven. This is easy for developers to forget, so instead dictionaries implement the Destruct trait to ensure that all dictionaries are squashed when going out of scope. As such, the following example will not compile:

use core::dict::Felt252Dict;

struct A {
    dict: Felt252Dict<u128>
}

fn main() {
    A { dict: Default::default() };
}

If you try to run this code, you will get a compile-time error:

$ scarb cairo-run 
   Compiling no_listing_06_no_destruct_compile_fails v0.1.0 (listings/ch04-understanding-ownership/no_listing_06_no_destruct_compile_fails/Scarb.toml)
error: Variable not dropped.
 --> listings/ch04-understanding-ownership/no_listing_06_no_destruct_compile_fails/src/lib.cairo:9:5
    A { dict: Default::default() };
    ^****************************^
note: Trait has no implementation in context: core::traits::Drop::<no_listing_06_no_destruct_compile_fails::A>.
note: Trait has no implementation in context: core::traits::Destruct::<no_listing_06_no_destruct_compile_fails::A>.

error: could not compile `no_listing_06_no_destruct_compile_fails` due to previous error
error: `scarb metadata` exited with error

When A goes out of scope, it can't be dropped as it implements neither the Drop (as it contains a dictionary and can't derive(Drop)) nor the Destruct trait. To fix this, we can derive the Destruct trait implementation for the A type:

use core::dict::Felt252Dict;

#[derive(Destruct)]
struct A {
    dict: Felt252Dict<u128>
}

fn main() {
    A { dict: Default::default() }; // No error here
}

Now, when A goes out of scope, its dictionary will be automatically squashed, and the program will compile.

Copy Array Data with clone

If we do want to deeply copy the data of an Array, we can use a common method called clone. We’ll discuss method syntax in a dedicated section in Chapter 5, but because methods are a common feature in many programming languages, you’ve probably seen them before.

Here’s an example of the clone method in action.

fn main() {
    let arr1: Array<u128> = array![];
    let arr2 = arr1.clone();
}

When you see a call to clone, you know that some arbitrary code is being executed and that code may be expensive. It’s a visual indicator that something different is going on. In this case, the value arr1 refers to is being copied, resulting in new memory cells being used, and a new variable arr2 is created, referring to the new copied value.

Return Values and Scope

Returning values is equivalent to moving them. Listing 4-2 shows an example of a function that returns some value, with similar annotations as those in Listing 4-1.

Filename: src/lib.cairo

#[derive(Drop)]
struct A {}

fn main() {
    let a1 = gives_ownership();           // gives_ownership moves its return
                                          // value into a1

    let a2 = A {};                        // a2 comes into scope

    let a3 = takes_and_gives_back(a2);    // a2 is moved into
                                          // takes_and_gives_back, which also
                                          // moves its return value into a3

} // Here, a3 goes out of scope and is dropped. a2 was moved, so nothing
  // happens. a1 goes out of scope and is dropped.

fn gives_ownership() -> A {               // gives_ownership will move its
                                          // return value into the function
                                          // that calls it

    let some_a = A {};                    // some_a comes into scope

    some_a                                // some_a is returned and
                                          // moves ownership to the calling
                                          // function
}

// This function takes an instance some_a of A and returns it
fn takes_and_gives_back(some_a: A) -> A { // some_a comes into scope

    some_a                                // some_a is returned and 
                                          // moves ownership to the calling
                                          // function
}

Listing 4-2: Moving return values

While this works, moving into and out of every function is a bit tedious. What if we want to let a function use a value but not move the value? It’s quite annoying that anything we pass in also needs to be passed back if we want to use it again, in addition to any data resulting from the body of the function that we might want to return as well.

Cairo does let us return multiple values using a tuple, as shown in Listing 4-3.

Filename: src/lib.cairo

fn main() {
    let arr1: Array<u128> = array![];

    let (arr2, len) = calculate_length(arr1);
}

fn calculate_length(arr: Array<u128>) -> (Array<u128>, usize) {
    let length = arr.len(); // len() returns the length of an array

    (arr, length)
}

Listing 4-3: Returning many values

But this is too much ceremony and a lot of work for a concept that should be common. Luckily for us, Cairo has two features for passing a value without destroying or moving it, called references and snapshots.

References and Snapshots

The issue with the tuple code in previous Listing 4-3 is that we have to return the Array to the calling function so we can still use the Array after the call to calculate_length, because the Array was moved into calculate_length.

Snapshots

In the previous chapter, we talked about how Cairo's ownership system prevents us from using a variable after we've moved it, protecting us from potentially writing twice to the same memory cell. However, it's not very convenient. Let's see how we can retain ownership of the variable in the calling function using snapshots.

In Cairo, a snapshot is an immutable view of a value at a certain point in time. Recall that memory is immutable, so modifying a value actually creates a new memory cell. The old memory cell still exists, and snapshots are variables that refer to that "old" value. In this sense, snapshots are a view "into the past".

Here is how you would define and use a calculate_length function that takes a snapshot of an array as a parameter instead of taking ownership of the underlying value. In this example, the calculate_length function returns the length of the array passed as a parameter. As we're passing it as a snapshot, which is an immutable view of the array, we can be sure that the calculate_length function will not mutate the array, and ownership of the array is kept in the main function.

Filename: src/lib.cairo

fn main() {
    let mut arr1: Array<u128> = array![];
    let first_snapshot = @arr1; // Take a snapshot of `arr1` at this point in time
    arr1.append(1); // Mutate `arr1` by appending a value
    let first_length = calculate_length(
        first_snapshot
    ); // Calculate the length of the array when the snapshot was taken
    let second_length = calculate_length(@arr1); // Calculate the current length of the array
    println!("The length of the array when the snapshot was taken is {}", first_length);
    println!("The current length of the array is {}", second_length);
}

fn calculate_length(arr: @Array<u128>) -> usize {
    arr.len()
}

Note: it is only possible to call the len() method on an array snapshot because it is defined as such in the ArrayTrait trait. If you try to call a method that is not defined for snapshots on a snapshot, you will get a compilation error. However, you can call methods expecting a snapshot on non-snapshot types.

The output of this program is:

$ scarb cairo-run 
   Compiling no_listing_09_snapshots v0.1.0 (listings/ch04-understanding-ownership/no_listing_09_snapshots/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running no_listing_09_snapshots
The length of the array when the snapshot was taken is 0
The current length of the array is 1
Run completed successfully, returning []

First, notice that all the tuple code in the variable declaration and the function return value is gone. Second, note that we pass @arr1 into calculate_length and, in its definition, we take @Array<u128> rather than Array<u128>.

Let’s take a closer look at the function call here:

let second_length = calculate_length(@arr1); // Calculate the current length of the array

The @arr1 syntax lets us create a snapshot of the value in arr1. Because a snapshot is an immutable view of a value at a specific point in time, the usual rules of the linear type system are not enforced. In particular, snapshot variables always implement the Drop trait, never the Destruct trait, even dictionary snapshots.

Similarly, the signature of the function uses @ to indicate that the type of the parameter arr is a snapshot. Let’s add some explanatory annotations:

fn calculate_length(
    array_snapshot: @Array<u128> // array_snapshot is a snapshot of an Array
) -> usize {
    array_snapshot.len()
} // Here, array_snapshot goes out of scope and is dropped.
// However, because it is only a view of what the original array `arr` contains, the original `arr` can still be used.

The scope in which the variable array_snapshot is valid is the same as any function parameter’s scope, but the underlying value of the snapshot is not dropped when array_snapshot stops being used. When functions have snapshots as parameters instead of the actual values, we won’t need to return the values in order to give back ownership of the original value, because we never had it.

Desnap Operator

To convert a snapshot back into a regular variable, you can use the desnap operator *, which serves as the opposite of the @ operator.

Only Copy types can be desnapped. However, in the general case, because the value is not modified, the new variable created by the desnap operator reuses the old value, and so desnapping is a completely free operation, just like Copy.

In the following example, we want to calculate the area of a rectangle, but we don't want to take ownership of the rectangle in the calculate_area function, because we might want to use the rectangle again after the function call. Since our function doesn't mutate the rectangle instance, we can pass the snapshot of the rectangle to the function, and then transform the snapshots back into values using the desnap operator *.

#[derive(Drop)]
struct Rectangle {
    height: u64,
    width: u64,
}

fn main() {
    let rec = Rectangle { height: 3, width: 10 };
    let area = calculate_area(@rec);
    println!("Area: {}", area);
}

fn calculate_area(rec: @Rectangle) -> u64 {
    // As rec is a snapshot to a Rectangle, its fields are also snapshots of the fields types.
    // We need to transform the snapshots back into values using the desnap operator `*`.
    // This is only possible if the type is copyable, which is the case for u64.
    // Here, `*` is used for both multiplying the height and width and for desnapping the snapshots.
    *rec.height * *rec.width
}

But, what happens if we try to modify something we’re passing as a snapshot? Try the code in Listing 4-4. Spoiler alert: it doesn’t work!

Filename: src/lib.cairo

#[derive(Copy, Drop)]
struct Rectangle {
    height: u64,
    width: u64,
}

fn main() {
    let rec = Rectangle { height: 3, width: 10 };
    flip(@rec);
}

fn flip(rec: @Rectangle) {
    let temp = rec.height;
    rec.height = rec.width;
    rec.width = temp;
}

Listing 4-4: Attempting to modify a snapshot value

Here’s the error:

$ scarb cairo-run 
   Compiling listing_04_04 v0.1.0 (listings/ch04-understanding-ownership/listing_04_attempt_modifying_snapshot/Scarb.toml)
error: Invalid left-hand side of assignment.
 --> listings/ch04-understanding-ownership/listing_04_attempt_modifying_snapshot/src/lib.cairo:15:5
    rec.height = rec.width;
    ^********^

error: Invalid left-hand side of assignment.
 --> listings/ch04-understanding-ownership/listing_04_attempt_modifying_snapshot/src/lib.cairo:16:5
    rec.width = temp;
    ^*******^

error: could not compile `listing_04_04` due to previous error
error: `scarb metadata` exited with error

The compiler prevents us from modifying values associated to snapshots.

Mutable References

We can achieve the behavior we want in Listing 4-4 by using a mutable reference instead of a snapshot. Mutable references are actually mutable values passed to a function that are implicitly returned at the end of the function, returning ownership to the calling context. By doing so, they allow you to mutate the value passed while keeping ownership of it by returning it automatically at the end of the execution. In Cairo, a parameter can be passed as mutable reference using the ref modifier.

Note: In Cairo, a parameter can only be passed as mutable reference using the ref modifier if the variable is declared as mutable with mut.

In Listing 4-5, we use a mutable reference to modify the value of the height and width fields of the Rectangle instance in the flip function.

#[derive(Drop)]
struct Rectangle {
    height: u64,
    width: u64,
}

fn main() {
    let mut rec = Rectangle { height: 3, width: 10 };
    flip(ref rec);
    println!("height: {}, width: {}", rec.height, rec.width);
}

fn flip(ref rec: Rectangle) {
    let temp = rec.height;
    rec.height = rec.width;
    rec.width = temp;
}

Listing 4-5: Use of a mutable reference to modify a value

First, we change rec to be mut. Then we pass a mutable reference of rec into flip with ref rec, and update the function signature to accept a mutable reference with ref rec: Rectangle. This makes it very clear that the flip function will mutate the value of the Rectangle instance passed as parameter.

The output of the program is:

$ scarb cairo-run 
   Compiling listing_04_05 v0.1.0 (listings/ch04-understanding-ownership/listing_05_mutable_reference/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running listing_04_05
height: 10, width: 3
Run completed successfully, returning []

As expected, the height and width fields of the rec variable have been swapped.

Small Recap

Let’s recap what we’ve discussed about the linear type system, ownership, snapshots, and references:

• At any given time, a variable can only have one owner.
• You can pass a variable by-value, by-snapshot, or by-reference to a function.
• If you pass-by-value, ownership of the variable is transferred to the function.
• If you want to keep ownership of the variable and know that your function won’t mutate it, you can pass it as a snapshot with @.
• If you want to keep ownership of the variable and know that your function will mutate it, you can pass it as a mutable reference with ref.

Using Structs to Structure Related Data

A struct, or structure, is a custom data type that lets you package together and name multiple related values that make up a meaningful group. If you’re familiar with an object-oriented language, a struct is like an object’s data attributes. In this chapter, we’ll compare and contrast tuples with structs to build on what you already know and demonstrate when structs are a better way to group data.

We’ll demonstrate how to define and instantiate structs. We’ll discuss how to define associated functions, especially the kind of associated functions called methods, to specify behavior associated with a struct type. Structs and enums (discussed in the next chapter) are the building blocks for creating new types in your program’s domain to take full advantage of Cairo's compile-time type checking.

Defining and Instantiating Structs

Structs are similar to tuples, discussed in the Data Types section, in that both hold multiple related values. Like tuples, the pieces of a struct can be different types. Unlike with tuples, in a struct you’ll name each piece of data so it’s clear what the values mean. Adding these names means that structs are more flexible than tuples: you don’t have to rely on the order of the data to specify or access the values of an instance.

To define a struct, we enter the keyword struct and name the entire struct. A struct’s name should describe the significance of the pieces of data being grouped together. Then, inside curly brackets, we define the names and types of the pieces of data, which we call fields. For example, Listing 5-1 shows a struct that stores information about a user account.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}

Listing 5-1: A User struct definition

To use a struct after we’ve defined it, we create an instance of that struct by specifying concrete values for each of the fields. We create an instance by stating the name of the struct and then add curly brackets containing key: value pairs, where the keys are the names of the fields and the values are the data we want to store in those fields. We don’t have to specify the fields in the same order in which we declared them in the struct. In other words, the struct definition is like a general template for the type, and instances fill in that template with particular data to create values of the type.

For example, we can declare two particular users as shown in Listing 5-2.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}

fn main() {
    let user1 = User {
        active: true, username: "someusername123", email: "someone@example.com", sign_in_count: 1
    };
    let user2 = User {
        sign_in_count: 1, username: "someusername123", active: true, email: "someone@example.com"
    };
}

Listing 5-2: Creating two instances of the User struct

To get a specific value from a struct, we use dot notation. For example, to access user1's email address, we use user1.email. If the instance is mutable, we can change a value by using the dot notation and assigning into a particular field. Listing 5-3 shows how to change the value in the email field of a mutable User instance.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}
fn main() {
    let mut user1 = User {
        active: true, username: "someusername123", email: "someone@example.com", sign_in_count: 1
    };
    user1.email = "anotheremail@example.com";
}

fn build_user(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username: username, email: email, sign_in_count: 1, }
}

fn build_user_short(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username, email, sign_in_count: 1, }
}

Listing 5-3: Changing the value in the email field of a User instance

Note that the entire instance must be mutable; Cairo doesn’t allow us to mark only certain fields as mutable.

As with any expression, we can construct a new instance of the struct as the last expression in the function body to implicitly return that new instance.

Listing 5-4 shows a build_user function that returns a User instance with the given email and username. The active field gets the value of true, and the sign_in_count gets a value of 1.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}
fn main() {
    let mut user1 = User {
        active: true, username: "someusername123", email: "someone@example.com", sign_in_count: 1
    };
    user1.email = "anotheremail@example.com";
}

fn build_user(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username: username, email: email, sign_in_count: 1, }
}

fn build_user_short(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username, email, sign_in_count: 1, }
}

Listing 5-4: A build_user function that takes an email and username and returns a User instance.

It makes sense to name the function parameters with the same name as the struct fields, but having to repeat the email and username field names and variables is a bit tedious. If the struct had more fields, repeating each name would get even more annoying. Luckily, there’s a convenient shorthand!

Using the Field Init Shorthand

Because the parameter names and the struct field names are exactly the same in Listing 5-4, we can use the field init shorthand syntax to rewrite build_user so it behaves exactly the same but doesn’t have the repetition of username and email, as shown in Listing 5-5.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}
fn main() {
    let mut user1 = User {
        active: true, username: "someusername123", email: "someone@example.com", sign_in_count: 1
    };
    user1.email = "anotheremail@example.com";
}

fn build_user(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username: username, email: email, sign_in_count: 1, }
}

fn build_user_short(email: ByteArray, username: ByteArray) -> User {
    User { active: true, username, email, sign_in_count: 1, }
}

Listing 5-5: A build_user function that uses field init shorthand because the username and email parameters have the same name as struct fields.

Here, we’re creating a new instance of the User struct, which has a field named email. We want to set the email field’s value to the value in the email parameter of the build_user function. Because the email field and the email parameter have the same name, we only need to write email rather than email: email.

Creating Instances from Other Instances with Struct Update Syntax

It’s often useful to create a new instance of a struct that includes most of the values from another instance, but changes some. You can do this using struct update syntax.

First, in Listing 5-6 we show how to create a new User instance in user2 regularly, without the update syntax. We set a new value for email but otherwise use the same values from user1 that we created in Listing 5-2.

Filename: src/lib.cairo

#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}

fn main() {
    // --snip--

    let user1 = User {
        email: "someone@example.com", username: "someusername123", active: true, sign_in_count: 1,
    };

    let user2 = User {
        active: user1.active,
        username: user1.username,
        email: "another@example.com",
        sign_in_count: user1.sign_in_count,
    };
}

Listing 5-6: Creating a new User instance using all but one of the values from user1

Using struct update syntax, we can achieve the same effect with less code, as shown in Listing 5-7. The syntax .. specifies that the remaining fields not explicitly set should have the same value as the fields in the given instance.

Filename: src/lib.cairo

use core::byte_array;
#[derive(Drop)]
struct User {
    active: bool,
    username: ByteArray,
    email: ByteArray,
    sign_in_count: u64,
}

fn main() {
    // --snip--

    let user1 = User {
        email: "someone@example.com", username: "someusername123", active: true, sign_in_count: 1,
    };

    let user2 = User { email: "another@example.com", ..user1 };
}

Listing 5-7: Using struct update syntax to set a new email value for a User instance but to use the rest of the values from user1

The code in Listing 5-7 also creates an instance of user2 that has a different value for email but has the same values for the username, active, and sign_in_count fields as user1. The ..user1 part must come last to specify that any remaining fields should get their values from the corresponding fields in user1, but we can choose to specify values for as many fields as we want in any order, regardless of the order of the fields in the struct’s definition.

Note that the struct update syntax uses = like an assignment; this is because it moves the data, just as we saw in the "Moving Values" section. In this example, we can no longer use user1 as a whole after creating user2 because the ByteArray in the username field of user1 was moved into user2. If we had given user2 new ByteArray values for both email and username, and thus only used the active and sign_in_count values from user1, then user1 would still be valid after creating user2. Both active and sign_in_count are types that implement the Copy trait, so the behavior we discussed in the "Copy Trait" section would apply.

An Example Program Using Structs

To understand when we might want to use structs, let’s write a program that calculates the area of a rectangle. We’ll start by using single variables, and then refactor the program until we’re using structs instead.

Let’s make a new project with Scarb called rectangles that will take the width and height of a rectangle specified in pixels and calculate the area of the rectangle. Listing 5-8 shows a short program with one way of doing exactly that in our project’s src/lib.cairo.

Filename: src/lib.cairo

fn main() {
    let width = 30;
    let height = 10;
    let area = area(width, height);
    println!("Area is {}", area);
}

fn area(width: u64, height: u64) -> u64 {
    width * height
}

Listing 5-8: Calculating the area of a rectangle specified by separate width and height variables.

Now run the program with scarb cairo-run:

$ scarb cairo-run 
   Compiling listing_04_06_no_struct v0.1.0 (listings/ch05-using-structs-to-structure-related-data/listing_03_no_struct/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running listing_04_06_no_struct
Area is 300
Run completed successfully, returning []

This code succeeds in figuring out the area of the rectangle by calling the area function with each dimension, but we can do more to make this code clear and readable.

The issue with this code is evident in the signature of area:

fn area(width: u64, height: u64) -> u64 {

The area function is supposed to calculate the area of one rectangle, but the function we wrote has two parameters, and it’s not clear anywhere in our program that the parameters are related. It would be more readable and more manageable to group width and height together. We’ve already discussed one way we might do that in the Tuple Section of Chapter 2.

Refactoring with Tuples

Listing 5-9 shows another version of our program that uses tuples.

Filename: src/lib.cairo

fn main() {
    let rectangle = (30, 10);
    let area = area(rectangle);
    println!("Area is {}", area);
}

fn area(dimension: (u64, u64)) -> u64 {
    let (x, y) = dimension;
    x * y
}

Listing 5-9: Specifying the width and height of the rectangle with a tuple.

In one way, this program is better. Tuples let us add a bit of structure, and we’re now passing just one argument. But in another way, this version is less clear: tuples don’t name their elements, so we have to index into the parts of the tuple, making our calculation less obvious.

Mixing up the width and height wouldn’t matter for the area calculation, but if we want to calculate the difference, it would matter! We would have to keep in mind that width is the tuple index 0 and height is the tuple index 1. This would be even harder for someone else to figure out and keep in mind if they were to use our code. Because we haven’t conveyed the meaning of our data in our code, it’s now easier to introduce errors.

Refactoring with Structs: Adding More Meaning

We use structs to add meaning by labeling the data. We can transform the tuple we’re using into a struct with a name for the whole as well as names for the parts.

Filename: src/lib.cairo

struct Rectangle {
    width: u64,
    height: u64,
}

fn main() {
    let rectangle = Rectangle { width: 30, height: 10, };
    let area = area(rectangle);
    println!("Area is {}", area);
}

fn area(rectangle: Rectangle) -> u64 {
    rectangle.width * rectangle.height
}

Listing 5-10: Defining a Rectangle struct.

Here we’ve defined a struct and named it Rectangle. Inside the curly brackets, we defined the fields as width and height, both of which have type u64. Then, in main, we created a particular instance of Rectangle that has a width of 30 and a height of 10. Our area function is now defined with one parameter, which we’ve named rectangle which is of type Rectangle struct. We can then access the fields of the instance with dot notation, and it gives descriptive names to the values rather than using the tuple index values of 0 and 1.

Conversions of Custom Types

We've already described how to perform type conversion on in-built types, see Data Types > Type Conversion. In this section, we will see how to define conversions for custom types.

Note: conversion can be defined for compound types, e.g. tuples, too.

Into

Defining a conversion for a custom type using the Into trait will typically require specification of the type to convert into, as the compiler is unable to determine this most of the time. However this is a small trade-off considering we get the functionality for free.

// Compiler automatically imports the core library, so you can omit this import
use core::traits::Into;

#[derive(Drop, PartialEq)]
struct Rectangle {
    width: u64,
    height: u64,
}

#[derive(Drop)]
struct Square {
    side_length: u64,
}

impl SquareIntoRectangle of Into<Square, Rectangle> {
    fn into(self: Square) -> Rectangle {
        Rectangle { width: self.side_length, height: self.side_length }
    }
}

fn main() {
    let square = Square { side_length: 5 };
    // Compiler will complain if you remove the type annotation
    let result: Rectangle = square.into();
    let expected = Rectangle { width: 5, height: 5 };
    assert!(
        result == expected,
        "A square is always convertible to a rectangle with the same width and height!"
    );
}

TryInto

Defining a conversion for TryInto is similar to defining it for Into.

// Compiler automatically imports the core library, so you can omit this import
use core::traits::TryInto;

#[derive(Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

#[derive(Drop, PartialEq)]
struct Square {
    side_length: u64,
}

impl RectangleIntoSquare of TryInto<Rectangle, Square> {
    fn try_into(self: Rectangle) -> Option<Square> {
        if self.height == self.width {
            Option::Some(Square { side_length: self.height })
        } else {
            Option::None
        }
    }
}

fn main() {
    let rectangle = Rectangle { width: 8, height: 8 };
    let result: Square = rectangle.try_into().unwrap();
    let expected = Square { side_length: 8 };
    assert!(
        result == expected,
        "Rectangle with equal width and height should be convertible to a square."
    );

    let rectangle = Rectangle { width: 5, height: 8 };
    let result: Option<Square> = rectangle.try_into();
    assert!(
        result.is_none(),
        "Rectangle with different width and height should not be convertible to a square."
    );
}

Method Syntax

Methods are similar to functions: we declare them with the fn keyword and a name, they can have parameters and a return value, and they contain some code that’s run when the method is called from somewhere else. Unlike functions, methods are defined within the context of a struct (or an enum which we cover in Chapter 6), and their first parameter is always self, which represents the instance of the type the method is being called on.

Defining Methods

Let’s change the area function that has a Rectangle instance as a parameter and instead make an area method defined on the Rectangle struct, as shown in Listing 5-11

#[derive(Copy, Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

trait RectangleTrait {
    fn area(self: @Rectangle) -> u64;
}

impl RectangleImpl of RectangleTrait {
    fn area(self: @Rectangle) -> u64 {
        (*self.width) * (*self.height)
    }
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50, };
    println!("Area is {}", rect1.area());
}

Listing 5-11: Defining an area method on the Rectangle struct.

To define the function within the context of Rectangle, we start an impl (implementation) block for a trait RectangleTrait that defines the methods that can be called on a Rectangle instance. As impl blocks can only be defined for traits and not types, we need to define this trait first - but it's not meant to be used for anything else.

Everything within this impl block will be associated with the Rectangle type. Then we move the area function within the impl curly brackets and change the first (and in this case, only) parameter to be self in the signature and everywhere within the body. In main, where we called the area function and passed rect1 as an argument, we can instead use method syntax to call the area method on our Rectangle instance. The method syntax goes after an instance: we add a dot followed by the method name, parentheses, and any arguments.

In the signature for area, we use self: @Rectangle instead of rectangle: @Rectangle. Methods must have a parameter named self, for their first parameter, and the type of self indicates the type that method can be called on. Methods can take ownership of self, but self can also be passed by snapshot or by reference, just like any other parameter.

There is no direct link between a type and a trait. Only the type of the self parameter of a method defines the type from which this method can be called. That means, it is technically possible to define methods on multiple types in a same trait (mixing Rectangle and Circle methods, for example). But this is not a recommended practice as it can lead to confusion.

The main reason for using methods instead of functions, in addition to providing method syntax, is for organization. We’ve put all the things we can do with an instance of a type in one impl block rather than making future users of our code search for capabilities of Rectangle in various places in the library we provide.

The generate_trait Attribute

If you are familiar with Rust, you may find Cairo's approach confusing because methods cannot be defined directly on types. Instead, you must define a trait and an implementation of this trait associated with the type for which the method is intended. However, defining a trait and then implementing it to define methods on a specific type is verbose, and unnecessary: the trait itself will not be reused.

So, to avoid defining useless traits, Cairo provides the #[generate_trait] attribute to add above a trait implementation, which tells to the compiler to generate the corresponding trait definition for you, and let's you focus on the implementation only. Both approaches are equivalent, but it's considered a best practice to not explicitly define traits in this case.

The previous example can also be written as follows:

#[derive(Copy, Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

#[generate_trait]
impl RectangleImpl of RectangleTrait {
    fn area(self: @Rectangle) -> u64 {
        (*self.width) * (*self.height)
    }
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50, };
    println!("Area is {}", rect1.area());
}

Let's use this #[generate_trait] in the following chapters to make our code cleaner.

Snapshots and References

As the area method does not modify the calling instance, self is declared as a snapshot of a Rectangle instance with the @ snapshot operator. But, of course, we can also define some methods receiving a mutable reference of this instance, to be able to modify it.

Let's write a new method scale which resizes a rectangle of a factor given as parameter:

#[generate_trait]
impl RectangleImpl of RectangleTrait {
    fn area(self: @Rectangle) -> u64 {
        (*self.width) * (*self.height)
    }
    fn scale(ref self: Rectangle, factor: u64) {
        self.width *= factor;
        self.height *= factor;
    }
}

fn main() {
    let mut rect2 = Rectangle { width: 10, height: 20 };
    rect2.scale(2);
    println!("The new size is (width: {}, height: {})", rect2.width, rect2.height);
}

It is also possible to define a method which takes ownership of the instance by using just self as the first parameter but it is rare. This technique is usually used when the method transforms self into something else and you want to prevent the caller from using the original instance after the transformation.

Look at the Understanding Ownership chapter for more details about these important notions.

Methods with Several Parameters

Let’s practice using methods by implementing another method on the Rectangle struct. This time we want to write the method can_hold which accepts another instance of Rectangle and returns true if this rectangle can fit completely within self; otherwise, it should return false.

#[generate_trait]
impl RectangleImpl of RectangleTrait {
    fn area(self: @Rectangle) -> u64 {
        *self.width * *self.height
    }

    fn scale(ref self: Rectangle, factor: u64) {
        self.width *= factor;
        self.height *= factor;
    }

    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width > *other.width && *self.height > *other.height
    }
}

fn main() {
    let rect1 = Rectangle { width: 30, height: 50, };
    let rect2 = Rectangle { width: 10, height: 40, };
    let rect3 = Rectangle { width: 60, height: 45, };

    println!("Can rect1 hold rect2? {}", rect1.can_hold(@rect2));
    println!("Can rect1 hold rect3? {}", rect1.can_hold(@rect3));
}

Here, we expect that rect1 can hold rect2 but not rect3.

Associated functions

We call associated functions all functions that are defined inside an impl block that are associated to a specific type. While this is not enforced by the compiler, it is a good practice to keep associated functions related to the same type in the same impl block - for example, all functions related to Rectangle will be grouped in the same impl block for RectangleTrait.

Methods are a special kind of associated function, but we can also define associated functions that don’t have self as their first parameter (and thus are not methods) because they don’t need an instance of the type to work with, but are still associated with that type.

Associated functions that aren’t methods are often used for constructors that will return a new instance of the type. These are often called new, but new isn’t a special name and isn’t built into the language. For example, we could choose to provide an associated function named square that would have one dimension parameter and use that as both width and height, thus making it easier to create a square Rectangle rather than having to specify the same value twice:

Let's create the function new which creates a Rectangle from a width and a height, square which creates a square Rectangle from a size and avg which computes the average of two Rectangle instances:

#[generate_trait]
impl RectangleImpl of RectangleTrait {
    fn area(self: @Rectangle) -> u64 {
        (*self.width) * (*self.height)
    }

    fn new(width: u64, height: u64) -> Rectangle {
        Rectangle { width, height }
    }

    fn square(size: u64) -> Rectangle {
        Rectangle { width: size, height: size }
    }

    fn avg(lhs: @Rectangle, rhs: @Rectangle) -> Rectangle {
        Rectangle {
            width: ((*lhs.width) + (*rhs.width)) / 2, height: ((*lhs.height) + (*rhs.height)) / 2
        }
    }
}

fn main() {
    let rect1 = RectangleTrait::new(30, 50);
    let rect2 = RectangleTrait::square(10);

    println!(
        "The average Rectangle of {:?} and {:?} is {:?}",
        @rect1,
        @rect2,
        RectangleTrait::avg(@rect1, @rect2)
    );
}

To call the square associated function, we use the :: syntax with the struct name; let sq = Rectangle::square(3); is an example. This function is namespaced by the struct: the :: syntax is used for both associated functions and namespaces created by modules. We’ll discuss modules in Chapter 7.

Note that the avg function could also be written as a method with self as the first rectangle. In this case, instead of using the method with RectangleTrait::avg(@rect1, @rect2), it would be called with rect1.avg(rect2).

Multiple Traits and impl Blocks

Each struct is allowed to have multiple trait and impl blocks. For example, the following code is equivalent to the code shown in the Methods with several parameters section, which has each method in its own trait and impl blocks.

#[generate_trait]
impl RectangleCalcImpl of RectangleCalc {
    fn area(self: @Rectangle) -> u64 {
        (*self.width) * (*self.height)
    }
}

#[generate_trait]
impl RectangleCmpImpl of RectangleCmp {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width > *other.width && *self.height > *other.height
    }
}

There’s no strong reason to separate these methods into multiple trait and impl blocks here, but this is valid syntax.

Enums and Pattern Matching

In this chapter, we’ll look at enumerations, also referred to as enums. Enums allow you to define a type by enumerating its possible variants. First, we’ll define and use an enum to show how an enum can encode meaning along with data. Next, we’ll explore a particularly useful enum, called Option, which expresses that a value can be either something or nothing. Finally, we’ll look at how pattern matching in the match expression makes it easy to run different code for different values of an enum.

Enums

Enums, short for "enumerations," are a way to define a custom data type that consists of a fixed set of named values, called variants. Enums are useful for representing a collection of related values where each value is distinct and has a specific meaning.

Enum Variants and Values

Here's a simple example of an enum:

#[derive(Drop)]
enum Direction {
    North,
    East,
    South,
    West,
}

In this example, we've defined an enum called Direction with four variants: North, East, South, and West. The naming convention is to use PascalCase for enum variants. Each variant represents a distinct value of the Direction type. In this particular example, variants don't have any associated value. One variant can be instantiated using this syntax:

#[derive(Drop)]
enum Direction {
    North,
    East,
    South,
    West,
}

fn main() {
    let direction = Direction::North;
}

Now let's imagine that our variants have associated values, that store the exact degree of the direction. We can define a new Direction enum:

#[derive(Drop)]
enum Direction {
    North: u128,
    East: u128,
    South: u128,
    West: u128,
}

fn main() {
    let direction = Direction::North(10);
}

and instantiate it as follows:

#[derive(Drop)]
enum Direction {
    North: u128,
    East: u128,
    South: u128,
    West: u128,
}

fn main() {
    let direction = Direction::North(10);
}

In this code, each variant is associated with a u128 value, representing the direction in degrees. In the next example, we will see that it is also possible to associate different data types with each variant.

It's easy to write code that acts differently depending on the variant of an enum instance, in this example to run specific code according to a direction. You can learn more about it in the Match Control Flow Construct section.

Enums Combined with Custom Types

Enums can also be used to store more interesting custom data associated with each variant. For example:

#[derive(Drop)]
enum Message {
    Quit,
    Echo: felt252,
    Move: (u128, u128),
}

In this example, the Message enum has three variants: Quit, Echo, and Move, all with different types:

• Quit doesn't have any associated value.
• Echo is a single felt252.
• Move is a tuple of two u128 values.

You could even use a Struct or another enum you defined inside one of your enum variants.

Trait Implementations for Enums

In Cairo, you can define traits and implement them for your custom enums. This allows you to define methods and behaviors associated with the enum. Here's an example of defining a trait and implementing it for the previous Message enum:

trait Processing {
    fn process(self: Message);
}

impl ProcessingImpl of Processing {
    fn process(self: Message) {
        match self {
            Message::Quit => { println!("quitting") },
            Message::Echo(value) => { println!("echoing {}", value) },
            Message::Move((x, y)) => { println!("moving from {} to {}", x, y) },
        }
    }
}

In this example, we implemented the Processing trait for Message. Here is how it could be used to process a Quit message:

#[derive(Drop)]
enum Message {
    Quit,
    Echo: felt252,
    Move: (u128, u128),
}

trait Processing {
    fn process(self: Message);
}

impl ProcessingImpl of Processing {
    fn process(self: Message) {
        match self {
            Message::Quit => { println!("quitting") },
            Message::Echo(value) => { println!("echoing {}", value) },
            Message::Move((x, y)) => { println!("moving from {} to {}", x, y) },
        }
    }
}
fn main() {
    let msg: Message = Message::Quit;
    msg.process(); // prints "quitting"
}

The Option Enum and Its Advantages

The Option enum is a standard Cairo enum that represents the concept of an optional value. It has two variants: Some: T and None. Some: T indicates that there's a value of type T, while None represents the absence of a value.

enum Option<T> {
    Some: T,
    None,
}

The Option enum is helpful because it allows you to explicitly represent the possibility of a value being absent, making your code more expressive and easier to reason about. Using Option can also help prevent bugs caused by using uninitialized or unexpected null values.

To give you an example, here is a function which returns the index of the first element of an array with a given value, or None if the element is not present.

We are demonstrating two approaches for the above function:

• Recursive approach with find_value_recursive.
• Iterative approach with find_value_iterative.

fn find_value_recursive(mut arr: Span<felt252>, value: felt252, index: usize) -> Option<usize> {
    match arr.pop_front() {
        Option::Some(index_value) => { if (*index_value == value) {
            return Option::Some(index);
        } },
        Option::None => { return Option::None; },
    };

    find_value_recursive(arr, value, index + 1)
}

fn find_value_iterative(mut arr: Span<felt252>, value: felt252) -> Option<usize> {
    let mut result = Option::None;
    let mut index = 0;

    while let Option::Some(array_value) = arr.pop_front() {
        if (*array_value == value) {
            result = Option::Some(index);
            break;
        };

        index += 1;
    };

    result
}

Enums can be useful in many situations, especially when using the match flow construct that we just used. We will describe it in the next section.

Other enums are used very often, such as the Result enum, allowing to handle errors gracefully. We will explain the Result enum in detail in the "Error Handling" chapter.

The Match Control Flow Construct

Cairo has an extremely powerful control flow construct called match that allows you to compare a value against a series of patterns and then execute code based on which pattern matches. Patterns can be made up of literal values, variable names, wildcards, and many other things. The power of match comes from the expressiveness of the patterns and the fact that the compiler confirms that all possible cases are handled.

Think of a match expression as being like a coin-sorting machine: coins slide down a track with variously sized holes along it, and each coin falls through the first hole it encounters that it fits into. In the same way, values go through each pattern in a match, and at the first pattern the value “fits”, the value falls into the associated code block to be used during execution.

Speaking of coins, let’s use them as an example using match! We can write a function that takes an unknown US coin and, in a similar way as the counting machine, determines which coin it is and returns its value in cents, as shown in Listing 6-1.

enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter,
}

fn value_in_cents(coin: Coin) -> felt252 {
    match coin {
        Coin::Penny => 1,
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter => 25,
    }
}

Listing 6-1: An enum and a match expression that has the variants of the enum as its patterns

Let’s break down the match expression in the value_in_cents function. First, we list the match keyword followed by an expression, which in this case is the value coin. This seems very similar to a conditional expression used with the if statement, but there’s a big difference: with if, the condition needs to evaluate to a boolean value, but here it can be any type. The type of coin in this example is the Coin enum that we defined on the first line.

Next are the match arms. An arm has two parts: a pattern and some code. The first arm here has a pattern that is the value Coin::Penny and then the => operator that separates the pattern and the code to run. The code in this case is just the value 1. Each arm is separated from the next with a comma.

When the match expression executes, it compares the resultant value against the pattern of each arm, in the order they are given. If a pattern matches the value, the code associated with that pattern is executed. If that pattern doesn’t match the value, execution continues to the next arm, much as in a coin-sorting machine. We can have as many arms as we need: in the above example, our match has four arms.

The code associated with each arm is an expression, and the resultant value of the expression in the matching arm is the value that gets returned for the entire match expression.

We don’t typically use curly brackets if the match arm code is short, as it is in our example where each arm just returns a value. If you want to run multiple lines of code in a match arm, you must use curly brackets, with a comma following the arm. For example, the following code prints “Lucky penny!” every time the method is called with a Coin::Penny, but still returns the last value of the block, 1:

fn value_in_cents(coin: Coin) -> felt252 {
    match coin {
        Coin::Penny => {
            println!("Lucky penny!");
            1
        },
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter => 25,
    }
}

Patterns That Bind to Values

Another useful feature of match arms is that they can bind to the parts of the values that match the pattern. This is how we can extract values out of enum variants.

As an example, let’s change one of our enum variants to hold data inside it. From 1999 through 2008, the United States minted quarters with different designs for each of the 50 states on one side. No other coins got state designs, so only quarters have this extra value. We can add this information to our enum by changing the Quarter variant to include a UsState value stored inside it, which we’ve done in Listing 6-2.

#[derive(Drop, Debug)] // Debug so we can inspect the state in a minute
enum UsState {
    Alabama,
    Alaska,
}

#[derive(Drop)]
enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter: UsState,
}

Listing 6-2: A Coin enum in which the Quarter variant also holds a UsState value

Let’s imagine that a friend is trying to collect all 50 state quarters. While we sort our loose change by coin type, we’ll also call out the name of the state associated with each quarter so that if it’s one our friend doesn’t have, they can add it to their collection.

In the match expression for this code, we add a variable called state to the pattern that matches values of the variant Coin::Quarter. When a Coin::Quarter matches, the state variable will bind to the value of that quarter’s state. Then we can use state in the code for that arm, like so:

fn value_in_cents(coin: Coin) -> felt252 {
    match coin {
        Coin::Penny => 1,
        Coin::Nickel => 5,
        Coin::Dime => 10,
        Coin::Quarter(state) => {
            println!("State quarter from {:?}!", state);
            25
        }
    }
}

Because state is an UsState enum which implements the Debug trait, we can print state value with println! macro.

Note: {:?} is a special formatting syntax that allows to print a debug form of the parameter passed to the println! macro. You can find more information about it in Appendix C.

If we were to call value_in_cents(Coin::Quarter(UsState::Alaska)), coin would be Coin::Quarter(UsState::Alaska). When we compare that value with each of the match arms, none of them match until we reach Coin::Quarter(state). At that point, the binding for state will be the value UsState::Alaska. We can then use that binding in println! macro, thus getting the inner state value out of the Coin enum variant for Quarter.

Matching with Option<T>

In the previous section, we wanted to get the inner T value out of the Some case when using Option<T>; we can also handle Option<T> using match, as we did with the Coin enum! Instead of comparing coins, we’ll compare the variants of Option<T>, but the way the match expression works remains the same.

Let’s say we want to write a function that takes an Option<u8> and, if there’s a value inside, adds 1 to that value. If there is no value inside, the function should return the None value and not attempt to perform any operations.

This function is very easy to write, thanks to match, and will look like Listing 6-3.

fn plus_one(x: Option<u8>) -> Option<u8> {
    match x {
        Option::Some(val) => Option::Some(val + 1),
        Option::None => Option::None,
    }
}

fn main() {
    let five: Option<u8> = Option::Some(5);
    let six: Option<u8> = plus_one(five);
    let none = plus_one(Option::None);
}

Listing 6-3: A function that uses a match expression on an Option<u8>

Let’s examine the first execution of plus_one in more detail. When we call plus_one(five), the variable x in the body of plus_one will have the value Some(5). We then compare that against each match arm:

        Option::Some(val) => Option::Some(val + 1),

Does Option::Some(5) value match the pattern Option::Some(val)? It does! We have the same variant. The val binds to the value contained in Option::Some, so val takes the value 5. The code in the match arm is then executed, so we add 1 to the value of val and create a new Option::Some value with our total 6 inside. Because the first arm matched, no other arms are compared.

Now let’s consider the second call of plus_one in our main function, where x is Option::None. We enter the match and compare to the first arm:

        Option::Some(val) => Option::Some(val + 1),

The Option::Some(val) value doesn’t match the pattern Option::None, so we continue to the next arm:

        Option::None => Option::None,

It matches! There’s no value to add to, so the matching construct ends and returns the Option::None value on the right side of =>.

Combining match and enums is useful in many situations. You’ll see this pattern a lot in Cairo code: match against an enum, bind a variable to the data inside, and then execute code based on it. It’s a bit tricky at first, but once you get used to it, you’ll wish you had it in all languages. It’s consistently a user favorite.

Matches Are Exhaustive

There’s one other aspect of match we need to discuss: the arms’ patterns must cover all possibilities. Consider this version of our plus_one function, which has a bug and won’t compile:

fn plus_one(x: Option<u8>) -> Option<u8> {
    match x {
        Option::Some(val) => Option::Some(val + 1),
    }
}

We didn’t handle the None case, so this code will cause a bug. Luckily, it’s a bug Cairo knows how to catch. If we try to compile this code, we’ll get this error:

$ scarb cairo-run 
   Compiling no_listing_08_missing_match_arm v0.1.0 (listings/ch06-enums-and-pattern-matching/no_listing_09_missing_match_arm/Scarb.toml)
error: Missing match arm: `None` not covered.
 --> listings/ch06-enums-and-pattern-matching/no_listing_09_missing_match_arm/src/lib.cairo:5:5
    match x {
    ^*******^

error: could not compile `no_listing_08_missing_match_arm` due to previous error
error: `scarb metadata` exited with error

Cairo knows that we didn’t cover every possible case, and even knows which pattern we forgot! Matches in Cairo are exhaustive: we must exhaust every last possibility in order for the code to be valid. Especially in the case of Option<T>, when Cairo prevents us from forgetting to explicitly handle the None case, it protects us from assuming that we have a value when we might have null, thus making the billion-dollar mistake discussed earlier impossible.

Catch-all with the _ Placeholder

Using enums, we can also take special actions for a few particular values, but for all other values take one default action. _ is a special pattern that matches any value and does not bind to that value. You can use it by simply adding a new arm with _ as the pattern for the last arm of the match expression.

Imagine we have a vending machine that only accepts Dime coins. We want to have a function that processes inserted coins and returns true only if the coin is accepted.

Here's a vending_machine_accept function that implements this logic:

fn vending_machine_accept(coin: Coin) -> bool {
    match coin {
        Coin::Dime => true,
        _ => false,
    }
}

This example also meets the exhaustiveness requirement because we’re explicitly ignoring all other values in the last arm; we haven’t forgotten anything.

There's no catch-all pattern in Cairo that allows you to use the value of the pattern.

Multiple Patterns with the | Operator

In match expressions, you can match multiple patterns using the | syntax, which is the pattern or operator.

For example, in the following code we modified the vending_machine_accept function to accept both Dime and Quarter coins in a single arm:

fn vending_machine_accept(coin: Coin) -> bool {
    match coin {
        Coin::Dime | Coin::Quarter => true,
        _ => false,
    }
}

Matching Tuples

It is possible to match tuples. Let's introduce a new DayType enum:

#[derive(Drop)]
enum DayType {
    Week,
    Weekend,
    Holiday
}

Now, let's suppose that our vending machine accepts any coin on weekdays, but only accepts quarters and dimes on weekends and holidays. We can modify the vending_machine_accept function to accept a tuple of a Coin and a Weekday and return true only if the given coin is accepted on the specified day:

fn vending_machine_accept(c: (DayType, Coin)) -> bool {
    match c {
        (DayType::Week, _) => true,
        (_, Coin::Dime) | (_, Coin::Quarter) => true,
        (_, _) => false,
    }
}

Writing (_, _) for the last arm of a tuple matching pattern might feel superfluous. Hence, we can use the _ => syntax if we want, for example, that our vending machine only accepts quarters on weekdays:

fn vending_week_machine(c: (DayType, Coin)) -> bool {
    match c {
        (DayType::Week, Coin::Quarter) => true,
        _ => false,
    }
}

Matching felt252 and Integer Variables

You can also match felt252 and integer variables. This is useful when you want to match against a range of values. However, there are some restrictions:

• Only integers that fit into a single felt252 are supported (i.e. u256 is not supported).
• The first arm must be 0.
• Each arm must cover a sequential segment, contiguously with other arms.

Imagine we’re implementing a game where you roll a six-sided die to get a number between 0 and 5. If you have 0, 1 or 2 you win. If you have 3, you can roll again. For all other values you lose.

Here's a match that implements that logic:

fn roll(value: u8) {
    match value {
        0 | 1 | 2 => println!("you won!"),
        3 => println!("you can roll again!"),
        _ => println!("you lost...")
    }
}

These restrictions are planned to be relaxed in future versions of Cairo.

Concise Control Flow with if let and while let

if let

The if let syntax lets you combine if and let into a less verbose way to handle values that match one pattern while ignoring the rest. Consider the program in Listing 6-4 that matches on an Option::Some<u8> value in the config_max variable but only wants to execute code if the value is Option::Some variant.

fn main() {
    let config_max = Option::Some(5);
    match config_max {
        Option::Some(max) => println!("The maximum is configured to be {}", max),
        _ => (),
    }
}

Listing 6-4: A match that only cares about executing code when the value is Option::Some

If the value is Option::Some, we print out the value in the Option::Some variant by binding the value to the variable max in the pattern. We don’t want to do anything with the None value. To satisfy the match expression, we have to add _ => () after processing just one variant, which is annoying boilerplate code to add.

Instead, we could write this in a shorter way using if let. The following code behaves the same as the match in Listing 6-4:

fn main() {
    let number = Option::Some(5);
    if let Option::Some(max) = number {
        println!("The maximum is configured to be {}", max);
    }
}

The syntax if let takes a pattern and an expression separated by an equal sign. It works the same way as a match, where the expression is given to the match and the pattern is its first arm. In this case, the pattern is Option::Some(max), and max binds to the value inside Option::Some. We can then use max in the body of the if let block in the same way we used max in the corresponding match arm. The code in the if let block isn’t run if the value doesn’t match the pattern.

Using if let means less typing, less indentation, and less boilerplate code. However, you lose the exhaustive checking that match enforces. Choosing between match and if let depends on what you’re doing in your particular situation and whether gaining conciseness is an appropriate trade-off for losing exhaustive checking.

In other words, you can think of if let as syntactic sugar for a match that runs code when the value matches one pattern and then ignores all other values.

We can include an else with an if let. The block of code that goes with else is the same as the block of code that would go with the _ case in the match expression. Recall the Coin enum definition in Listing 6-2, where the Quarter variant also held a UsState value. If we wanted to count all non-quarter coins we see while also announcing the state of the quarters, we could do that with a match expression, like this:

#[derive(Drop)]
enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter,
}

fn main() {
    let coin = Coin::Quarter;
    let mut count = 0;
    match coin {
        Coin::Quarter => println!("You got a quarter!"),
        _ => count += 1,
    }
}

Or we could use an if let and else expression, like this:

#[derive(Drop)]
enum Coin {
    Penny,
    Nickel,
    Dime,
    Quarter,
}

fn main() {
    let coin = Coin::Quarter;
    let mut count = 0;
    if let Coin::Quarter = coin {
        println!("You got a quarter!");
    } else {
        count += 1;
    }
    println!("{}", count);
}

If you have a situation in which your program has logic that is too verbose to express using match, remember that if let is in your Cairo toolbox as well.

while let

The while let syntax is similar to the if let syntax, but it allows you to loop over a collection of values and execute a block of code for each value that matches a specified pattern. In the case below, the pattern is Option::Some(x), which matches any Some variant of the Option enum.

fn main() {
    let mut arr = array![1, 2, 3, 4, 5, 6, 7, 8, 9];
    let mut sum = 0;
    while let Option::Some(value) = arr.pop_front() {
        sum += value;
    };
    println!("{}", sum);
}

Using while let provides a more concise and idiomatic way of writing this loop compared to a traditional while loop with explicit pattern matching or handling of the Option type. However, as with if let, you lose the exhaustive checking that a match expression provides, so you need to be careful to handle any remaining cases outside the while let loop if necessary.

Managing Cairo Projects with Packages, Crates and Modules

As you write large programs, organizing your code will become increasingly important. By grouping related functionality and separating code with distinct features, you’ll clarify where to find code that implements a particular feature and where to go to change how a feature works.

The programs we’ve written so far have been in one module in one file. As a project grows, you should organize code by splitting it into multiple modules and then multiple files. As a package grows, you can extract parts into separate crates that become external dependencies. This chapter covers all these techniques.

We’ll also discuss encapsulating implementation details, which lets you reuse code at a higher level: once you’ve implemented an operation, other code can call your code without having to know how the implementation works.

A related concept is scope: the nested context in which code is written has a set of names that are defined as “in scope”. When reading, writing, and compiling code, programmers and compilers need to know whether a particular name at a particular spot refers to a variable, function, struct, enum, module, constant, or other item and what that item means. You can create scopes and change which names are in or out of scope. You can’t have two items with the same name in the same scope.

Cairo has a number of features that allow you to manage your code’s organization. These features, sometimes collectively referred to as the module system, include:

• Packages: A Scarb feature that lets you build, test, and share crates.
• Crates: A tree of modules that corresponds to a single compilation unit. It has a root directory, and a root module defined at the lib.cairo file under this directory.
• Modules and use: Let you control the organization and scope of items.
• Paths: A way of naming an item, such as a struct, function, or module.

In this chapter, we’ll cover all these features, discuss how they interact, and explain how to use them to manage scope. By the end, you should have a solid understanding of the module system and be able to work with scopes like a pro!

Packages and Crates

What is a Crate?

A crate is a subset of a package that is used in the actual Cairo compilation. This includes:

• The package source code, identified by the package name and the crate root, which is the main entry point of the package.
• A subset of the package metadata that identifies crate-level settings of the Cairo compiler, for example, the edition field in the Scarb.toml file.

Crates can contain modules, and the modules may be defined in other files that get compiled with the crate, as will be discussed in the subsequent sections.

What is the Crate Root?

The crate root is the lib.cairo source file that the Cairo compiler starts from and makes up the root module of your crate. We’ll explain modules in depth in the "Defining Modules to Control Scope" chapter.

What is a Package?

A Cairo package is a directory (or equivalent) containing:

• A Scarb.toml manifest file with a [package] section.
• Associated source code.

This definition implies that a package might contain other packages, with a corresponding Scarb.toml file for each package.

Creating a Package with Scarb

You can create a new Cairo package using the Scarb command-line tool. To create a new package, run the following command:

scarb new my_package

This command will generate a new package directory named my_package with the following structure:

my_package/
├── Scarb.toml
└── src
    └── lib.cairo

• src/ is the main directory where all the Cairo source files for the package will be stored.
• lib.cairo is the default root module of the crate, which is also the main entry point of the package.
• Scarb.toml is the package manifest file, which contains metadata and configuration options for the package, such as dependencies, package name, version, and authors. You can find documentation about it on the Scarb reference.

[package]
name = "my_package"
version = "0.1.0"
edition = "2024_07"

[dependencies]
# foo = { path = "vendor/foo" }

As you develop your package, you may want to organize your code into multiple Cairo source files. You can do this by creating additional .cairo files within the src directory or its subdirectories.

Defining Modules to Control Scope

In this section, we’ll talk about modules and other parts of the module system, namely paths that allow you to name items and the use keyword that brings a path into scope.

First, we’re going to start with a list of rules for easy reference when you’re organizing your code in the future. Then we’ll explain each of the rules in detail.

Modules Cheat Sheet

Here we provide a quick reference on how modules, paths and the use keyword work in the compiler, and how most developers organize their code. We’ll be going through examples of each of these rules throughout this chapter, but this is a great place to refer to as a reminder of how modules work. You can create a new Scarb project with scarb new backyard to follow along.

• Start from the crate root: When compiling a crate, the compiler first looks in the crate root file (src/lib.cairo) for code to compile.

• Declaring modules: In the crate root file, you can declare new modules; say, you declare a “garden” module with mod garden;. The compiler will look for the module’s code in these places:

  • Inline, within curly brackets that replace the semicolon following mod garden.

      // crate root file (src/lib.cairo)
      mod garden {
          // code defining the garden module goes here
      }
    

  • In the file src/garden.cairo.

• Declaring submodules: In any file other than the crate root, you can declare submodules. For example, you might declare mod vegetables; in src/garden.cairo. The compiler will look for the submodule’s code within the directory named for the parent module in these places:

  • Inline, directly following mod vegetables, within curly brackets instead of the semicolon.

    // src/garden.cairo file
    mod vegetables {
        // code defining the vegetables submodule goes here
    }
    

  • In the file src/garden/vegetables.cairo.

• Paths to code in modules: Once a module is part of your crate, you can refer to code in that module from anywhere else in that same crate, using the path to the code. For example, an Asparagus type in the vegetables submodule would be found at crate::garden::vegetables::Asparagus.

• Private vs public: Code within a module is private from its parent modules by default. This means that it may only be accessed by the current module and its descendants. To make a module public, declare it with pub mod instead of mod. To make items within a public module public as well, use pub before their declarations. Cairo also provides the pub(crate) keyword, allowing an item or module to be only visible within the crate in which the definition is included.

• The use keyword: Within a scope, the use keyword creates shortcuts to items to reduce repetition of long paths. In any scope that can refer to crate::garden::vegetables::Asparagus, you can create a shortcut with use crate::garden::vegetables::Asparagus; and from then on you only need to write Asparagus to make use of that type in the scope.

Here we create a crate named backyard that illustrates these rules. The crate’s directory, also named backyard, contains these files and directories:

backyard/
├── Scarb.toml
└── src
    ├── garden
    │   └── vegetables.cairo
    ├── garden.cairo
    └── lib.cairo

The crate root file in this case is src/lib.cairo, and it contains:

Filename: src/lib.cairo

pub mod garden;
use crate::garden::vegetables::Asparagus;

fn main() {
    let plant = Asparagus {};
    println!("I'm growing {:?}!", plant);
}

The pub mod garden; line imports the garden module. Using pub to make garden publicly accessible, or pub(crate) if you really want to make garden only available for your crate, is optional to run our program here, as the main function resides in the same module as pub mod garden; declaration. Nevertheless, not declaring garden as pub will make it not accessible from any other package. This line tells the compiler to include the code it finds in src/garden.cairo, which is:

Filename: src/garden.cairo

pub mod vegetables;

Here, pub mod vegetables; means the code in src/garden/vegetables.cairo is included too. That code is:

#[derive(Drop, Debug)]
pub struct Asparagus {}

The line use crate::garden::vegetables::Asparagus; lets us bring the Asparagus type into scope, so we can use it in the main function.

Now let’s get into the details of these rules and demonstrate them in action!

Grouping Related Code in Modules

Modules let us organize code within a crate for readability and easy reuse. Modules also allow us to control the privacy of items, because code within a module is private by default. Private items are internal implementation details not available for outside use. We can choose to make modules and the items within them public, which exposes them to allow external code to use and depend on them.

As an example, let’s write a library crate that provides the functionality of a restaurant. We’ll define the signatures of functions but leave their bodies empty to concentrate on the organization of the code, rather than the implementation of a restaurant.

In the restaurant industry, some parts of a restaurant are referred to as front of house and others as back of house. Front of house is where customers are; this encompasses where the hosts seat customers, servers take orders and payment, and bartenders make drinks. Back of house is where the chefs and cooks work in the kitchen, dishwashers clean up, and managers do administrative work.

To structure our crate in this way, we can organize its functions into nested modules. Create a new package named restaurant by running scarb new restaurant; then enter the code in Listing 7-1 into src/lib.cairo to define some modules and function signatures. Here’s the front of house section:

Filename: src/lib.cairo

mod front_of_house {
    mod hosting {
        fn add_to_waitlist() {}

        fn seat_at_table() {}
    }

    mod serving {
        fn take_order() {}

        fn serve_order() {}

        fn take_payment() {}
    }
}

Listing 7-1: A front_of_house module containing other modules that then contain functions

We define a module with the mod keyword followed by the name of the module (in this case, front_of_house). The body of the module then goes inside curly brackets. Inside modules, we can place other modules, as in this case with the modules hosting and serving. Modules can also hold definitions for other items, such as structs, enums, constants, traits, and functions.

By using modules, we can group related definitions together and name why they’re related. Programmers using this code can navigate the code based on the groups rather than having to read through all the definitions, making it easier to find the definitions relevant to them. Programmers adding new functionality to this code would know where to place the code to keep the program organized.

Earlier, we mentioned that src/lib.cairo is called the crate root. The reason for this name is that the content of this file forms a module named after the crate name at the root of the crate’s module structure, known as the module tree.

Listing 7-2 shows the module tree for the structure in Listing 7-1.

restaurant
 └── front_of_house
     ├── hosting
     │   ├── add_to_waitlist
     │   └── seat_at_table
     └── serving
         ├── take_order
         ├── serve_order
         └── take_payment

Listing 7-2: The module tree for the code in Listing 7-1

This tree shows how some of the modules nest inside one another; for example, hosting nests inside front_of_house. The tree also shows that some modules are siblings to each other, meaning they’re defined in the same module; hosting and serving are siblings defined within front_of_house. If module A is contained inside module B, we say that module A is the child of module B and that module B is the parent of module A. Notice that the entire module tree is rooted under the explicit name of the crate restaurant.

The module tree might remind you of the filesystem’s directory tree on your computer; this is a very apt comparison! Just like directories in a filesystem, you use modules to organize your code. And just like files in a directory, we need a way to find our modules.

Paths for Referring to an Item in the Module Tree

To show Cairo where to find an item in a module tree, we use a path in the same way we use a path when navigating a filesystem. To call a function, we need to know its path.

A path can take two forms:

• An absolute path is the full path starting from a crate root. The absolute path begins with the crate name.
• A relative path starts from the current module.

Both absolute and relative paths are followed by one or more identifiers separated by double colons (::).

To illustrate this notion let's take back our example Listing 7-1 for the restaurant we used in the last chapter. We have a crate named restaurant in which we have a module named front_of_house that contains a module named hosting. The hosting module contains a function named add_to_waitlist. We want to call the add_to_waitlist function from the eat_at_restaurant function. We need to tell Cairo the path to the add_to_waitlist function so it can find it.

Filename: src/lib.cairo

mod front_of_house {
    mod hosting {
        fn add_to_waitlist() {}

        fn seat_at_table() {}
    }

    mod serving {
        fn take_order() {}

        fn serve_order() {}

        fn take_payment() {}
    }
}


pub fn eat_at_restaurant() {
    // Absolute path
    crate::front_of_house::hosting::add_to_waitlist();

    // Relative path
    front_of_house::hosting::add_to_waitlist();
}

Listing 7-3: Calling the add_to_waitlist function using absolute and relative paths

The eat_at_restaurant function is part of our library's public API, so we mark it with the pub keyword. We’ll go into more detail about pub in the "Exposing Paths with the pub Keyword" section.

The first time we call the add_to_waitlist function in eat_at_restaurant, we use an absolute path. The add_to_waitlist function is defined in the same crate as eat_at_restaurant. In Cairo, absolute paths start from the crate root, which you need to refer to by using the crate name. You can imagine a filesystem with the same structure: we’d specify the path /front_of_house/hosting/add_to_waitlist to run the add_to_waitlist program; using the crate name to start from the crate root is like using a slash (/) to start from the filesystem root in your shell.

The second time we call add_to_waitlist, we use a relative path. The path starts with front_of_house, the name of the module defined at the same level of the module tree as eat_at_restaurant. Here the filesystem equivalent would be using the path ./front_of_house/hosting/add_to_waitlist. Starting with a module name means that the path is relative to the current module.

Let’s try to compile Listing 7-3 and find out why it won’t compile yet! We get the following error:

$ scarb cairo-run 
   Compiling listing_07_02 v0.1.0 (listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_02_paths/Scarb.toml)
error: Item `listing_07_02::front_of_house::hosting` is not visible in this context.
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_02_paths/src/lib.cairo:22:28
    crate::front_of_house::hosting::add_to_waitlist();
                           ^*****^

error: Item `listing_07_02::front_of_house::hosting::add_to_waitlist` is not visible in this context.
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_02_paths/src/lib.cairo:22:37
    crate::front_of_house::hosting::add_to_waitlist();
                                    ^*************^

error: Item `listing_07_02::front_of_house::hosting` is not visible in this context.
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_02_paths/src/lib.cairo:25:21
    front_of_house::hosting::add_to_waitlist();
                    ^*****^

error: Item `listing_07_02::front_of_house::hosting::add_to_waitlist` is not visible in this context.
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_02_paths/src/lib.cairo:25:30
    front_of_house::hosting::add_to_waitlist();
                             ^*************^

error: could not compile `listing_07_02` due to previous error
error: `scarb metadata` exited with error

The error messages say that module hosting and the add_to_waitlist function are not visible. In other words, we have the correct paths for the hosting module and the add_to_waitlist function, but Cairo won’t let us use them because it doesn’t have access to them. In Cairo, all items (functions, methods, structs, enums, modules, and constants) are private to parent modules by default. If you want to make an item like a function or struct private, you put it in a module.

Items in a parent module can’t use the private items inside child modules, but items in child modules can use the items in their ancestor modules. This is because child modules wrap and hide their implementation details, but the child modules can see the context in which they’re defined. To continue with our metaphor, think of the privacy rules as being like the back office of a restaurant: what goes on in there is private to restaurant customers, but office managers can see and do everything in the restaurant they operate.

Cairo chose to have the module system function this way so that hiding inner implementation details is the default. That way, you know which parts of the inner code you can change without breaking outer code. However, Cairo does give you the option to expose inner parts of child modules’ code to outer ancestor modules by using the pub keyword to make an item public.

Exposing Paths with the pub Keyword

Let’s return to the previous error that told us the hosting module and the add_to_waitlist function are not visible. We want the eat_at_restaurant function in the parent module to have access to the add_to_waitlist function in the child module, so we mark the hosting module with the pub keyword, as shown in Listing 7-4.

Filename: src/lib.cairo

mod front_of_house {
    pub mod hosting {
        fn add_to_waitlist() {}
    }
}

pub fn eat_at_restaurant() {
    // Absolute path
    crate::front_of_house::hosting::add_to_waitlist();

    // Relative path
    front_of_house::hosting::add_to_waitlist();
}

Listing 7-4: Declaring the hosting module as pub to use it from eat_at_restaurant

Unfortunately, the code in Listing 7-4 still results in an error.

What happened? Adding the pub keyword in front of mod hosting; makes the module public. With this change, if we can access front_of_house, we can access hosting. But the contents of hosting are still private; making the module public doesn’t make its contents public. The pub keyword on a module only lets code in its ancestor modules refer to it, not access its inner code. Because modules are containers, there’s not much we can do by only making the module public; we need to go further and choose to make one or more of the items within the module public as well.

Let’s also make the add_to_waitlist function public by adding the pub keyword before its definition, as in Listing 7-5.

Filename: src/lib.cairo

mod front_of_house {
    pub mod hosting {
        pub fn add_to_waitlist() {}
    }
}

pub fn eat_at_restaurant() {
    // Absolute path
    crate::front_of_house::hosting::add_to_waitlist(); // ✅ Compiles

    // Relative path
    front_of_house::hosting::add_to_waitlist(); // ✅ Compiles
}

Listing 7-5: Declaring the hosting module as pub to use it from eat_at_restaurant

Now the code will compile! To see why adding the pub keyword lets us use these paths in add_to_waitlist with respect to the privacy rules, let’s look at the absolute and the relative paths.

In the absolute path, we start with the crate root, the root of our crate’s module tree. The front_of_house module is defined in the crate root. While front_of_house isn’t public, because the eat_at_restaurant function is defined in the same module as front_of_house (that is, front_of_house and eat_at_restaurant are siblings), we can refer to front_of_house from eat_at_restaurant. Next is the hosting module marked with pub. We can access the parent module of hosting, so we can access hosting itself. Finally, the add_to_waitlist function is marked with pub and we can access its parent module, so this function call works!

In the relative path, the logic is the same as the absolute path except for the first step: rather than starting from the crate root, the path starts from front_of_house. The front_of_house module is defined within the same module as eat_at_restaurant, so the relative path starting from the module in which eat_at_restaurant is defined works. Then, because hosting and add_to_waitlist are marked with pub, the rest of the path works, and this function call is valid!

Starting Relative Paths with super

We can construct relative paths that begin in the parent module, rather than the current module or the crate root, by using super at the start of the path. This is like starting a filesystem path with the .. syntax. Using super allows us to reference an item that we know is in the parent module, which can make rearranging the module tree easier when the module is closely related to the parent, but the parent might be moved elsewhere in the module tree someday.

Consider the code in Listing 7-6 that models the situation in which a chef fixes an incorrect order and personally brings it out to the customer. The function fix_incorrect_order defined in the back_of_house module calls the function deliver_order defined in the parent module by specifying the path to deliver_order starting with super:

Filename: src/lib.cairo

fn deliver_order() {}

mod back_of_house {
    fn fix_incorrect_order() {
        cook_order();
        super::deliver_order();
    }

    fn cook_order() {}
}

Listing 7-6: Calling a function using a relative path starting with super

Here you can see directly that you access a parent's module easily using super, which wasn't the case previously. Note that the back_of_house is kept private, as external users are not supposed to interact with the back of house directly.

Making Structs and Enums Public

We can also use pub to designate structs and enums as public, but there are a few extra details to consider when using pub with structs and enums.

• If we use pub before a struct definition, we make the struct public, but the struct’s fields will still be private. We can make each field public or not on a case-by-case basis.
• In contrast, if we make an enum public, all of its variants are then public. We only need the pub before the enum keyword.

There’s one more situation involving pub that we haven’t covered, and that is our last module system feature: the use keyword. We’ll cover use by itself first, and then we’ll show how to combine pub and use.

Bringing Paths into Scope with the use Keyword

Having to write out the paths to call functions can feel inconvenient and repetitive. Fortunately, there’s a way to simplify this process: we can create a shortcut to a path with the use keyword once, and then use the shorter name everywhere else in the scope.

In Listing 7-7, we bring the crate::front_of_house::hosting module into the scope of the eat_at_restaurant function so we only have to specify hosting::add_to_waitlist to call the add_to_waitlist function in eat_at_restaurant.

Filename: src/lib.cairo

// section "Defining Modules to Control Scope"

mod front_of_house {
    pub mod hosting {
        pub fn add_to_waitlist() {}
    }
}

use crate::front_of_house::hosting;

pub fn eat_at_restaurant() {
    hosting::add_to_waitlist(); // ✅ Shorter path
}

Listing 7-7: Bringing a module into scope with use

Adding use and a path in a scope is similar to creating a symbolic link in the filesystem. By adding use crate::front_of_house::hosting; in the crate root, hosting is now a valid name in that scope, just as though the hosting module had been defined in the crate root.

Note that use only creates the shortcut for the particular scope in which the use occurs. Listing 7-8 moves the eat_at_restaurant function into a new child module named customer, which is then a different scope than the use statement, so the function body won’t compile:

Filename: src/lib.cairo

mod front_of_house {
    pub mod hosting {
        pub fn add_to_waitlist() {}
    }
}

use crate::front_of_house::hosting;

mod customer {
    pub fn eat_at_restaurant() {
        hosting::add_to_waitlist();
    }
}

Listing 7-8: A use statement only applies in the scope it’s in.

The compiler error shows that the shortcut no longer applies within the customer module:

$ scarb build 
   Compiling listing_07_05 v0.1.0 (listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_07_use_and_scope/Scarb.toml)
warn: Unused import: `listing_07_05::hosting`
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_07_use_and_scope/src/lib.cairo:10:28
use crate::front_of_house::hosting;
                           ^*****^

error: Identifier not found.
 --> listings/ch07-managing-cairo-projects-with-packages-crates-and-modules/listing_07_use_and_scope/src/lib.cairo:14:9
        hosting::add_to_waitlist();
        ^*****^

error: could not compile `listing_07_05` due to previous error

Creating Idiomatic use Paths

In Listing 7-7, you might have wondered why we specified use crate::front_of_house::hosting and then called hosting::add_to_waitlist in eat_at_restaurant rather than specifying the use path all the way out to the add_to_waitlist function to achieve the same result, as in Listing 7-9.

Filename: src/lib.cairo

mod front_of_house {
    pub mod hosting {
        pub fn add_to_waitlist() {}
    }
}

use crate::front_of_house::hosting::add_to_waitlist;

pub fn eat_at_restaurant() {
    add_to_waitlist();
}

Listing 7-9: Bringing the add_to_waitlist function into scope with use, which is unidiomatic

Although both Listing 7-7 and 7-9 accomplish the same task, Listing 7-7 is the idiomatic way to bring a function into scope with use. Bringing the function’s parent module into scope with use means we have to specify the parent module when calling the function. Specifying the parent module when calling the function makes it clear that the function isn’t locally defined while still minimizing repetition of the full path. The code in Listing 7-9 is unclear as to where add_to_waitlist is defined.

On the other hand, when bringing in structs, enums, traits, and other items with use, it’s idiomatic to specify the full path. Listing 7-10 shows the idiomatic way to bring the core library’s BitSize trait into the scope, allowing to call bits method to retrieve the size in bits of a type.

use core::num::traits::BitSize;

fn main() {
    let u8_size: usize = BitSize::<u8>::bits();
    println!("A u8 variable has {} bits", u8_size)
}

Listing 7-10: Bringing BitSize trait into scope in an idiomatic way

There’s no strong reason behind this idiom: it’s just the convention that has emerged in the Rust community, and folks have gotten used to reading and writing Rust code this way. As Cairo shares many idioms with Rust, we follow this convention as well.

The exception to this idiom is if we’re bringing two items with the same name into scope with use statements, because Cairo doesn’t allow that.

Providing New Names with the as Keyword

There’s another solution to the problem of bringing two types of the same name into the same scope with use: after the path, we can specify as and a new local name, or alias, for the type. Listing 7-11 shows how you can rename an import with as:

Filename: src/lib.cairo

use core::array::ArrayTrait as Arr;

fn main() {
    let mut arr = Arr::new(); // ArrayTrait was renamed to Arr
    arr.append(1);
}

Listing 7-11: Renaming a trait when it’s brought into scope with the as keyword

Here, we brought ArrayTrait into scope with the alias Arr. We can now access the trait's methods with the Arr identifier.

Importing Multiple Items from the Same Module

When you want to import multiple items (like functions, structs or enums) from the same module in Cairo, you can use curly braces {} to list all of the items that you want to import. This helps to keep your code clean and easy to read by avoiding a long list of individual use statements.

The general syntax for importing multiple items from the same module is:

use module::{item1, item2, item3};

Here is an example where we import three structures from the same module:

// Assuming we have a module called `shapes` with the structures `Square`, `Circle`, and `Triangle`.
mod shapes {
    #[derive(Drop)]
    pub struct Square {
        pub side: u32
    }

    #[derive(Drop)]
    pub struct Circle {
        pub radius: u32
    }

    #[derive(Drop)]
    pub struct Triangle {
        pub base: u32,
        pub height: u32,
    }
}

// We can import the structures `Square`, `Circle`, and `Triangle` from the `shapes` module like
// this:
use shapes::{Square, Circle, Triangle};

// Now we can directly use `Square`, `Circle`, and `Triangle` in our code.
fn main() {
    let sq = Square { side: 5 };
    let cr = Circle { radius: 3 };
    let tr = Triangle { base: 5, height: 2 };
    // ...
}

Listing 7-12: Importing multiple items from the same module

Re-exporting Names in Module Files

When we bring a name into scope with the use keyword, the name available in the new scope can be imported as if it had been defined in that code’s scope. This technique is called re-exporting because we’re bringing an item into scope, but also making that item available for others to bring into their scope, with the pub keyword.

For example, let's re-export the add_to_waitlist function in the restaurant example:

Filename: src/lib.cairo

mod front_of_house {
    pub mod hosting {
        pub fn add_to_waitlist() {}
    }
}

pub use crate::front_of_house::hosting;

fn eat_at_restaurant() {
    hosting::add_to_waitlist();
}

Listing 7-13: Making a name available for any code to use from a new scope with pub use

Before this change, external code would have to call the add_to_waitlist function by using the path restaurant::front_of_house::hosting::add_to_waitlist(). Now that this pub use has re-exported the hosting module from the root module, external code can now use the path restaurant::hosting::add_to_waitlist() instead.

Re-exporting is useful when the internal structure of your code is different from how programmers calling your code would think about the domain. For example, in this restaurant metaphor, the people running the restaurant think about “front of house” and “back of house.” But customers visiting a restaurant probably won’t think about the parts of the restaurant in those terms. With pub use, we can write our code with one structure but expose a different structure. Doing so makes our library well organized for programmers working on the library and programmers calling the library.

Using External Packages in Cairo with Scarb

You might need to use external packages to leverage the functionality provided by the community. Scarb allows you to use dependencies by cloning packages from their Git repositories. To use an external package in your project with Scarb, simply declare the Git repository URL of the dependency you want to add in a dedicated [dependencies] section in your Scarb.toml configuration file. Note that the URL might correspond to the main branch, or any specific commit, branch or tag. For this, you will have to pass an extra rev, branch, or tag field, respectively. For example, the following code imports the main branch of alexandria_math crate from alexandria package:

[dependencies]
alexandria_math = { git = "https://github.com/keep-starknet-strange/alexandria.git" }

while the following code imports a specific branch (which is deprecated and should not be used):

[dependencies]
alexandria_math = { git = "https://github.com/keep-starknet-strange/alexandria.git", branch = "cairo-v2.3.0-rc0" }

If you want to import multiple packages in your project, you need to create only one [dependencies] section and list all the desired packages beneath it. You can also specify development dependencies by declaring a [dev-dependencies] section.

After that, simply run scarb build to fetch all external dependencies and compile your package with all the dependencies included.

Note that it is also possible to add dependencies with the scarb add command, which will automatically edit the Scarb.toml file for you. For development dependencies, just use the scarb add --dev command.

To remove a dependency, simply remove the corresponding line from your Scarb.toml file, or use the scarb rm command.

Separating Modules into Different Files

So far, all the examples in this chapter defined multiple modules in one file. When modules get large, you might want to move their definitions to a separate file to make the code easier to navigate.

For example, let’s start from the code in Listing 7-7 that had multiple restaurant modules. We’ll extract modules into files instead of having all the modules defined in the crate root file. In this case, the crate root file is src/lib.cairo.

First, we’ll extract the front_of_house module to its own file. Remove the code inside the curly brackets for the front_of_house module, leaving only the mod front_of_house; declaration, so that src/lib.cairo contains the code shown in Listing 7-14. Note that this won’t compile until we create the src/front_of_house.cairo file.

Filename: src/lib.cairo

mod front_of_house;

use crate::front_of_house::hosting;

fn eat_at_restaurant() {
    hosting::add_to_waitlist();
}

Listing 7-14: Declaring the front_of_house module whose body will be in src/front_of_house.cairo

Next, place the code that was in the curly brackets into a new file named src/front_of_house.cairo, as shown in Listing 7-15. The compiler knows to look in this file because it came across the module declaration in the crate root with the name front_of_house.

Filename: src/front_of_house.cairo

pub mod hosting {
    pub fn add_to_waitlist() {}
}

Listing 7-15: Definitions inside the front_of_house module in src/front_of_house.cairo

Note that you only need to load a file using a mod declaration once in your module tree. Once the compiler knows the file is part of the project (and knows where in the module tree the code resides because of where you’ve put the mod statement), other files in your project should refer to the loaded file’s code using a path to where it was declared, as covered in the "Paths for Referring to an Item in the Module Tree" chapter. In other words, mod is not an “include” operation that you may have seen in other programming languages.

Next, we’ll extract the hosting module to its own file. The process is a bit different because hosting is a child module of front_of_house, not of the root module. We’ll place the file for hosting in a new directory that will be named for its ancestors in the module tree, in this case src/front_of_house/.

To start moving hosting, we change src/front_of_house.cairo to contain only the declaration of the hosting module:

Filename: src/front_of_house.cairo

pub mod hosting;

Then we create a src/front_of_house directory and a file hosting.cairo to contain the definitions made in the hosting module:

Filename: src/front_of_house/hosting.cairo

pub fn add_to_waitlist() {}

If we instead put hosting.cairo in the src directory, the compiler would expect the hosting.cairo code to be in a hosting module declared in the crate root, and not declared as a child of the front_of_house module. The compiler’s rules for which files to check for which modules’ code means the directories and files more closely match the module tree.

We’ve moved each module’s code to a separate file, and the module tree remains the same. The function calls in eat_at_restaurant will work without any modification, even though the definitions live in different files. This technique lets you move modules to new files as they grow in size.

Note that the use crate::front_of_house::hosting; statement in src/lib.cairo also hasn’t changed, nor does use have any impact on what files are compiled as part of the crate. The mod keyword declares modules, and Cairo looks in a file with the same name as the module for the code that goes into that module.

Summary

Cairo lets you split a package into multiple crates and a crate into modules so you can refer to items defined in one module from another module. You can do this by specifying absolute or relative paths. These paths can be brought into scope with a use statement so you can use a shorter path for multiple uses of the item in that scope. Module code is private by default.

Generic Types and Traits

Every programming language has tools for effectively handling the duplication of concepts. In Cairo, one such tool is generics: abstract stand-ins for concrete types or other properties. We can express the behavior of generics or how they relate to other generics without knowing what will be in their place when compiling and running the code.

Functions can take parameters of some generic type, instead of a concrete type like u32 or bool, in the same way a function takes parameters with unknown values to run the same code on multiple concrete values. In fact, we’ve already used generics in Chapter 6 with Option<T>.

In this chapter, you’ll explore how to define your own types, functions, and traits with generics.

Generics allow us to replace specific types with a placeholder that represents multiple types to remove code duplication. Upon compilation, the compiler creates a new definition for each concrete type that replaces a generic type, reducing development time for the programmer, but code duplication at compile level still exists. This may be of importance if you are writing Starknet contracts and using a generic for multiple types which will cause contract size to increment.

Then you’ll learn how to use traits to define behavior in a generic way. You can combine traits with generic types to constrain a generic type to accept only those types that have a particular behavior, as opposed to just any type.

Removing Duplication by Extracting a Function

Generics allow us to replace specific types with a placeholder that represents multiple types to remove code duplication. Before diving into generics syntax, let’s first look at how to remove duplication in a way that doesn’t involve generic types by extracting a function that replaces specific values with a placeholder that represents multiple values. Then we’ll apply the same technique to extract a generic function! By learning how to identify duplicated code that can be extracted into a function, you'll start to recognize instances where generics can be used to reduce duplication.

We begin with a short program that finds the largest number in an array of u8:

fn main() {
    let mut number_list: Array<u8> = array![34, 50, 25, 100, 65];

    let mut largest = number_list.pop_front().unwrap();

    while let Option::Some(number) = number_list.pop_front() {
        if number > largest {
            largest = number;
        }
    };

    println!("The largest number is {}", largest);
}

We store an array of u8 in the variable number_list and extract the first number in the array in a variable named largest. We then iterate through all the numbers in the array, and if the current number is greater than the number stored in largest, we update the value of largest. However, if the current number is less than or equal to the largest number seen so far, the variable doesn’t change, and the code moves on to the next number in the list. After considering all the numbers in the array, largest should contain the largest number, which in this case is 100.

We've now been tasked with finding the largest number in two different arrays of numbers. To do so, we can choose to duplicate the previous code and use the same logic at two different places in the program, as follows:

fn main() {
    let mut number_list: Array<u8> = array![34, 50, 25, 100, 65];

    let mut largest = number_list.pop_front().unwrap();

    while let Option::Some(number) = number_list.pop_front() {
        if number > largest {
            largest = number;
        }
    };

    println!("The largest number is {}", largest);

    let mut number_list: Array<u8> = array![102, 34, 255, 89, 54, 2, 43, 8];

    let mut largest = number_list.pop_front().unwrap();

    while let Option::Some(number) = number_list.pop_front() {
        if number > largest {
            largest = number;
        }
    };

    println!("The largest number is {}", largest);
}

Although this code works, duplicating code is tedious and error-prone. We also have to remember to update the code in multiple places when we want to change it.

To eliminate this duplication, we’ll create an abstraction by defining a function that operates on any array of u8 passed in a parameter. This solution makes our code clearer and lets us express the concept of finding the largest number in an array abstractly.

To do that, we extract the code that finds the largest number into a function named largest. Then we call the function to find the largest number in the two arrays. We could also use the function on any other array of u8 values we might have in the future.

fn largest(ref number_list: Array<u8>) -> u8 {
    let mut largest = number_list.pop_front().unwrap();

    while let Option::Some(number) = number_list.pop_front() {
        if number > largest {
            largest = number;
        }
    };

    largest
}

fn main() {
    let mut number_list = array![34, 50, 25, 100, 65];

    let result = largest(ref number_list);
    println!("The largest number is {}", result);

    let mut number_list = array![102, 34, 255, 89, 54, 2, 43, 8];

    let result = largest(ref number_list);
    println!("The largest number is {}", result);
}

The largest function has a parameter called number_list, passed by reference, which represents any concrete array of u8 values we might pass into the function. As a result, when we call the function, the code runs on the specific values that we pass in.

In summary, here are the steps we took to change the code:

• Identify duplicate code.
• Extract the duplicate code into the body of the function and specify the inputs and return values of that code in the function signature.
• Update the two instances of duplicated code to call the function instead.

Next, we’ll use these same steps with generics to reduce code duplication. In the same way that the function body can operate on an abstract Array<T> instead of specific u8 values, generics allow code to operate on abstract types.

Generic Data Types

We use generics to create definitions for item declarations, such as structs and functions, which we can then use with many different concrete data types. In Cairo, we can use generics when defining functions, structs, enums, traits, implementations and methods. In this chapter, we are going to take a look at how to effectively use generic types with all of them.

Generics allow us to write reusable code that works with many types, thus avoiding code duplication, while enhancing code maintainability.

Generic Functions

Making a function generic means it can operate on different types, avoiding the need for multiple, type-specific implementations. This leads to significant code reduction and increases the flexibility of the code.

When defining a function that uses generics, we place the generics in the function signature, where we would usually specify the data types of the parameter and return value. For example, imagine we want to create a function which given two Array of items, will return the largest one. If we need to perform this operation for lists of different types, then we would have to redefine the function each time. Luckily we can implement the function once using generics and move on to other tasks.

// Specify generic type T between the angulars
fn largest_list<T>(l1: Array<T>, l2: Array<T>) -> Array<T> {
    if l1.len() > l2.len() {
        l1
    } else {
        l2
    }
}

fn main() {
    let mut l1 = array![1, 2];
    let mut l2 = array![3, 4, 5];

    // There is no need to specify the concrete type of T because
    // it is inferred by the compiler
    let l3 = largest_list(l1, l2);
}

The largest_list function compares two lists of the same type and returns the one with more elements and drops the other. If you compile the previous code, you will notice that it will fail with an error saying that there are no traits defined for dropping an array of a generic type. This happens because the compiler has no way to guarantee that an Array<T> is droppable when executing the main function. In order to drop an array of T, the compiler must first know how to drop T. This can be fixed by specifying in the function signature of largest_list that T must implement the Drop trait. The correct function definition of largest_list is as follows:

fn largest_list<T, impl TDrop: Drop<T>>(l1: Array<T>, l2: Array<T>) -> Array<T> {
    if l1.len() > l2.len() {
        l1
    } else {
        l2
    }
}

The new largest_list function includes in its definition the requirement that whatever generic type is placed there, it must be droppable. This is what we call trait bounds. The main function remains unchanged, the compiler is smart enough to deduce which concrete type is being used and if it implements the Drop trait.

Constraints for Generic Types

When defining generic types, it is useful to have information about them. Knowing which traits a generic type implements allows us to use it more effectively in a function's logic at the cost of constraining the generic types that can be used with the function. We saw an example of this previously by adding the TDrop implementation as part of the generic arguments of largest_list. While TDrop was added to satisfy the compiler's requirements, we can also add constraints to benefit our function logic.

Imagine that we want, given a list of elements of some generic type T, to find the smallest element among them. Initially, we know that for an element of type T to be comparable, it must implement the PartialOrd trait. The resulting function would be:

// Given a list of T get the smallest one
// The PartialOrd trait implements comparison operations for T
fn smallest_element<T, impl TPartialOrd: PartialOrd<T>>(list: @Array<T>) -> T {
    // This represents the smallest element through the iteration
    // Notice that we use the desnap (*) operator
    let mut smallest = *list[0];

    // The index we will use to move through the list
    let mut index = 1;

    // Iterate through the whole list storing the smallest
    while index < list.len() {
        if *list[index] < smallest {
            smallest = *list[index];
        }
        index = index + 1;
    };

    smallest
}

fn main() {
    let list: Array<u8> = array![5, 3, 10];

    // We need to specify that we are passing a snapshot of `list` as an argument
    let s = smallest_element(@list);
    assert!(s == 3);
}

The smallest_element function uses a generic type T that implements the PartialOrd trait, takes a snapshot of an Array<T> as a parameter and returns a copy of the smallest element. Because the parameter is of type @Array<T>, we no longer need to drop it at the end of the execution and so we are not required to implement the Drop trait for T as well. Why does it not compile then?

When indexing on list, the value results in a snap of the indexed element, and unless PartialOrd is implemented for @T we need to desnap the element using *. The * operation requires a copy from @T to T, which means that T needs to implement the Copy trait. After copying an element of type @T to T, there are now variables with type T that need to be dropped, requiring T to implement the Drop trait as well. We must then add both Drop and Copy traits implementation for the function to be correct. After updating the smallest_element function the resulting code would be:

fn smallest_element<T, impl TPartialOrd: PartialOrd<T>, impl TCopy: Copy<T>, impl TDrop: Drop<T>>(
    list: @Array<T>
) -> T {
    let mut smallest = *list[0];
    let mut index = 1;

    while index < list.len() {
        if *list[index] < smallest {
            smallest = *list[index];
        }
        index = index + 1;
    };

    smallest
}

Anonymous Generic Implementation Parameter (+ Operator)

Until now, we have always specified a name for each implementation of the required generic trait: TPartialOrd for PartialOrd<T>, TDrop for Drop<T>, and TCopy for Copy<T>.

However, most of the time, we don't use the implementation in the function body; we only use it as a constraint. In these cases, we can use the + operator to specify that the generic type must implement a trait without naming the implementation. This is referred to as an anonymous generic implementation parameter.

For example, +PartialOrd<T> is equivalent to impl TPartialOrd: PartialOrd<T>.

We can rewrite the smallest_element function signature as follows:

fn smallest_element<T, +PartialOrd<T>, +Copy<T>, +Drop<T>>(list: @Array<T>) -> T {
    let mut smallest = *list[0];
    let mut index = 1;
    loop {
        if index >= list.len() {
            break smallest;
        }
        if *list[index] < smallest {
            smallest = *list[index];
        }
        index = index + 1;
    }
}

Structs

We can also define structs to use a generic type parameter for one or more fields using the <> syntax, similar to function definitions. First, we declare the name of the type parameter inside the angle brackets just after the name of the struct. Then we use the generic type in the struct definition where we would otherwise specify concrete data types. The next code example shows the definition Wallet<T> which has a balance field of type T.

#[derive(Drop)]
struct Wallet<T> {
    balance: T
}

fn main() {
    let w = Wallet { balance: 3 };
}

The above code derives the Drop trait for the Wallet type automatically. It is equivalent to writing the following code:

struct Wallet<T> {
    balance: T
}

impl WalletDrop<T, +Drop<T>> of Drop<Wallet<T>>;

fn main() {
    let w = Wallet { balance: 3 };
}

We avoid using the derive macro for Drop implementation of Wallet and instead define our own WalletDrop implementation. Notice that we must define, just like functions, an additional generic type for WalletDrop saying that T implements the Drop trait as well. We are basically saying that the struct Wallet<T> is droppable as long as T is also droppable.

Finally, if we want to add a field to Wallet representing its address and we want that field to be different than T but generic as well, we can simply add another generic type between the <>:

#[derive(Drop)]
struct Wallet<T, U> {
    balance: T,
    address: U,
}

fn main() {
    let w = Wallet { balance: 3, address: 14 };
}

We add to the Wallet struct definition a new generic type U and then assign this type to the new field member address. Notice that the derive attribute for the Drop trait works for U as well.

Enums

As we did with structs, we can define enums to hold generic data types in their variants. For example the Option<T> enum provided by the Cairo core library:

enum Option<T> {
    Some: T,
    None,
}

The Option<T> enum is generic over a type T and has two variants: Some, which holds one value of type T and None that doesn't hold any value. By using the Option<T> enum, it is possible for us to express the abstract concept of an optional value and because the value has a generic type T we can use this abstraction with any type.

Enums can use multiple generic types as well, like the definition of the Result<T, E> enum that the core library provides:

enum Result<T, E> {
    Ok: T,
    Err: E,
}

The Result<T, E> enum has two generic types, T and E, and two variants: Ok which holds the value of type T and Err which holds the value of type E. This definition makes it convenient to use the Result enum anywhere we have an operation that might succeed (by returning a value of type T) or fail (by returning a value of type E).

Generic Methods

We can implement methods on structs and enums, and use the generic types in their definitions, too. Using our previous definition of Wallet<T> struct, we define a balance method for it:

#[derive(Copy, Drop)]
struct Wallet<T> {
    balance: T
}

trait WalletTrait<T> {
    fn balance(self: @Wallet<T>) -> T;
}

impl WalletImpl<T, +Copy<T>> of WalletTrait<T> {
    fn balance(self: @Wallet<T>) -> T {
        return *self.balance;
    }
}

fn main() {
    let w = Wallet { balance: 50 };
    assert!(w.balance() == 50);
}

We first define WalletTrait<T> trait using a generic type T which defines a method that returns the value of the field balance from Wallet. Then we give an implementation for the trait in WalletImpl<T>. Note that you need to include a generic type in both definitions of the trait and the implementation.

We can also specify constraints on generic types when defining methods on the type. We could, for example, implement methods only for Wallet<u128> instances rather than Wallet<T>. In the code example, we define an implementation for wallets which have a concrete type of u128 for the balance field.

#[derive(Copy, Drop)]
struct Wallet<T> {
    balance: T
}

/// Generic trait for wallets
trait WalletTrait<T> {
    fn balance(self: @Wallet<T>) -> T;
}

impl WalletImpl<T, +Copy<T>> of WalletTrait<T> {
    fn balance(self: @Wallet<T>) -> T {
        return *self.balance;
    }
}

/// Trait for wallets of type u128
trait WalletReceiveTrait {
    fn receive(ref self: Wallet<u128>, value: u128);
}

impl WalletReceiveImpl of WalletReceiveTrait {
    fn receive(ref self: Wallet<u128>, value: u128) {
        self.balance += value;
    }
}

fn main() {
    let mut w = Wallet { balance: 50 };
    assert!(w.balance() == 50);

    w.receive(100);
    assert!(w.balance() == 150);
}

The new method receive increments the size of balance of any instance of a Wallet<u128>. Notice that we changed the main function making w a mutable variable in order for it to be able to update its balance. If we were to change the initialization of w by changing the type of balance the previous code wouldn't compile.

Cairo allows us to define generic methods inside generic traits as well. Using the past implementation from Wallet<U, V> we are going to define a trait that picks two wallets of different generic types and creates a new one with a generic type of each. First, let's rewrite the struct definition:

struct Wallet<T, U> {
    balance: T,
    address: U,
}

Next, we are going to naively define the mixup trait and implementation:

// This does not compile!
trait WalletMixTrait<T1, U1> {
    fn mixup<T2, U2>(self: Wallet<T1, U1>, other: Wallet<T2, U2>) -> Wallet<T1, U2>;
}

impl WalletMixImpl<T1, U1> of WalletMixTrait<T1, U1> {
    fn mixup<T2, U2>(self: Wallet<T1, U1>, other: Wallet<T2, U2>) -> Wallet<T1, U2> {
        Wallet { balance: self.balance, address: other.address }
    }
}

We are creating a trait WalletMixTrait<T1, U1> with the mixup<T2, U2> method which given an instance of Wallet<T1, U1> and Wallet<T2, U2> creates a new Wallet<T1, U2>. As mixup signature specifies, both self and other are getting dropped at the end of the function, which is why this code does not compile. If you have been following from the start until now you would know that we must add a requirement for all the generic types specifying that they will implement the Drop trait for the compiler to know how to drop instances of Wallet<T, U>. The updated implementation is as follows:

trait WalletMixTrait<T1, U1> {
    fn mixup<T2, +Drop<T2>, U2, +Drop<U2>>(
        self: Wallet<T1, U1>, other: Wallet<T2, U2>
    ) -> Wallet<T1, U2>;
}

impl WalletMixImpl<T1, +Drop<T1>, U1, +Drop<U1>> of WalletMixTrait<T1, U1> {
    fn mixup<T2, +Drop<T2>, U2, +Drop<U2>>(
        self: Wallet<T1, U1>, other: Wallet<T2, U2>
    ) -> Wallet<T1, U2> {
        Wallet { balance: self.balance, address: other.address }
    }
}

We add the requirements for T1 and U1 to be droppable on WalletMixImpl declaration. Then we do the same for T2 and U2, this time as part of mixup signature. We can now try the mixup function:

fn main() {
    let w1: Wallet<bool, u128> = Wallet { balance: true, address: 10 };
    let w2: Wallet<felt252, u8> = Wallet { balance: 32, address: 100 };

    let w3 = w1.mixup(w2);

    assert!(w3.balance);
    assert!(w3.address == 100);
}

We first create two instances: one of Wallet<bool, u128> and the other of Wallet<felt252, u8>. Then, we call mixup and create a new Wallet<bool, u8> instance.

Traits in Cairo

A trait defines a set of methods that can be implemented by a type. These methods can be called on instances of the type when this trait is implemented. A trait combined with a generic type defines functionality a particular type has and can share with other types. We can use traits to define shared behavior in an abstract way. We can use trait bounds to specify that a generic type can be any type that has certain behavior.

Note: Traits are similar to a feature often called interfaces in other languages, although with some differences.

While traits can be written to not accept generic types, they are most useful when used with generic types. We already covered generics in the previous chapter, and we will use them in this chapter to demonstrate how traits can be used to define shared behavior for generic types.

Defining a Trait

A type’s behavior consists of the methods we can call on that type. Different types share the same behavior if we can call the same methods on all of those types. Trait definitions are a way to group method signatures together to define a set of behaviors necessary to accomplish some purpose.

For example, let’s say we have a struct NewsArticle that holds a news story in a particular location. We can define a trait Summary that describes the behavior of something that can summarize the NewsArticle type.

#[derive(Drop, Clone)]
struct NewsArticle {
    headline: ByteArray,
    location: ByteArray,
    author: ByteArray,
    content: ByteArray,
}

pub trait Summary {
    fn summarize(self: @NewsArticle) -> ByteArray;
}

impl NewsArticleSummary of Summary {
    fn summarize(self: @NewsArticle) -> ByteArray {
        format!("{:?} by {:?} ({:?})", self.headline, self.author, self.location)
    }
}

Listing 8-1: A Summary trait that consists of the behavior provided by a summarize method

In Listing 8-1, we declare a trait using the trait keyword and then the trait’s name, which is Summary in this case. We’ve also declared the trait as pub so that crates depending on this crate can make use of this trait too, as we’ll see in a few examples.

Inside the curly brackets, we declare the method signatures that describe the behaviors of the types that implement this trait, which in this case is fn summarize(self: @NewsArticle) -> ByteArray;. After the method signature, instead of providing an implementation within curly brackets, we use a semicolon.

Note: the ByteArray type is the type used to represent strings in Cairo.

As the trait is not generic, the self parameter is not generic either and is of type @NewsArticle. This means that the summarize method can only be called on instances of NewsArticle.

Now, consider that we want to make a media aggregator library crate named aggregator that can display summaries of data that might be stored in a NewsArticle or Tweet instance. To do this, we need a summary from each type, and we’ll request that summary by calling a summarize method on an instance of that type. By defining the Summary trait on generic type T, we can implement the summarize method on any type we want to be able to summarize.

mod aggregator {
    pub trait Summary<T> {
        fn summarize(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct NewsArticle {
        pub headline: ByteArray,
        pub location: ByteArray,
        pub author: ByteArray,
        pub content: ByteArray,
    }

    impl NewsArticleSummary of Summary<NewsArticle> {
        fn summarize(self: @NewsArticle) -> ByteArray {
            format!("{} by {} ({})", self.headline, self.author, self.location)
        }
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize(self: @Tweet) -> ByteArray {
            format!("{}: {}", self.username, self.content)
        }
    }
}

use aggregator::{Summary, NewsArticle, Tweet};

fn main() {
    let news = NewsArticle {
        headline: "Cairo has become the most popular language for developers",
        location: "Worldwide",
        author: "Cairo Digger",
        content: "Cairo is a new programming language for zero-knowledge proofs",
    };

    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    }; // Tweet instantiation

    println!("New article available! {}", news.summarize());
    println!("New tweet! {}", tweet.summarize());
}

Listing 8-2: A Summary trait that consists of the behavior provided by a summarize method for a generic type

Each type implementing this trait must provide its own custom behavior for the body of the method. The compiler will enforce that any type that implements the Summary trait will have the method summarize defined with this signature exactly.

A trait can have multiple methods in its body: the method signatures are listed one per line and each line ends in a semicolon.

Implementing a Trait on a Type

Now that we’ve defined the desired signatures of the Summary trait’s methods, we can implement it on the types in our media aggregator. The following code shows an implementation of the Summary trait on the NewsArticle struct that uses the headline, the author, and the location to create the return value of summarize. For the Tweet struct, we define summarize as the username followed by the entire text of the tweet, assuming that tweet content is already limited to 280 characters.

mod aggregator {
    pub trait Summary<T> {
        fn summarize(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct NewsArticle {
        pub headline: ByteArray,
        pub location: ByteArray,
        pub author: ByteArray,
        pub content: ByteArray,
    }

    impl NewsArticleSummary of Summary<NewsArticle> {
        fn summarize(self: @NewsArticle) -> ByteArray {
            format!("{} by {} ({})", self.headline, self.author, self.location)
        }
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize(self: @Tweet) -> ByteArray {
            format!("{}: {}", self.username, self.content)
        }
    }
}

use aggregator::{Summary, NewsArticle, Tweet};

fn main() {
    let news = NewsArticle {
        headline: "Cairo has become the most popular language for developers",
        location: "Worldwide",
        author: "Cairo Digger",
        content: "Cairo is a new programming language for zero-knowledge proofs",
    };

    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    }; // Tweet instantiation

    println!("New article available! {}", news.summarize());
    println!("New tweet! {}", tweet.summarize());
}

Listing 8-3: Implementation of the Summary trait on NewsArticle and Tweet

Implementing a trait on a type is similar to implementing regular methods. The difference is that after impl, we put a name for the implementation, then use the of keyword, and then specify the name of the trait we are writing the implementation for. If the implementation is for a generic type, we place the generic type name in the angle brackets after the trait name.

Note that for the trait method to be accessible, there must be an implementation of that trait visible from the scope where the method is called. If the trait is pub and the implementation is not, and the implementation is not visible in the scope where the trait method is called, this will cause a compilation error.

Within the impl block, we put the method signatures that the trait definition has defined. Instead of adding a semicolon after each signature, we use curly brackets and fill in the method body with the specific behavior that we want the methods of the trait to have for the particular type.

Now that the library has implemented the Summary trait on NewsArticle and Tweet, users of the crate can call the trait methods on instances of NewsArticle and Tweet in the same way we call regular methods. The only difference is that the user must bring the trait into scope as well as the types. Here’s an example of how a crate could use our aggregator crate:

mod aggregator {
    pub trait Summary<T> {
        fn summarize(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct NewsArticle {
        pub headline: ByteArray,
        pub location: ByteArray,
        pub author: ByteArray,
        pub content: ByteArray,
    }

    impl NewsArticleSummary of Summary<NewsArticle> {
        fn summarize(self: @NewsArticle) -> ByteArray {
            format!("{} by {} ({})", self.headline, self.author, self.location)
        }
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize(self: @Tweet) -> ByteArray {
            format!("{}: {}", self.username, self.content)
        }
    }
}

use aggregator::{Summary, NewsArticle, Tweet};

fn main() {
    let news = NewsArticle {
        headline: "Cairo has become the most popular language for developers",
        location: "Worldwide",
        author: "Cairo Digger",
        content: "Cairo is a new programming language for zero-knowledge proofs",
    };

    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    }; // Tweet instantiation

    println!("New article available! {}", news.summarize());
    println!("New tweet! {}", tweet.summarize());
}

This code prints the following:

$ scarb cairo-run 
   Compiling no_listing_15_traits v0.1.0 (listings/ch08-generic-types-and-traits/no_listing_15_traits/Scarb.toml)
    Finished `dev` profile target(s) in 2 seconds
     Running no_listing_15_traits
New article available! Cairo has become the most popular language for developers by Cairo Digger (Worldwide)
New tweet! EliBenSasson: Crypto is full of short-term maximizing projects. 
 @Starknet and @StarkWareLtd are about long-term vision maximization.
Run completed successfully, returning []

Other crates that depend on the aggregator crate can also bring the Summary trait into scope to implement Summary on their own types.

Default Implementations

Sometimes it’s useful to have default behavior for some or all of the methods in a trait instead of requiring implementations for all methods on every type. Then, as we implement the trait on a particular type, we can keep or override each method’s default behavior.

In Listing 8-5 we specify a default string for the summarize method of the Summary trait instead of only defining the method signature, as we did in Listing 8-2.

Filename: src/lib.cairo

mod aggregator {
    pub trait Summary<T> {
        fn summarize(self: @T) -> ByteArray {
            "(Read more...)"
        }
    }

    #[derive(Drop)]
    pub struct NewsArticle {
        pub headline: ByteArray,
        pub location: ByteArray,
        pub author: ByteArray,
        pub content: ByteArray,
    }

    impl NewsArticleSummary of Summary<NewsArticle> {}

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize(self: @Tweet) -> ByteArray {
            format!("{}: {}", self.username, self.content)
        }
    }
}

use aggregator::{Summary, NewsArticle};

fn main() {
    let news = NewsArticle {
        headline: "Cairo has become the most popular language for developers",
        location: "Worldwide",
        author: "Cairo Digger",
        content: "Cairo is a new programming language for zero-knowledge proofs",
    };

    println!("New article available! {}", news.summarize());
}

Listing 8-5: Defining a Summary trait with a default implementation of the summarize method

To use a default implementation to summarize instances of NewsArticle, we specify an empty impl block with impl NewsArticleSummary of Summary<NewsArticle> {}.

Even though we’re no longer defining the summarize method on NewsArticle directly, we’ve provided a default implementation and specified that NewsArticle implements the Summary trait. As a result, we can still call the summarize method on an instance of NewsArticle, like this:

mod aggregator {
    pub trait Summary<T> {
        fn summarize(self: @T) -> ByteArray {
            "(Read more...)"
        }
    }

    #[derive(Drop)]
    pub struct NewsArticle {
        pub headline: ByteArray,
        pub location: ByteArray,
        pub author: ByteArray,
        pub content: ByteArray,
    }

    impl NewsArticleSummary of Summary<NewsArticle> {}

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize(self: @Tweet) -> ByteArray {
            format!("{}: {}", self.username, self.content)
        }
    }
}

use aggregator::{Summary, NewsArticle};

fn main() {
    let news = NewsArticle {
        headline: "Cairo has become the most popular language for developers",
        location: "Worldwide",
        author: "Cairo Digger",
        content: "Cairo is a new programming language for zero-knowledge proofs",
    };

    println!("New article available! {}", news.summarize());
}

This code prints New article available! (Read more...).

Creating a default implementation doesn’t require us to change anything about the previous implementation of Summary on Tweet. The reason is that the syntax for overriding a default implementation is the same as the syntax for implementing a trait method that doesn’t have a default implementation.

Default implementations can call other methods in the same trait, even if those other methods don’t have a default implementation. In this way, a trait can provide a lot of useful functionality and only require implementors to specify a small part of it. For example, we could define the Summary trait to have a summarize_author method whose implementation is required, and then define a summarize method that has a default implementation that calls the summarize_author method:

mod aggregator {
    pub trait Summary<T> {
        fn summarize(
            self: @T
        ) -> ByteArray {
            format!("(Read more from {}...)", Self::summarize_author(self))
        }
        fn summarize_author(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize_author(self: @Tweet) -> ByteArray {
            format!("@{}", self.username)
        }
    }
}

use aggregator::{Summary, Tweet};

fn main() {
    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    };

    println!("1 new tweet: {}", tweet.summarize());
}

To use this version of Summary, we only need to define summarize_author when we implement the trait on a type:

mod aggregator {
    pub trait Summary<T> {
        fn summarize(
            self: @T
        ) -> ByteArray {
            format!("(Read more from {}...)", Self::summarize_author(self))
        }
        fn summarize_author(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize_author(self: @Tweet) -> ByteArray {
            format!("@{}", self.username)
        }
    }
}

use aggregator::{Summary, Tweet};

fn main() {
    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    };

    println!("1 new tweet: {}", tweet.summarize());
}

After we define summarize_author, we can call summarize on instances of the Tweet struct, and the default implementation of summarize will call the definition of summarize_author that we’ve provided. Because we’ve implemented summarize_author, the Summary trait has given us the behavior of the summarize method without requiring us to write any more code.

mod aggregator {
    pub trait Summary<T> {
        fn summarize(
            self: @T
        ) -> ByteArray {
            format!("(Read more from {}...)", Self::summarize_author(self))
        }
        fn summarize_author(self: @T) -> ByteArray;
    }

    #[derive(Drop)]
    pub struct Tweet {
        pub username: ByteArray,
        pub content: ByteArray,
        pub reply: bool,
        pub retweet: bool,
    }

    impl TweetSummary of Summary<Tweet> {
        fn summarize_author(self: @Tweet) -> ByteArray {
            format!("@{}", self.username)
        }
    }
}

use aggregator::{Summary, Tweet};

fn main() {
    let tweet = Tweet {
        username: "EliBenSasson",
        content: "Crypto is full of short-term maximizing projects. \n @Starknet and @StarkWareLtd are about long-term vision maximization.",
        reply: false,
        retweet: false
    };

    println!("1 new tweet: {}", tweet.summarize());
}

This code prints 1 new tweet: (Read more from @EliBenSasson...).

Note that it isn’t possible to call the default implementation from an overriding implementation of that same method.

Managing and Using External Trait

To use traits methods, you need to make sure the correct traits/implementation(s) are imported. In some cases you might need to import not only the trait but also the implementation if they are declared in separate modules. If CircleGeometry implementation was in a separate module/file named circle, then to define boundary method on Circle struct, we'd need to import ShapeGeometry trait in the circle module.

If the code were to be organized into modules like in Listing 8-6 where the implementation of a trait is defined in a different module than the trait itself, explicitly importing the relevant trait or implementation would be required.

// Here T is an alias type which will be provided during implementation
pub trait ShapeGeometry<T> {
    fn boundary(self: T) -> u64;
    fn area(self: T) -> u64;
}

mod rectangle {
    // Importing ShapeGeometry is required to implement this trait for Rectangle
    use super::ShapeGeometry;

    #[derive(Copy, Drop)]
    pub struct Rectangle {
        pub height: u64,
        pub width: u64,
    }

    // Implementation RectangleGeometry passes in <Rectangle>
    // to implement the trait for that type
    impl RectangleGeometry of ShapeGeometry<Rectangle> {
        fn boundary(self: Rectangle) -> u64 {
            2 * (self.height + self.width)
        }
        fn area(self: Rectangle) -> u64 {
            self.height * self.width
        }
    }
}

mod circle {
    // Importing ShapeGeometry is required to implement this trait for Circle
    use super::ShapeGeometry;

    #[derive(Copy, Drop)]
    pub struct Circle {
        pub radius: u64
    }

    // Implementation CircleGeometry passes in <Circle>
    // to implement the imported trait for that type
    impl CircleGeometry of ShapeGeometry<Circle> {
        fn boundary(self: Circle) -> u64 {
            (2 * 314 * self.radius) / 100
        }
        fn area(self: Circle) -> u64 {
            (314 * self.radius * self.radius) / 100
        }
    }
}

use rectangle::Rectangle;
use circle::Circle;

fn main() {
    let rect = Rectangle { height: 5, width: 7 };
    println!("Rectangle area: {}", ShapeGeometry::area(rect)); //35
    println!("Rectangle boundary: {}", ShapeGeometry::boundary(rect)); //24

    let circ = Circle { radius: 5 };
    println!("Circle area: {}", ShapeGeometry::area(circ)); //78
    println!("Circle boundary: {}", ShapeGeometry::boundary(circ)); //31
}

Listing 8-6: Implementing an external trait

Note that in Listing 8-6, CircleGeometry and RectangleGeometry implementations don't need to be declared as pub. Indeed, ShapeGeometry trait, which is public, is used to print the result in the main function. The compiler will find the appropriate implementation for the ShapeGeometry public trait, regardless of the implementation visibility.

Impl Aliases

Implementations can be aliased when imported. This is most useful when you want to instantiate generic implementations with concrete types. For example, let's say we define a trait Two that is used to return the value 2 for a type T. We can write a trivial generic implementation of Two for all types that implement the One trait, simply by adding twice the value of one and returning it. However, in our public API, we may only want to expose the Two implementation for the u8 and u128 types.

trait Two<T> {
    fn two() -> T;
}

mod one_based {
    pub impl TwoImpl<
        T, +Copy<T>, +Drop<T>, +Add<T>, impl One: core::num::traits::One<T>
    > of super::Two<T> {
        fn two() -> T {
            One::one() + One::one()
        }
    }
}

pub impl U8Two = one_based::TwoImpl<u8>;
pub impl U128Two = one_based::TwoImpl<u128>;

Listing 8-7: Using impl aliases to instantiate generic impls with concrete types

We can define the generic implementation in a private module, use an impl alias to instantiate the generic implementation for these two concrete types, and make these two implementations public, while keeping the generic implementation private and unexposed. This way, we can avoid code duplication using the generic implementation, while keeping the public API clean and simple.

Negative Impls

Note: This is still an experimental feature and can only be used if experimental-features = ["negative_impls"] is enabled in your Scarb.toml file, under the [package] section.

Negative implementations, also known as negative traits or negative bounds, are a mechanism that allows you to express that a type does not implement a certain trait when defining the implementation of a trait over a generic type. Negative impls enable you to write implementations that are applicable only when another implementation does not exist in the current scope.

For example, let's say we have a trait Producer and a trait Consumer, and we want to define a generic behavior where all types implement the Consumer trait by default. However, we want to ensure that no type can be both a Consumer and a Producer. We can use negative impls to express this restriction.

In Listing 8-8, we define a ProducerType that implements the Producer trait, and two other types, AnotherType and AThirdType, which do not implement the Producer trait. We then use negative impls to create a default implementation of the Consumer trait for all types that do not implement the Producer trait.

#[derive(Drop)]
struct ProducerType {}

#[derive(Drop, Debug)]
struct AnotherType {}

#[derive(Drop, Debug)]
struct AThirdType {}

trait Producer<T> {
    fn produce(self: T) -> u32;
}

trait Consumer<T> {
    fn consume(self: T, input: u32);
}

impl ProducerImpl of Producer<ProducerType> {
    fn produce(self: ProducerType) -> u32 {
        42
    }
}

impl TConsumerImpl<T, +core::fmt::Debug<T>, +Drop<T>, -Producer<T>> of Consumer<T> {
    fn consume(self: T, input: u32) {
        println!("{:?} consumed value: {}", self, input);
    }
}

fn main() {
    let producer = ProducerType {};
    let another_type = AnotherType {};
    let third_type = AThirdType {};
    let production = producer.produce();

    // producer.consume(production); Invalid: ProducerType does not implement Consumer
    another_type.consume(production);
    third_type.consume(production);
}

Listing 8-8: Using negative impls to enforce that a type cannot implement both Producer and Consumer traits simultaneously

In the main function, we create instances of ProducerType, AnotherType, and AThirdType. We then call the produce method on the producer instance and pass the result to the consume method on the another_type and third_type instances. Finally, we try to call the consume method on the producer instance, which results in a compile-time error because ProducerType does not implement the Consumer trait.

Error handling

In this chapter, we will explore various error handling techniques provided by Cairo, which not only allow you to address potential issues in your code, but also make it easier to create programs that are adaptable and maintainable. By examining different approaches to managing errors, such as pattern matching with the Result enum, using the ? operator for more ergonomic error propagation, and employing the unwrap or expect methods for handling recoverable errors, you'll gain a deeper understanding of Cairo's error handling features. These concepts are crucial for building robust applications that can effectively handle unexpected situations, ensuring your code is ready for production.

Unrecoverable Errors with panic

In Cairo, unexpected issues may arise during program execution, resulting in runtime errors. While the panic function from the core library doesn't provide a resolution for these errors, it does acknowledge their occurrence and terminates the program. There are two primary ways that a panic can be triggered in Cairo: inadvertently, through actions causing the code to panic (e.g., accessing an array beyond its bounds), or deliberately, by invoking the panic function.

When a panic occurs, it leads to an abrupt termination of the program. The panic function takes an array as an argument, which can be used to provide an error message and performs an unwind process where all variables are dropped and dictionaries squashed to ensure the soundness of the program to safely terminate the execution.

Here is how we can call panic from inside a program and return the error code 2:

Filename: src/lib.cairo

fn main() {
    let mut data = array![2];

    if true {
        panic(data);
    }
    println!("This line isn't reached");
}

Running the program will produce the following output:

$ scarb cairo-run 
   Compiling no_listing_01_panic v0.1.0 (listings/ch09-error-handling/no_listing_01_panic/Scarb.toml)
    Finished `dev` profile target(s) in 3 seconds
     Running no_listing_01_panic
Run panicked with [2, ].

As you can notice in the output, the call to println! macro is never reached, as the program terminates after encountering the panic statement.

An alternative and more idiomatic approach to panic in Cairo would be to use the panic_with_felt252 function. This function serves as an abstraction of the array-defining process and is often preferred due to its clearer and more concise expression of intent. By using panic_with_felt252, developers can panic in a one-liner by providing a felt252 error message as an argument, making the code more readable and maintainable.

Let's consider an example:

use core::panic_with_felt252;

fn main() {
    panic_with_felt252(2);
}

Executing this program will yield the same error message as before. In that case, if there is no need for an array and multiple values to be returned within the error, panic_with_felt252 is a more succinct alternative.

panic! Macro

panic! macro can be really helpful. The previous example returning the error code 2 shows how convenient panic! macro is. There is no need to create an array and pass it as an argument like with the panic function.

fn main() {
    if true {
        panic!("2");
    }
    println!("This line isn't reached");
}

Unlike the panic_with_felt252 function, using panic! allows the input, which is ultimately the panic error, to be a literal longer than 31 bytes. This is because panic! takes a string as a parameter. For example, the following line of code will successfully compile:

panic!("the error for panic! macro is not limited to 31 characters anymore");

nopanic Notation

You can use the nopanic notation to indicate that a function will never panic. Only nopanic functions can be called in a function annotated as nopanic.

Here is an example:

fn function_never_panic() -> felt252 nopanic {
    42
}

This function will always return 42 and is guaranteed to never panic. Conversely, the following function is not guaranteed to never panic:

fn function_never_panic() nopanic {
    assert(1 == 1, 'what');
}

If you try to compile this function that includes code that may panic, you will get the following error:

$ scarb cairo-run 
   Compiling no_listing_04_nopanic_wrong v0.1.0 (listings/ch09-error-handling/no_listing_05_nopanic_wrong/Scarb.toml)
error: Function is declared as nopanic but calls a function that may panic.
 --> listings/ch09-error-handling/no_listing_05_nopanic_wrong/src/lib.cairo:4:12
    assert(1 == 1, 'what');
           ^****^

error: Function is declared as nopanic but calls a function that may panic.
 --> listings/ch09-error-handling/no_listing_05_nopanic_wrong/src/lib.cairo:4:5
    assert(1 == 1, 'what');
    ^********************^

error: could not compile `no_listing_04_nopanic_wrong` due to previous error
error: `scarb metadata` exited with error

Note that there are two functions that may panic here, assert and equality with ==. We usually don't use assert function in practice and use assert! macro instead. We will discuss assert! macro in more detail in the Testing Cairo Programs chapter.

panic_with Attribute

You can use the panic_with attribute to mark a function that returns an Option or Result. This attribute takes two arguments, which are the data that is passed as the panic reason as well as the name for a wrapping function. It will create a wrapper for your annotated function which will panic if the function returns None or Err, with the given data as the panic error.

Example:

#[panic_with('value is 0', wrap_not_zero)]
fn wrap_if_not_zero(value: u128) -> Option<u128> {
    if value == 0 {
        Option::None
    } else {
        Option::Some(value)
    }
}

fn main() {
    wrap_if_not_zero(0); // this returns None
    wrap_not_zero(0); // this panics with 'value is 0'
}

Recoverable Errors with Result

Most errors aren’t serious enough to require the program to stop entirely. Sometimes, when a function fails, it’s for a reason that you can easily interpret and respond to. For example, if you try to add two large integers and the operation overflows because the sum exceeds the maximum representable value, you might want to return an error or a wrapped result instead of causing undefined behavior or terminating the process.

The Result Enum

Recall from Generic data types section in Chapter 8 that the Result enum is defined as having two variants, Ok and Err, as follows:

enum Result<T, E> {
    Ok: T,
    Err: E,
}

The Result<T, E> enum has two generic types, T and E, and two variants: Ok which holds the value of type T and Err which holds the value of type E. This definition makes it convenient to use the Result enum anywhere we have an operation that might succeed (by returning a value of type T) or fail (by returning a value of type E).

The ResultTrait

The ResultTrait trait provides methods for working with the Result<T, E> enum, such as unwrapping values, checking whether the Result is Ok or Err, and panicking with a custom message. The ResultTraitImpl implementation defines the logic of these methods.

trait ResultTrait<T, E> {
    fn expect<+Drop<E>>(self: Result<T, E>, err: felt252) -> T;

    fn unwrap<+Drop<E>>(self: Result<T, E>) -> T;

    fn expect_err<+Drop<T>>(self: Result<T, E>, err: felt252) -> E;

    fn unwrap_err<+Drop<T>>(self: Result<T, E>) -> E;

    fn is_ok(self: @Result<T, E>) -> bool;

    fn is_err(self: @Result<T, E>) -> bool;
}

The expect and unwrap methods are similar in that they both attempt to extract the value of type T from a Result<T, E> when it is in the Ok variant. If the Result is Ok(x), both methods return the value x. However, the key difference between the two methods lies in their behavior when the Result is in the Err variant. The expect method allows you to provide a custom error message (as a felt252 value) that will be used when panicking, giving you more control and context over the panic. On the other hand, the unwrap method panics with a default error message, providing less information about the cause of the panic.

The expect_err and unwrap_err methods have the exact opposite behavior. If the Result is Err(x), both methods return the value x. However, the key difference between the two methods is in case of Result::Ok(). The expect_err method allows you to provide a custom error message (as a felt252 value) that will be used when panicking, giving you more control and context over the panic. On the other hand, the unwrap_err method panics with a default error message, providing less information about the cause of the panic.

A careful reader may have noticed the <+Drop<T>> and <+Drop<E>> in the first four methods signatures. This syntax represents generic type constraints in the Cairo language, as seen in the previous chapter. These constraints indicate that the associated functions require an implementation of the Drop trait for the generic types T and E, respectively.

Finally, the is_ok and is_err methods are utility functions provided by the ResultTrait trait to check the variant of a Result enum value.

• is_ok takes a snapshot of a Result<T, E> value and returns true if the Result is the Ok variant, meaning the operation was successful. If the Result is the Err variant, it returns false.
• is_err takes a snapshot of a Result<T, E> value and returns true if the Result is the Err variant, meaning the operation encountered an error. If the Result is the Ok variant, it returns false.

These methods are helpful when you want to check the success or failure of an operation without consuming the Result value, allowing you to perform additional operations or make decisions based on the variant without unwrapping it.

You can find the implementation of the ResultTrait here.

It is always easier to understand with examples. Have a look at this function signature:

fn u128_overflowing_add(a: u128, b: u128) -> Result<u128, u128>;

It takes two u128 integers, a and b, and returns a Result<u128, u128> where the Ok variant holds the sum if the addition does not overflow, and the Err variant holds the overflowed value if the addition does overflow.

Now, we can use this function elsewhere. For instance:

fn u128_checked_add(a: u128, b: u128) -> Option<u128> {
    match u128_overflowing_add(a, b) {
        Result::Ok(r) => Option::Some(r),
        Result::Err(r) => Option::None,
    }
}

Here, it accepts two u128 integers, a and b, and returns an Option<u128>. It uses the Result returned by u128_overflowing_add to determine the success or failure of the addition operation. The match expression checks the Result from u128_overflowing_add. If the result is Ok(r), it returns Option::Some(r) containing the sum. If the result is Err(r), it returns Option::None to indicate that the operation has failed due to overflow. The function does not panic in case of an overflow.

Let's take another example:

fn parse_u8(s: felt252) -> Result<u8, felt252> {
    match s.try_into() {
        Option::Some(value) => Result::Ok(value),
        Option::None => Result::Err('Invalid integer'),
    }
}

In this example, the parse_u8 function takes a felt252 and tries to convert it into a u8 integer using the try_into method. If successful, it returns Result::Ok(value), otherwise it returns Result::Err('Invalid integer').

Our two test cases are:

fn parse_u8(s: felt252) -> Result<u8, felt252> {
    match s.try_into() {
        Option::Some(value) => Result::Ok(value),
        Option::None => Result::Err('Invalid integer'),
    }
}

#[cfg(test)]
mod tests {
    use super::parse_u8;

    #[test]
    fn test_felt252_to_u8() {
        let number: felt252 = 5;
        // should not panic
        let res = parse_u8(number).unwrap();
    }

    #[test]
    #[should_panic]
    fn test_felt252_to_u8_panic() {
        let number: felt252 = 256;
        // should panic
        let res = parse_u8(number).unwrap();
    }
}

Don't worry about the #[cfg(test)] attribute for now. We'll explain in more detail its meaning in the next Testing Cairo Programs chapter.

#[test] attribute means the function is a test function, and #[should_panic] attribute means this test will pass if the test execution panics.

The first one tests a valid conversion from felt252 to u8, expecting the unwrap method not to panic. The second test function attempts to convert a value that is out of the u8 range, expecting the unwrap method to panic with the error message Invalid integer.

The ? Operator

The last operator we will talk about is the ? operator. The ? operator is used for more idiomatic and concise error handling. When you use the ? operator on a Result or Option type, it will do the following:

• If the value is Result::Ok(x) or Option::Some(x), it will return the inner value x directly.
• If the value is Result::Err(e) or Option::None, it will propagate the error or None by immediately returning from the function.

The ? operator is useful when you want to handle errors implicitly and let the calling function deal with them.

Here is an example:

fn do_something_with_parse_u8(input: felt252) -> Result<u8, felt252> {
    let input_to_u8: u8 = parse_u8(input)?;
    // DO SOMETHING
    let res = input_to_u8 - 1;
    Result::Ok(res)
}

We can see that do_something_with_parse_u8 function takes a felt252 value as input and calls parse_u8 function. The ? operator is used to propagate the error, if any, or unwrap the successful value.

And with a little test case:

fn parse_u8(s: felt252) -> Result<u8, felt252> {
    match s.try_into() {
        Option::Some(value) => Result::Ok(value),
        Option::None => Result::Err('Invalid integer'),
    }
}

fn do_something_with_parse_u8(input: felt252) -> Result<u8, felt252> {
    let input_to_u8: u8 = parse_u8(input)?;
    // DO SOMETHING
    let res = input_to_u8 - 1;
    Result::Ok(res)
}

#[cfg(test)]
mod tests {
    use super::do_something_with_parse_u8;
    #[test]
    fn test_function_2() {
        let number: felt252 = 258;
        match do_something_with_parse_u8(number) {
            Result::Ok(value) => println!("Result: {}", value),
            Result::Err(e) => println!("Error: {}", e),
        }
    }
}

The console will print the error Invalid Integer.

Summary

We saw that recoverable errors can be handled in Cairo using the Result enum, which has two variants: Ok and Err. The Result<T, E> enum is generic, with types T and E representing the successful and error values, respectively. The ResultTrait provides methods for working with Result<T, E>, such as unwrapping values, checking if the result is Ok or Err, and panicking with custom messages.

To handle recoverable errors, a function can return a Result type and use pattern matching to handle the success or failure of an operation. The ? operator can be used to implicitly handle errors by propagating the error or unwrapping the successful value. This allows for more concise and clear error handling, where the caller is responsible for managing errors raised by the called function.

Testing Cairo Programs

Correctness in our programs is the extent to which our code does what we intend it to do. Cairo is designed with a high degree of concern about the correctness of programs, but correctness is complex and not easy to prove. Cairo's linear type system shoulders a huge part of this burden, but the type system cannot catch everything. As such, Cairo includes support for writing tests.

Testing is a complex skill: although we can’t cover every detail about how to write good tests in one chapter, we’ll discuss the mechanics of Cairo's testing facilities. We’ll talk about the annotations and macros available to you when writing your tests, the default behavior and options provided for running your tests, and how to organize tests into unit tests and integration tests.

How To Write Tests

The Anatomy of a Test Function

Tests are Cairo functions that verify that the non-test code is functioning in the expected manner. The bodies of test functions typically perform these three actions:

• Set up any needed data or state.
• Run the code you want to test.
• Assert the results are what you expect.

Let’s look at the features Cairo provides for writing tests that take these actions, which include:

• #[test] attribute.
• assert!macro.
• assert_eq!, assert_ne!, assert_lt!, assert_le!, assert_gt! and assert_ge! macros. In order to use them, you will need to add assert_macros = "2.8.2" as a dev dependency.
• #[should_panic] attribute.

Note: Make sure to select Starknet Foundry as a test runner when creating your project.

The Anatomy of a Test Function

At its simplest, a test in Cairo is a function that’s annotated with the #[test] attribute. Attributes are metadata about pieces of Cairo code; one example is the #[derive()] attribute we used with structs in Chapter 5. To change a function into a test function, add #[test] on the line before fn. When you run your tests with the scarb test command, Scarb runs Starknet Foundry's test runner binary that runs the annotated functions and reports on whether each test function passes or fails.

Let's create a new project called adder using Scarb with the command scarb new adder. Remove the tests folder.

adder
├── Scarb.toml
└── src
    └── lib.cairo

In lib.cairo, let's remove the existing content and add a tests module containing the first test, as shown in Listing 10-1.

Filename: src/lib.cairo

pub fn add(left: usize, right: usize) -> usize {
    left + right
}

#[cfg(test)]
mod tests {
    use super::add;

    #[test]
    fn it_works() {
        let result = add(2, 2);
        assert_eq!(result, 4);
    }
}

Listing 10-1: A simple test function

Note the #[test] annotation: this attribute indicates this is a test function, so the test runner knows to treat this function as a test. We might also have non-test functions to help set up common scenarios or perform common operations, so we always need to indicate which functions are tests.

We use the #[cfg(test)] attribute for the tests module, so that the compiler knows the code it contains needs to be compiled only when running tests. This is actually not an option: if you put a simple test with the #[test] attribute in a lib.cairo file, it will not compile. We will talk more about the #[cfg(test)] attribute in the next Test Organization section.

The example function body uses the assert_eq! macro, which contains the result of adding 2 and 2, which equals 4. This assertion serves as an example of the format for a typical test. We'll explain in more detail how assert_eq! works later in this chapter. Let’s run it to see that this test passes.

The scarb test command runs all tests found in our project, and shows the following output:

$ scarb test 
     Running test listing_10_01 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.90s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_01/Scarb.toml)
    Finished `dev` profile target(s) in 6 seconds


Collected 2 test(s) from listing_10_01 package
Running 2 test(s) from src/
[PASS] listing_10_01::tests::it_works (gas: ~1)
[PASS] listing_10_01::other_tests::exploration (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

scarb test compiled and ran the test. We see the line Collected 1 test(s) from adder package followed by the line Running 1 test(s) from src/. The next line shows the name of the test function, called it_works, and that the result of running that test is ok. The test runner also provides an estimation of the gas consumption. The overall summary shows that all the tests passed, and the portion that reads 1 passed; 0 failed totals the number of tests that passed or failed.

It’s possible to mark a test as ignored so it doesn’t run in a particular instance; we’ll cover that in the Ignoring Some Tests Unless Specifically Requested section later in this chapter. Because we haven’t done that here, the summary shows 0 ignored. We can also pass an argument to the scarb test command to run only a test whose name matches a string; this is called filtering and we’ll cover that in the Running Single Tests section. Since we haven’t filtered the tests being run, the end of the summary shows 0 filtered out.

Let’s start to customize the test to our own needs. First change the name of the it_works function to a different name, such as exploration, like so:

    #[test]
    fn exploration() {
        let result = 2 + 2;
        assert_eq!(result, 4);
    }

Then run scarb test again. The output now shows exploration instead of it_works:

$ scarb test 
     Running test listing_10_01 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.90s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_01/Scarb.toml)
    Finished `dev` profile target(s) in 6 seconds


Collected 2 test(s) from listing_10_01 package
Running 2 test(s) from src/
[PASS] listing_10_01::tests::it_works (gas: ~1)
[PASS] listing_10_01::other_tests::exploration (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Now we’ll add another test, but this time we’ll make a test that fails! Tests fail when something in the test function panics. Each test is run in a new thread, and when the main thread sees that a test thread has died, the test is marked as failed. Enter the new test as a function named another, so your src/lib.cairo file looks like in Listing 10-2.

Filename: src/lib.cairo

#[cfg(test)]
mod tests {
    #[test]
    fn exploration() {
        let result = 2 + 2;
        assert_eq!(result, 4);
    }

    #[test]
    fn another() {
        let result = 2 + 2;
        assert!(result == 6, "Make this test fail");
    }
}

Listing 10-2: Adding a second test in lib.cairo that will fail

Run scarb test and you will see the following output:

Collected 2 test(s) from adder package
Running 2 test(s) from src/
[FAIL] adder::tests::another

Failure data:
    "Make this test fail"

[PASS] adder::tests::exploration (gas: ~1)
Tests: 1 passed, 1 failed, 0 skipped, 0 ignored, 0 filtered out

Failures:
    adder::tests::another

Instead of [PASS], the line adder::tests::another shows [FAIL]. A new section appears between the individual results and the summary. It displays the detailed reason for each test failure. In this case, we get the details that another failed because it panicked with "Make this test fail" error.

After that, the summary line is displayed: we had one test pass and one test fail. At the end, we see a list of the failing tests.

Now that you've seen what the test results look like in different scenarios, let’s look at some functions that are useful in tests.

Checking Results with the assert! Macro

The assert! macro, provided by Cairo, is useful when you want to ensure that some condition in a test evaluates to true. We give the assert! macro the first argument that evaluates to a boolean. If the value is true, nothing happens and the test passes. If the value is false, the assert! macro calls panic() to cause the test to fail with a message we defined as the second argument. Using the assert! macro helps us check that our code is functioning in the way we intended.

Remember in Chapter 5, we used a Rectangle struct and a can_hold method, which are repeated here in Listing 10-3. Let’s put this code in the src/lib.cairo file, then write some tests for it using the assert! macro.

Filename: src/lib.cairo

#[derive(Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

trait RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool;
}

impl RectangleImpl of RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width > *other.width && *self.height > *other.height
    }
}

Listing 10-3: Using the Rectangle struct and its can_hold method from Chapter 5

The can_hold method returns a bool, which means it’s a perfect use case for the assert! macro. We can write a test that exercises the can_hold method by creating a Rectangle instance that has a width of 8 and a height of 7 and asserting that it can hold another Rectangle instance that has a width of 5 and a height of 1.

#[derive(Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

trait RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool;
}

impl RectangleImpl of RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width > *other.width && *self.height > *other.height
    }
}

#[cfg(test)]
mod tests {
    use super::Rectangle;
    use super::RectangleTrait;

    #[test]
    fn larger_can_hold_smaller() {
        let larger = Rectangle { height: 7, width: 8, };
        let smaller = Rectangle { height: 1, width: 5, };

        assert!(larger.can_hold(@smaller), "rectangle cannot hold");
    }
}
#[cfg(test)]
mod tests2 {
    use super::Rectangle;
    use super::RectangleTrait;

    #[test]
    fn smaller_cannot_hold_larger() {
        let larger = Rectangle { height: 7, width: 8, };
        let smaller = Rectangle { height: 1, width: 5, };

        assert!(!smaller.can_hold(@larger), "rectangle cannot hold");
    }
}

We’ve named our test larger_can_hold_smaller, and we’ve created the two Rectangle instances that we need. Then we called the assert! macro and passed it the result of calling larger.can_hold(@smaller). This expression is supposed to return true, so our test should pass. Let’s find out!

$ scarb test 
     Running test listing_10_03 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Finished `release` profile [optimized] target(s) in 1.04s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_03/Scarb.toml)
    Finished `dev` profile target(s) in 5 seconds


Collected 2 test(s) from listing_10_03 package
Running 2 test(s) from src/
[PASS] listing_10_03::tests::larger_can_hold_smaller (gas: ~1)
[PASS] listing_10_03::tests2::smaller_cannot_hold_larger (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

It does pass! Let’s add another test, this time asserting that a smaller rectangle cannot hold a larger rectangle:

Filename: src/lib.cairo

#[derive(Drop)]
struct Rectangle {
    width: u64,
    height: u64,
}

trait RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool;
}

impl RectangleImpl of RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width > *other.width && *self.height > *other.height
    }
}

#[cfg(test)]
mod tests {
    use super::Rectangle;
    use super::RectangleTrait;

    #[test]
    fn larger_can_hold_smaller() {
        let larger = Rectangle { height: 7, width: 8, };
        let smaller = Rectangle { height: 1, width: 5, };

        assert!(larger.can_hold(@smaller), "rectangle cannot hold");
    }
}
#[cfg(test)]
mod tests2 {
    use super::Rectangle;
    use super::RectangleTrait;

    #[test]
    fn smaller_cannot_hold_larger() {
        let larger = Rectangle { height: 7, width: 8, };
        let smaller = Rectangle { height: 1, width: 5, };

        assert!(!smaller.can_hold(@larger), "rectangle cannot hold");
    }
}

Listing 10-4: Adding another test in lib.cairo that will pass

Because the correct result of the can_hold method, in this case, is false, we need to negate that result before we pass it to the assert! macro. As a result, our test will pass if can_hold returns false:

$ scarb test 
     Running test listing_10_03 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Finished `release` profile [optimized] target(s) in 1.04s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_03/Scarb.toml)
    Finished `dev` profile target(s) in 5 seconds


Collected 2 test(s) from listing_10_03 package
Running 2 test(s) from src/
[PASS] listing_10_03::tests::larger_can_hold_smaller (gas: ~1)
[PASS] listing_10_03::tests2::smaller_cannot_hold_larger (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Two tests that pass! Now let’s see what happens to our test results when we introduce a bug in our code. We’ll change the implementation of the can_hold method by replacing the > sign with a < sign when it compares the widths:

impl RectangleImpl of RectangleTrait {
    fn can_hold(self: @Rectangle, other: @Rectangle) -> bool {
        *self.width < *other.width && *self.height > *other.height
    }
}

Running the tests now produces the following:

$ scarb test 
     Running test no_listing_01_wrong_can_hold_impl (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 1.11s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_01_wrong_can_hold_impl/Scarb.toml)
    Finished `dev` profile target(s) in 6 seconds


Collected 0 test(s) from no_listing_01_wrong_can_hold_impl package
Running 0 test(s) from src/
Tests: 0 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Our tests caught the bug! Because larger.width is 8 and smaller.width is 5, the comparison of the widths in can_hold now returns false (8 is not less than 5) in the larger_can_hold_smaller test. Notice that the smaller_cannot_hold_larger test still passes: to make this test fail, the height comparison should also be modified in can_hold method, replacing the > sign with a < sign.

Testing Equality and Comparisons with the assert_xx! Macros

assert_eq! and assert_ne! Macros

A common way to verify functionality is to test for equality between the result of the code under test and the value you expect the code to return. You could do this using the assert! macro and passing it an expression using the == operator. However, this is such a common test that the standard library provides a pair of macros — assert_eq! and assert_ne! — to perform this test more conveniently. These macros compare two arguments for equality or inequality, respectively. They’ll also print the two values if the assertion fails, which makes it easier to see why the test failed; conversely, the assert! macro only indicates that it got a false value for the == expression, without printing the values that led to the false value.

In Listing 10-5, we write a function named add_two that adds 2 to its parameter, then we test this function using assert_eq! and assert_ne! macros.

Filename: src/lib.cairo

pub fn add_two(a: u32) -> u32 {
    a + 2
}

#[cfg(test)]
mod tests {
    use super::add_two;

    #[test]
    fn it_adds_two() {
        assert_eq!(4, add_two(2));
    }

    #[test]
    fn wrong_check() {
        assert_ne!(0, add_two(2));
    }
}

Listing 10-5: Testing the function add_two using assert_eq! and assert_ne! macros

Let’s check that it passes!

$ scarb test 
     Running test listing_10_04 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 1.00s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_04/Scarb.toml)
    Finished `dev` profile target(s) in 5 seconds


Collected 2 test(s) from listing_10_04 package
Running 2 test(s) from src/
[PASS] listing_10_04::add_two::tests::it_adds_two (gas: ~1)
[PASS] listing_10_04::add_two::tests::wrong_check (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

In the it_adds_two test, we pass 4 as argument to assert_eq! macro, which is equal to the result of calling add_two(2). The line for this test is [PASS] adder::tests::it_adds_two (gas: ~1).

In the wrong_check test, we pass 0 as argument to assert_ne! macro, which is not equal to the result of calling add_two(2). Tests that use the assert_ne! macro will pass if the two values we give it are not equal and fail if they’re equal. This macro is most useful for cases when we’re not sure what a value will be, but we know what the value definitely shouldn’t be. For example, if we’re testing a function that is guaranteed to change its input in some way, but how the input is changed depends on the day of the week that we run our tests, the best thing to assert might be that the output of the function is not equal to the input.

Let’s introduce a bug into our code to see what assert_eq! looks like when it fails. Change the implementation of the add_two function to instead add 3:

pub fn add_two(a: u32) -> u32 {
    a + 3
}

Run the tests again:

$ scarb test 
     Running test listing_10_04 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 1.00s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_04/Scarb.toml)
    Finished `dev` profile target(s) in 5 seconds


Collected 2 test(s) from listing_10_04 package
Running 2 test(s) from src/
[PASS] listing_10_04::add_two::tests::it_adds_two (gas: ~1)
[PASS] listing_10_04::add_two::tests::wrong_check (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Our test caught the bug! The it_adds_two test failed with the following message: "assertion `4 == add_two(2)` failed. It tells us that the assertion that failed was "assertion `left == right` failed and the left and right values are printed on the next lines as left: left_value and right: right_value. This helps us start debugging: the left argument was 4 but the right argument, where we had add_two(2), was 5. You can imagine that this would be especially helpful when we have a lot of tests going on.

Note that in some languages and test frameworks, the parameters for equality assertion functions are called expected and actual, and the order in which we specify the arguments matters. However, in Cairo, they’re called left and right, and the order in which we specify the value we expect and the value the code produces doesn’t matter. We could write the assertion in this test as assert_eq!(add_two(2), 4), which would result in the same failure message that displays assertion failed: `(left == right)`.

Here is a simple example comparing two structs, showing how to use assert_eq! and assert_ne! macros:

#[derive(Drop, Debug, PartialEq)]
struct MyStruct {
    var1: u8,
    var2: u8
}

#[cfg(test)]
#[test]
fn test_struct_equality() {
    let first = MyStruct { var1: 1, var2: 2 };
    let second = MyStruct { var1: 1, var2: 2 };
    let third = MyStruct { var1: 1, var2: 3 };

    assert_eq!(first, second);
    assert_eq!(first, second, "{:?},{:?} should be equal", first, second);
    assert_ne!(first, third);
    assert_ne!(first, third, "{:?},{:?} should not be equal", first, third);
}

Under the surface, assert_eq! and assert_ne! macros use the operators == and !=, respectively. They both take snapshots of values as arguments. When the assertions fail, these macros print their arguments using debug formatting ({:?} syntax), which means the values being compared must implement PartialEq and Debug traits. All primitive types and most of the core library types implement these traits. For structs and enums that you define yourself, you’ll need to implement PartialEq to assert equality of those types. You’ll also need to implement Debug to print the values when the assertion fails. Because both traits are derivable, this is usually as straightforward as adding the #[derive(Drop, Debug, PartialEq)] annotation to your struct or enum definition. See Appendix C for more details about these and other derivable traits.

assert_lt!, assert_le!, assert_gt! and assert_ge! Macros

Comparisons in tests can be done using the assert_xx! macros:

• assert_lt! checks if a given value is lower than another value, and reverts otherwise.
• assert_le! checks if a given value is lower or equal than another value, and reverts otherwise.
• assert_gt! checks if a given value is greater than another value, and reverts otherwise.
• assert_ge! checks if a given value is greater or equal than another value, and reverts otherwise.

Listing 10-6 demonstrates how to use these macros:

#[derive(Drop, Copy, Debug, PartialEq)]
struct Dice {
    number: u8,
}

impl DicePartialOrd of PartialOrd<Dice> {
    fn lt(lhs: Dice, rhs: Dice) -> bool {
        lhs.number < rhs.number
    }

    fn le(lhs: Dice, rhs: Dice) -> bool {
        lhs.number <= rhs.number
    }

    fn gt(lhs: Dice, rhs: Dice) -> bool {
        lhs.number > rhs.number
    }

    fn ge(lhs: Dice, rhs: Dice) -> bool {
        lhs.number >= rhs.number
    }
}

#[cfg(test)]
#[test]
fn test_struct_equality() {
    let first_throw = Dice { number: 5 };
    let second_throw = Dice { number: 2 };
    let third_throw = Dice { number: 6 };
    let fourth_throw = Dice { number: 5 };

    assert_gt!(first_throw, second_throw);
    assert_ge!(first_throw, fourth_throw);
    assert_lt!(second_throw, third_throw);
    assert_le!(
        first_throw, fourth_throw, "{:?},{:?} should be lower or equal", first_throw, fourth_throw
    );
}

Listing 10-6: Example of tests that use the assert_xx! macros for comparisons

In this example, we roll a Dice struct multiple times and compare the results. We need to manually implement the PartialOrd trait for our struct so that we can compare Dice instances with lt, le, gt and ge functions, which are used by assert_lt!, assert_le!, assert_gt! and assert_ge! macros, respectively. We also need to derive the Copy trait on our Dice struct to use the instantiated structs multiple times, as the comparison functions take ownership of the variables.

Adding Custom Failure Messages

You can also add a custom message to be printed with the failure message as optional arguments to assert!, assert_eq!, and assert_ne! macros. Any arguments specified after the required arguments are passed along to the format! macro (discussed in the Printing chapter), so you can pass a format string that contains {} placeholders and values to go in those placeholders. Custom messages are useful for documenting what an assertion means; when a test fails, you’ll have a better idea of what the problem is with the code.

Let’s add a custom failure message composed of a format string with a placeholder filled in with the actual value we got from the previous add_two function:

    #[test]
    fn it_adds_two() {
        assert_eq!(4, add_two(2), "Expected {}, got add_two(2)={}", 4, add_two(2));
    }

Now when we run the test, we’ll get a more informative error message:

$ scarb test 
     Running test no_listing_02_custom_messages (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.39s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_02_custom_messages/Scarb.toml)
    Finished `dev` profile target(s) in 4 seconds


Collected 1 test(s) from no_listing_02_custom_messages package
Running 1 test(s) from src/
[FAIL] no_listing_02_custom_messages::tests::it_adds_two

Failure data:
    "assertion `4 == add_two(2)` failed: Expected 4, got add_two(2)=5
    4: 4
    add_two(2): 5"

Tests: 0 passed, 1 failed, 0 skipped, 0 ignored, 0 filtered out

Failures:
    no_listing_02_custom_messages::tests::it_adds_two

We can see the value we actually got in the test output, which would help us debug what happened instead of what we were expecting to happen.

Checking for panics with should_panic

In addition to checking return values, it’s important to check that our code handles error conditions as we expect. For example, consider the Guess type in Listing 10-7:

Filename: src/lib.cairo

#[derive(Drop)]
struct Guess {
    value: u64,
}

pub trait GuessTrait {
    fn new(value: u64) -> Guess;
}

impl GuessImpl of GuessTrait {
    fn new(value: u64) -> Guess {
        if value < 1 || value > 100 {
            panic!("Guess must be >= 1 and <= 100");
        }

        Guess { value, }
    }
}

Listing 10-7: Guess struct and its new method

Other code that uses Guess depends on the guarantee that Guess instances will contain only values between 1 and 100. We can write a test that ensures that attempting to create a Guess instance with a value outside that range panics.

We do this by adding the attribute should_panic to our test function. The test passes if the code inside the function panics; the test fails if the code inside the function doesn’t panic.

#[cfg(test)]
mod tests {
    use super::GuessTrait;

    #[test]
    #[should_panic]
    fn greater_than_100() {
        GuessTrait::new(200);
    }
}

We place the #[should_panic] attribute after the #[test] attribute and before the test function it applies to. Let’s look at the result to see that this test passes:

$ scarb test 
     Running test listing_09_08 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.21s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_05/Scarb.toml)
    Finished `dev` profile target(s) in 4 seconds


Collected 1 test(s) from listing_09_08 package
Running 1 test(s) from src/
[PASS] listing_09_08::tests::greater_than_100 (gas: ~1)

Success data:
    "Guess must be >= 1 and <= 100"

Tests: 1 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Looks good! Now let’s introduce a bug in our code by removing the condition that the new function will panic if the value is greater than 100:

#[derive(Drop)]
struct Guess {
    value: u64,
}

trait GuessTrait {
    fn new(value: u64) -> Guess;
}

impl GuessImpl of GuessTrait {
    fn new(value: u64) -> Guess {
        if value < 1 {
            panic!("Guess must be >= 1 and <= 100");
        }

        Guess { value, }
    }
}

When we run the test, it will fail:

$ scarb test 
     Running test no_listing_03_wrong_new_impl (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Finished `release` profile [optimized] target(s) in 0.14s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_03_wrong_new_impl/Scarb.toml)
    Finished `dev` profile target(s) in 4 seconds


Collected 0 test(s) from no_listing_03_wrong_new_impl package
Running 0 test(s) from src/
Tests: 0 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

We don’t get a very helpful message in this case, but when we look at the test function, we see that it’s annotated with #[should_panic] attribute. The failure we got means that the code in the test function did not cause a panic.

Tests that use should_panic can be imprecise. A should_panic test would pass even if the test panics for a different reason from the one we were expecting. To make should_panic tests more precise, we can add an optional expected parameter to the #[should_panic] attribute. The test harness will make sure that the failure message contains the provided text. For example, consider the modified code for GuessImpl in Listing 10-8 where the new function panics with different messages depending on whether the value is too small or too large:

Filename: src/lib.cairo

#[derive(Drop)]
struct Guess {
    value: u64,
}

trait GuessTrait {
    fn new(value: u64) -> Guess;
}

impl GuessImpl of GuessTrait {
    fn new(value: u64) -> Guess {
        if value < 1 {
            panic!("Guess must be >= 1");
        } else if value > 100 {
            panic!("Guess must be <= 100");
        }

        Guess { value, }
    }
}

#[cfg(test)]
mod tests {
    use super::GuessTrait;

    #[test]
    #[should_panic(expected: "Guess must be <= 100")]
    fn greater_than_100() {
        GuessTrait::new(200);
    }
}

Listing 10-8: new implementation that panics with different error messages

The test will pass because the value we put in the should_panic attribute’s expected parameter is the string that the Guess::new method panics with. We need to specify the entire panic message that we expect.

To see what happens when a should_panic test with an expected message fails, let’s again introduce a bug into our code by swapping the bodies of the if value < 1 and the else if value > 100 blocks:

impl GuessImpl of GuessTrait {
    fn new(value: u64) -> Guess {
        if value < 1 {
            panic!("Guess must be <= 100");
        } else if value > 100 {
            panic!("Guess must be >= 1");
        }

        Guess { value, }
    }
}

#[cfg(test)]
mod tests {
    use super::GuessTrait;

    #[test]
    #[should_panic(expected: "Guess must be <= 100")]
    fn greater_than_100() {
        GuessTrait::new(200);
    }
}

This time when we run the should_panic test, it will fail:

$ scarb test 
     Running test no_listing_04_new_bug (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 1.01s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_04_new_bug/Scarb.toml)
    Finished `dev` profile target(s) in 6 seconds


Collected 1 test(s) from no_listing_04_new_bug package
Running 1 test(s) from src/
[FAIL] no_listing_04_new_bug::tests::greater_than_100

Failure data:
    Incorrect panic data
    Actual:    [0x46a6158a16a947e5916b2a2ca68501a45e93d7110e81aa2d6438b1c57c879a3, 0x0, 0x4775657373206d757374206265203e3d2031, 0x12] (Guess must be >= 1)
    Expected:  [0x46a6158a16a947e5916b2a2ca68501a45e93d7110e81aa2d6438b1c57c879a3, 0x0, 0x4775657373206d757374206265203c3d20313030, 0x14] (Guess must be <= 100)

Tests: 0 passed, 1 failed, 0 skipped, 0 ignored, 0 filtered out

Failures:
    no_listing_04_new_bug::tests::greater_than_100

The failure message indicates that this test did indeed panic as we expected, but the panic message did not include the expected string. The panic message that we did get in this case was Guess must be >= 1. Now we can start figuring out where our bug is!

Running Single Tests

Sometimes, running a full test suite can take a long time. If you’re working on code in a particular area, you might want to run only the tests pertaining to that code. You can choose which tests to run by passing scarb test the name of the test you want to run as an argument.

To demonstrate how to run a single test, we’ll first create two test functions, as shown in Listing 10-9, and choose which ones to run.

Filename: src/lib.cairo

#[cfg(test)]
mod tests {
    #[test]
    fn add_two_and_two() {
        let result = 2 + 2;
        assert_eq!(result, 4);
    }

    #[test]
    fn add_three_and_two() {
        let result = 3 + 2;
        assert!(result == 5, "result is not 5");
    }
}

Listing 10-9: Two tests with two different names

We can pass the name of any test function to scarb test to run only that test:

$ scarb test 
     Running test listing_10_07 (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.71s
   Compiling test(listings/ch10-testing-cairo-programs/listing_10_07/Scarb.toml)
    Finished `dev` profile target(s) in 6 seconds


Collected 2 test(s) from listing_10_07 package
Running 2 test(s) from src/
[PASS] listing_10_07::tests::add_two_and_two (gas: ~1)
[PASS] listing_10_07::tests::add_three_and_two (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

Only the test with the name add_two_and_two ran; the other test didn’t match that name. The test output lets us know we had one more test that didn’t run by displaying 1 filtered out; at the end.

We can also specify part of a test name, and any test whose name contains that value will be run.

Ignoring Some Tests Unless Specifically Requested

Sometimes a few specific tests can be very time-consuming to execute, so you might want to exclude them during most runs of scarb test. Rather than listing as arguments all tests you do want to run, you can instead annotate the time-consuming tests using the #[ignore] attribute to exclude them, as shown here:

pub fn add(left: usize, right: usize) -> usize {
    left + right
}

#[cfg(test)]
mod tests {
    use super::add;

    #[test]
    fn it_works() {
        let result = add(2, 2);
        assert_eq!(result, 4);
    }

    #[test]
    #[ignore]
    fn expensive_test() { // code that takes an hour to run
    }
}

After #[test] we add the #[ignore] line to the test we want to exclude. Now when we run our tests, it_works runs, but expensive_test doesn’t:

$ scarb test 
     Running test no_listing_05_ignore_tests (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.79s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_05_ignore_tests/Scarb.toml)
    Finished `dev` profile target(s) in 5 seconds


Collected 2 test(s) from no_listing_05_ignore_tests package
Running 2 test(s) from src/
[IGNORE] no_listing_05_ignore_tests::tests::expensive_test
[PASS] no_listing_05_ignore_tests::tests::it_works (gas: ~1)
Tests: 1 passed, 0 failed, 0 skipped, 1 ignored, 0 filtered out

The expensive_test function is listed as ignored.

When you’re at a point where it makes sense to check the results of the ignored tests and you have time to wait for the results, you can run scarb test --include-ignored to run all tests, whether they’re ignored or not.

Testing Recursive Functions or Loops

When testing recursive functions or loops, the test is instantiated by default with a maximum amount of gas that it can consume. This prevents running infinite loops or consuming too much gas, and can help you benchmark the efficiency of your implementations. This value is assumed reasonably large enough, but you can override it by adding the #[available_gas(<Number>)] attribute to the test function. The following example shows how to use it:

fn sum_n(n: usize) -> usize {
    let mut i = 0;
    let mut sum = 0;
    while i <= n {
        sum += i;
        i += 1;
    };
    sum
}

#[cfg(test)]
mod tests {
    use super::sum_n;

    #[test]
    #[available_gas(2000000)]
    fn test_sum_n() {
        let result = sum_n(10);
        assert!(result == 55, "result is not 55");
    }
}

Benchmarking Cairo Programs

Starknet Foundry contains a profiling feature that is useful to analyze and optimize the performance of your Cairo programs.

The profiling feature generates execution traces for successful tests, which are used to create profile outputs. This allows you to benchmark specific parts of your code.

To use the profiler, you will need to:

1. Install Cairo Profiler from Software Mansion.
2. Install Go, Graphviz and pprof, all of them are required to visualize the generated profile output.
3. Run snforge test --build-profile command, which generates a trace file for each passing test, stored in the snfoundry_trace directory of your project. This command also generates the corresponding output files in the profile directory.
4. Run go tool pprof -http=":8000" path/to/profile/output.pb.gz to analyse a profile. This will start a web server at the specified port.

Let's reuse the sum_n function studied above:

fn sum_n(n: usize) -> usize {
    let mut i = 0;
    let mut sum = 0;
    while i <= n {
        sum += i;
        i += 1;
    };
    sum
}

#[cfg(test)]
mod tests {
    use super::sum_n;

    #[test]
    #[available_gas(2000000)]
    fn test_sum_n() {
        let result = sum_n(10);
        assert!(result == 55, "result is not 55");
    }
}

After generating the trace file and the profile output, running go tool pprof in your project will start the web server where you can find many useful information about the test that you ran:

• The test includes one function call, corresponding to the call to the test function. Calling sum_n multiple times in the test function will still return 1 call. This is because snforge simulates a contract call when executing a test.

• The sum_n function execution uses 256 Cairo steps:

Other information is also available such as memory holes (i.e., unused memory cells) or builtins usage. The Cairo Profiler is under active development, and many other features will be made available in the future.

Test Organization

We'll think about tests in terms of two main categories: unit tests and integration tests. Unit tests are small and more focused, testing one module in isolation at a time, and can test private functions. Integration tests use your code in the same way any other external code would, using only the public interface and potentially exercising multiple modules per test.

Writing both kinds of tests is important to ensure that the pieces of your library are doing what you expect them to, separately and together.

Unit Tests

The purpose of unit tests is to test each unit of code in isolation from the rest of the code to quickly pinpoint where code is and isn’t working as expected. You’ll put unit tests in the src directory in each file with the code that they’re testing.

The convention is to create a module named tests in each file to contain the test functions and to annotate the module with #[cfg(test)] attribute.

The Tests Module and #[cfg(test)]

The #[cfg(test)] annotation on the tests module tells Cairo to compile and run the test code only when you run scarb test, not when you run scarb build. This saves compile time when you only want to build the project and saves space in the resulting compiled artifact because the tests are not included. You’ll see that because integration tests go in a different directory, they don’t need the #[cfg(test)] annotation. However, because unit tests go in the same files as the code, you’ll use #[cfg(test)] to specify that they shouldn’t be included in the compiled result.

Recall that when we created the new adder project in the first section of this chapter, we wrote this first test:

pub fn add(left: usize, right: usize) -> usize {
    left + right
}

#[cfg(test)]
mod tests {
    use super::add;

    #[test]
    fn it_works() {
        let result = add(2, 2);
        assert_eq!(result, 4);
    }
}

The attribute cfg stands for configuration and tells Cairo that the following item should only be included given a certain configuration option. In this case, the configuration option is test, which is provided by Cairo for compiling and running tests. By using the cfg attribute, Cairo compiles our test code only if we actively run the tests with scarb test. This includes any helper functions that might be within this module, in addition to the functions annotated with #[test].

Testing Private Functions

There’s debate within the testing community about whether or not private functions should be tested directly, and other languages make it difficult or impossible to test private functions. Regardless of which testing ideology you adhere to, Cairo's privacy rules do allow you to test private functions. Consider the code below with the private function internal_adder.

Filename: src/lib.cairo

pub fn add(a: u32, b: u32) -> u32 {
    internal_adder(a, 2)
}

fn internal_adder(a: u32, b: u32) -> u32 {
    a + b
}

#[cfg(test)]
mod tests {
    use super::internal_adder;

    #[test]
    fn add() {
        assert_eq!(4, internal_adder(2, 2));
    }
}

Listing 10-10: Testing a private function

Note that the internal_adder function is not marked as pub. Tests are just Cairo code, and the tests module is just another module. As we discussed in the "Paths for Referring to an Item in the Module Tree" section, items in child modules can use the items in their ancestor modules. In this test, we bring the tests module’s parent internal_adder into scope with use super::internal_adder; and then the test can call internal_adder. If you don’t think private functions should be tested, there’s nothing in Cairo that will compel you to do so.

Integration Tests

Integration tests use your library in the same way any other code would. Their purpose is to test whether many parts of your library work together correctly. Units of code that work correctly on their own could have problems when integrated, so test coverage of the integrated code is important as well. To create integration tests, you first need a tests directory.

The tests Directory

We create a tests directory at the top level of our project directory, next to src. Scarb knows to look for integration test files in this directory. We can then make as many test files as we want, and Scarb will compile each of the files as an individual crate.

Let’s create an integration test. With the code in Listing 10-10 still in the src/lib.cairo file, make a tests directory, and create a new file named tests/integration_test.cairo. Your directory structure should look like this:

adder
├── Scarb.lock
├── Scarb.toml
├── src
│   └── lib.cairo
└── tests
    └── integration_tests.cairo

Enter the code in Listing 10-11 into the tests/integration_test.cairo file:

Filename: tests/integration_tests.cairo

use adder::add_two;

#[test]
fn it_adds_two() {
    assert_eq!(4, add_two(2));
}

Listing 10-11: An integration test of a function in the adder crate

Each file in the tests directory is a separate crate, so we need to bring our library into each test crate’s scope. For that reason we add use adder::add_two at the top of the code, which we didn’t need in the unit tests.

We don’t need to annotate any code in tests/integration_test.cairo with #[cfg(test)]. Scarb treats the tests directory specially and compiles files in this directory only when we run scarb test. Run scarb test now:

$ scarb test 
     Running test adder (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Blocking waiting for file lock on build directory
    Finished `release` profile [optimized] target(s) in 0.88s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_09_integration_test/Scarb.toml)
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_09_integration_test/Scarb.toml)
    Finished `dev` profile target(s) in 9 seconds


Collected 2 test(s) from adder package
Running 1 test(s) from tests/
[PASS] adder_integrationtest::integration_tests::it_adds_two (gas: ~1)
Running 1 test(s) from src/
[PASS] adder::tests::internal (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

The two sections of output include the unit tests and the integration tests. Note that if any test in a section fails, the following sections will not be run. For example, if a unit test fails, there won’t be any output for integration tests because those tests will only be run if all unit tests are passing.

The first displayed section is for the integration tests. Each integration test file has its own section, so if we add more files in the tests directory, there will be more integration test sections.

The second displayed section is the same as we’ve been seeing: one line for each unit test (one named add that we added just above) and then a summary line for the unit tests.

We can still run a particular integration test function by specifying the test function’s name as an argument of the option -f to scarb test like for instance scarb test -f integration_tests::internal. To run all the tests in a particular integration test file, we use the same option of scarb test but using only the name of the file.

Then, to run all of our integration tests, we can just add a filter to only run tests whose path contains integration_tests.

$ scarb test -f integration_tests
     Running cairo-test adder
   Compiling test(adder_unittest) adder v0.1.0 (cairo-book/listings/ch10-testing-cairo-programs/no_listing_09_integration_test/Scarb.toml)
   Compiling test(adder_integration_tests) adder_integration_tests v0.1.0 (cairo-book/listings/ch10-testing-cairo-programs/no_listing_09_integration_test/Scarb.toml)
    Finished release target(s) in 7 seconds
testing adder ...
running 1 test
test adder_integration_tests::integration_tests::internal ... ok (gas usage est.: 23110)
test result: ok. 1 passed; 0 failed; 0 ignored; 0 filtered out;

running 0 tests
test result: ok. 0 passed; 0 failed; 0 ignored; 1 filtered out;

We see that in the second section for the unit tests, 1 has been filtered out because it is not in the integration_tests file.

Submodules in Integration Tests

As you add more integration tests, you might want to make more files in the tests directory to help organize them; for example, you can group the test functions by the functionality they’re testing. As mentioned earlier, each file in the tests directory is compiled as its own separate crate, which is useful for creating separate scopes to more closely imitate the way end users will be using your crate. However, this means files in the tests directory don’t share the same behavior as files in src do, as you learned in Chapter 7 regarding how to separate code into modules and files.

The different behavior of tests directory files is most noticeable when you have a set of helper functions to use in multiple integration test files and you try to follow the steps in the Separating Modules into Different Files section of Chapter 7 to extract them into a common module. For example, if we create tests/common.cairo and place a function named setup in it, we can add some code to setup that we want to call from multiple test functions in multiple test files:

Filename: tests/common.cairo

pub fn setup() {
    println!("Setting up tests...");
}

Filename: tests/integration_tests.cairo

use adder::it_adds_two;

#[test]
fn internal() {
    assert!(it_adds_two(2, 2) == 4, "internal_adder failed");
}

Filename: src/lib.cairo

pub fn it_adds_two(a: u8, b: u8) -> u8 {
    a + b
}

#[cfg(test)]
mod tests {
    #[test]
    fn add() {
        assert_eq!(4, super::it_adds_two(2, 2));
    }
}

When we run the tests with scarb test, we’ll see a new section in the test output for the common.cairo file, even though this file doesn’t contain any test functions nor did we call the setup function from anywhere:

$ scarb test 
     Running test adder (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Finished `release` profile [optimized] target(s) in 0.64s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_12_submodules/Scarb.toml)
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_12_submodules/Scarb.toml)
    Finished `dev` profile target(s) in 8 seconds


Collected 2 test(s) from adder package
Running 1 test(s) from src/
[PASS] adder::tests::add (gas: ~1)
Running 1 test(s) from tests/
[PASS] adder_integrationtest::integration_tests::internal (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

To avoid systematically getting a section for each file of the tests folder, we also have the option of making the tests/ directory behave like a regular crate, by adding a tests/lib.cairo file. In that case, the tests directory will no longer compile as one crate per file, but as one crate for the whole directory.

Let's create this tests/lib.cairo file :

Filename: tests/lib.cairo

mod integration_tests;
mod common;

The project directory will now look like this :

adder
├── Scarb.lock
├── Scarb.toml
├── src
│   └── lib.cairo
└── tests
    ├── common.cairo
    ├── integration_tests.cairo
    └── lib.cairo

When we run the scarb test command again, here is the output :

$ scarb test 
     Running test adder (snforge test)
   Compiling snforge_scarb_plugin v0.31.0 (git+https://github.com/foundry-rs/starknet-foundry.git?tag=v0.31.0#72ea785ca354e9e506de3e5d687da9fb2c1b3c67)
    Finished `release` profile [optimized] target(s) in 0.51s
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_13_single_integration_crate/Scarb.toml)
   Compiling test(listings/ch10-testing-cairo-programs/no_listing_13_single_integration_crate/Scarb.toml)
    Finished `dev` profile target(s) in 7 seconds


Collected 2 test(s) from adder package
Running 1 test(s) from tests/
[PASS] adder_tests::integration_tests::internal (gas: ~1)
Running 1 test(s) from src/
[PASS] adder::tests::add (gas: ~1)
Tests: 2 passed, 0 failed, 0 skipped, 0 ignored, 0 filtered out

This way, only the test functions will be tested and the setup function can be imported without being tested.

Summary

Cairo's testing features provide a way to specify how code should function to ensure it continues to work as you expect, even as you make changes. Unit tests exercise different parts of a library separately and can test private implementation details. Integration tests check that many parts of the library work together correctly, and they use the library’s public API to test the code in the same way external code will use it. Even though Cairo's type system and ownership rules help prevent some kinds of bugs, tests are still important to reduce logic bugs having to do with how your code is expected to behave.

Advanced Features

Now, let's learn about more advanced features offered by Cairo.

Custom Data Structures

When you first start programming in Cairo, you'll likely want to use arrays (Array<T>) to store collections of data. However, you will quickly realize that arrays have one big limitation - the data stored in them is immutable. Once you append a value to an array, you can't modify it.

This can be frustrating when you want to use a mutable data structure. For example, say you're making a game where the players have a level, and they can level up. You might try to store the level of the players in an array:

    let mut level_players = array![5, 1, 10];

But then you realize you can't increase the level at a specific index once it's set. If a player dies, you cannot remove it from the array unless he happens to be in the first position.

Fortunately, Cairo provides a handy built-in dictionary type called Felt252Dict<T> that allows us to simulate the behavior of mutable data structures. Let's first explore how to create a struct that contains, among others, a Felt252Dict<T>.

Note: Several concepts used in this chapter were already presented earlier in the book. We recommend checking out the following chapters if you need to revise them: Structs, Methods, Generic types, Traits.

Dictionaries as Struct Members

Defining dictionaries as struct members is possible in Cairo but correctly interacting with them may not be entirely seamless. Let's try implementing a custom user database that will allow us to add users and query them. We will need to define a struct to represent the new type and a trait to define its functionality:

struct UserDatabase<T> {
    users_updates: u64,
    balances: Felt252Dict<T>,
}

trait UserDatabaseTrait<T> {
    fn new() -> UserDatabase<T>;
    fn update_user<+Drop<T>>(ref self: UserDatabase<T>, name: felt252, balance: T);
    fn get_balance<+Copy<T>>(ref self: UserDatabase<T>, name: felt252) -> T;
}

Our new type UserDatabase<T> represents a database of users. It is generic over the balances of the users, giving major flexibility to whoever uses our data type. Its two members are:

• users_updates, the number of users updates in the dictionary.
• balances, a mapping of each user to its balance.

The database core functionality is defined by UserDatabaseTrait. The following methods are defined:

• new for easily creating new UserDatabase types.
• update_user to update the balance of users in the database.
• get_balance to find user's balance in the database.

The only remaining step is to implement each of the methods in UserDatabaseTrait, but since we are working with Generic types we also need to correctly establish the requirements of T so it can be a valid Felt252Dict<T> value type:

1. T should implement the Copy<T> since it's required for getting values from a Felt252Dict<T>.
2. All value types of a dictionary implement the Felt252DictValue<T>, our generic type should do as well.
3. To insert values, Felt252DictTrait<T> requires all value types to be droppable (implement the Drop<T> trait).

The implementation, with all restrictions in place, would be as follows:

impl UserDatabaseImpl<T, +Felt252DictValue<T>> of UserDatabaseTrait<T> {
    // Creates a database
    fn new() -> UserDatabase<T> {
        UserDatabase { users_updates: 0, balances: Default::default() }
    }

    // Get the user's balance
    fn get_balance<+Copy<T>>(ref self: UserDatabase<T>, name: felt252) -> T {
        self.balances.get(name)
    }

    // Add a user
    fn update_user<+Drop<T>>(ref self: UserDatabase<T>, name: felt252, balance: T) {
        self.balances.insert(name, balance);
        self.users_updates += 1;
    }
}

Our database implementation is almost complete, except for one thing: the compiler doesn't know how to make a UserDatabase<T> go out of scope, since it doesn't implement the Drop<T> trait, nor the Destruct<T> trait. Since it has a Felt252Dict<T> as a member, it cannot be dropped, so we are forced to implement the Destruct<T> trait manually (refer to the Ownership chapter for more information). Using #[derive(Destruct)] on top of the UserDatabase<T> definition won't work because of the use of Generic types in the struct definition. We need to code the Destruct<T> trait implementation by ourselves:

impl UserDatabaseDestruct<T, +Drop<T>, +Felt252DictValue<T>> of Destruct<UserDatabase<T>> {
    fn destruct(self: UserDatabase<T>) nopanic {
        self.balances.squash();
    }
}

Implementing Destruct<T> for UserDatabase was our last step to get a fully functional database. We can now try it out:

use core::dict::Felt252Dict;

struct UserDatabase<T> {
    users_updates: u64,
    balances: Felt252Dict<T>,
}

trait UserDatabaseTrait<T> {
    fn new() -> UserDatabase<T>;
    fn update_user<+Drop<T>>(ref self: UserDatabase<T>, name: felt252, balance: T);
    fn get_balance<+Copy<T>>(ref self: UserDatabase<T>, name: felt252) -> T;
}

impl UserDatabaseImpl<T, +Felt252DictValue<T>> of UserDatabaseTrait<T> {
    // Creates a database
    fn new() -> UserDatabase<T> {
        UserDatabase { users_updates: 0, balances: Default::default() }
    }

    // Get the user's balance
    fn get_balance<+Copy<T>>(ref self: UserDatabase<T>, name: felt252) -> T {
        self.balances.get(name)
    }

    // Add a user
    fn update_user<+Drop<T>>(ref self: UserDatabase<T>, name: felt252, balance: T) {
        self.balances.insert(name, balance);
        self.users_updates += 1;
    }
}

impl UserDatabaseDestruct<T, +Drop<T>, +Felt252DictValue<T>> of Destruct<UserDatabase<T>> {
    fn destruct(self: UserDatabase<T>) nopanic {
        self.balances.squash();
    }
}

fn main() {
    let mut db = UserDatabaseTrait::<u64>::new();

    db.update_user('Alex', 100);
    db.update_user('Maria', 80);

    db.update_user('Alex', 40);
    db.update_user('Maria', 0);

    let alex_latest_balance = db.get_balance('Alex');
    let maria_latest_balance = db.get_balance('Maria');

    assert!(alex_latest_balance == 40, "Expected 40");
    assert!(maria_latest_balance == 0, "Expected 0");
}

Simulating a Dynamic Array with Dicts

First, let's think about how we want our mutable dynamic array to behave. What operations should it support?

It should:

• Allow us to append items at the end.
• Let us access any item by index.
• Allow setting the value of an item at a specific index.
• Return the current length.

We can define this interface in Cairo like:

trait MemoryVecTrait<V, T> {
    fn new() -> V;
    fn get(ref self: V, index: usize) -> Option<T>;
    fn at(ref self: V, index: usize) -> T;
    fn push(ref self: V, value: T) -> ();
    fn set(ref self: V, index: usize, value: T);
    fn len(self: @V) -> usize;
}

This provides a blueprint for the implementation of our dynamic array. We named it MemoryVec as it is similar to the Vec<T> data structure in Rust.

Note: The core library of Cairo already includes a Vec<T> data structure, strictly used as a storage type in smart contracts. To differentiate our data structure from the core library's one, we named our implementation MemoryVec.

Implementing a Dynamic Array in Cairo

To store our data, we'll use a Felt252Dict<T> which maps index numbers (felts) to values. We'll also store a separate len field to track the length.

Here is what our struct looks like. We wrap the type T inside Nullable pointer to allow using any type T in our data structure, as explained in the Dictionaries section:

use core::dict::Felt252Dict;
use core::nullable::NullableTrait;
use core::num::traits::WrappingAdd;

trait MemoryVecTrait<V, T> {
    fn new() -> V;
    fn get(ref self: V, index: usize) -> Option<T>;
    fn at(ref self: V, index: usize) -> T;
    fn push(ref self: V, value: T) -> ();
    fn set(ref self: V, index: usize, value: T);
    fn len(self: @V) -> usize;
}

struct MemoryVec<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize
}

impl DestructMemoryVec<T, +Drop<T>> of Destruct<MemoryVec<T>> {
    fn destruct(self: MemoryVec<T>) nopanic {
        self.data.squash();
    }
}

impl MemoryVecImpl<T, +Drop<T>, +Copy<T>> of MemoryVecTrait<MemoryVec<T>, T> {
    fn new() -> MemoryVec<T> {
        MemoryVec { data: Default::default(), len: 0 }
    }

    fn get(ref self: MemoryVec<T>, index: usize) -> Option<T> {
        if index < self.len() {
            Option::Some(self.data.get(index.into()).deref())
        } else {
            Option::None
        }
    }

    fn at(ref self: MemoryVec<T>, index: usize) -> T {
        assert!(index < self.len(), "Index out of bounds");
        self.data.get(index.into()).deref()
    }

    fn push(ref self: MemoryVec<T>, value: T) -> () {
        self.data.insert(self.len.into(), NullableTrait::new(value));
        self.len.wrapping_add(1_usize);
    }
    fn set(ref self: MemoryVec<T>, index: usize, value: T) {
        assert!(index < self.len(), "Index out of bounds");
        self.data.insert(index.into(), NullableTrait::new(value));
    }
    fn len(self: @MemoryVec<T>) -> usize {
        *self.len
    }
}

Since we again have Felt252Dict<T> as a struct member, we need to implement the Destruct<T> trait to tell the compiler how to make MemoryVec<T> go out of scope.

use core::dict::Felt252Dict;
use core::nullable::NullableTrait;
use core::num::traits::WrappingAdd;

trait MemoryVecTrait<V, T> {
    fn new() -> V;
    fn get(ref self: V, index: usize) -> Option<T>;
    fn at(ref self: V, index: usize) -> T;
    fn push(ref self: V, value: T) -> ();
    fn set(ref self: V, index: usize, value: T);
    fn len(self: @V) -> usize;
}

struct MemoryVec<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize
}

impl DestructMemoryVec<T, +Drop<T>> of Destruct<MemoryVec<T>> {
    fn destruct(self: MemoryVec<T>) nopanic {
        self.data.squash();
    }
}

impl MemoryVecImpl<T, +Drop<T>, +Copy<T>> of MemoryVecTrait<MemoryVec<T>, T> {
    fn new() -> MemoryVec<T> {
        MemoryVec { data: Default::default(), len: 0 }
    }

    fn get(ref self: MemoryVec<T>, index: usize) -> Option<T> {
        if index < self.len() {
            Option::Some(self.data.get(index.into()).deref())
        } else {
            Option::None
        }
    }

    fn at(ref self: MemoryVec<T>, index: usize) -> T {
        assert!(index < self.len(), "Index out of bounds");
        self.data.get(index.into()).deref()
    }

    fn push(ref self: MemoryVec<T>, value: T) -> () {
        self.data.insert(self.len.into(), NullableTrait::new(value));
        self.len.wrapping_add(1_usize);
    }
    fn set(ref self: MemoryVec<T>, index: usize, value: T) {
        assert!(index < self.len(), "Index out of bounds");
        self.data.insert(index.into(), NullableTrait::new(value));
    }
    fn len(self: @MemoryVec<T>) -> usize {
        *self.len
    }
}

The key thing that makes this vector mutable is that we can insert values into the dictionary to set or update values in our data structure. For example, to update a value at a specific index, we do:

use core::dict::Felt252Dict;
use core::nullable::NullableTrait;
use core::num::traits::WrappingAdd;

trait MemoryVecTrait<V, T> {
    fn new() -> V;
    fn get(ref self: V, index: usize) -> Option<T>;
    fn at(ref self: V, index: usize) -> T;
    fn push(ref self: V, value: T) -> ();
    fn set(ref self: V, index: usize, value: T);
    fn len(self: @V) -> usize;
}

struct MemoryVec<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize
}

impl DestructMemoryVec<T, +Drop<T>> of Destruct<MemoryVec<T>> {
    fn destruct(self: MemoryVec<T>) nopanic {
        self.data.squash();
    }
}

impl MemoryVecImpl<T, +Drop<T>, +Copy<T>> of MemoryVecTrait<MemoryVec<T>, T> {
    fn new() -> MemoryVec<T> {
        MemoryVec { data: Default::default(), len: 0 }
    }

    fn get(ref self: MemoryVec<T>, index: usize) -> Option<T> {
        if index < self.len() {
            Option::Some(self.data.get(index.into()).deref())
        } else {
            Option::None
        }
    }

    fn at(ref self: MemoryVec<T>, index: usize) -> T {
        assert!(index < self.len(), "Index out of bounds");
        self.data.get(index.into()).deref()
    }

    fn push(ref self: MemoryVec<T>, value: T) -> () {
        self.data.insert(self.len.into(), NullableTrait::new(value));
        self.len.wrapping_add(1_usize);
    }
    fn set(ref self: MemoryVec<T>, index: usize, value: T) {
        assert!(index < self.len(), "Index out of bounds");
        self.data.insert(index.into(), NullableTrait::new(value));
    }
    fn len(self: @MemoryVec<T>) -> usize {
        *self.len
    }
}

This overwrites the previously existing value at that index in the dictionary.

While arrays are immutable, dictionaries provide the flexibility we need for modifiable data structures like vectors.

The implementation of the rest of the interface is straightforward. The implementation of all the methods defined in our interface can be done as follow :

use core::dict::Felt252Dict;
use core::nullable::NullableTrait;
use core::num::traits::WrappingAdd;

trait MemoryVecTrait<V, T> {
    fn new() -> V;
    fn get(ref self: V, index: usize) -> Option<T>;
    fn at(ref self: V, index: usize) -> T;
    fn push(ref self: V, value: T) -> ();
    fn set(ref self: V, index: usize, value: T);
    fn len(self: @V) -> usize;
}

struct MemoryVec<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize
}

impl DestructMemoryVec<T, +Drop<T>> of Destruct<MemoryVec<T>> {
    fn destruct(self: MemoryVec<T>) nopanic {
        self.data.squash();
    }
}

impl MemoryVecImpl<T, +Drop<T>, +Copy<T>> of MemoryVecTrait<MemoryVec<T>, T> {
    fn new() -> MemoryVec<T> {
        MemoryVec { data: Default::default(), len: 0 }
    }

    fn get(ref self: MemoryVec<T>, index: usize) -> Option<T> {
        if index < self.len() {
            Option::Some(self.data.get(index.into()).deref())
        } else {
            Option::None
        }
    }

    fn at(ref self: MemoryVec<T>, index: usize) -> T {
        assert!(index < self.len(), "Index out of bounds");
        self.data.get(index.into()).deref()
    }

    fn push(ref self: MemoryVec<T>, value: T) -> () {
        self.data.insert(self.len.into(), NullableTrait::new(value));
        self.len.wrapping_add(1_usize);
    }
    fn set(ref self: MemoryVec<T>, index: usize, value: T) {
        assert!(index < self.len(), "Index out of bounds");
        self.data.insert(index.into(), NullableTrait::new(value));
    }
    fn len(self: @MemoryVec<T>) -> usize {
        *self.len
    }
}

The full implementation of the MemoryVec structure can be found in the community-maintained library Alexandria.

Simulating a Stack with Dicts

We will now look at a second example and its implementation details: a Stack.

A Stack is a LIFO (Last-In, First-Out) collection. The insertion of a new element and removal of an existing element takes place at the same end, represented as the top of the stack.

Let us define what operations we need to create a stack:

• Push an item to the top of the stack.
• Pop an item from the top of the stack.
• Check whether there are still any elements in the stack.

From these specifications we can define the following interface :

trait StackTrait<S, T> {
    fn push(ref self: S, value: T);
    fn pop(ref self: S) -> Option<T>;
    fn is_empty(self: @S) -> bool;
}

Implementing a Mutable Stack in Cairo

To create a stack data structure in Cairo, we can again use a Felt252Dict<T> to store the values of the stack along with a usize field to keep track of the length of the stack to iterate over it.

The Stack struct is defined as:

struct NullableStack<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize,
}

Next, let's see how our main functions push and pop are implemented.

use core::dict::Felt252Dict;
use core::nullable::{match_nullable, FromNullableResult, NullableTrait};

trait StackTrait<S, T> {
    fn push(ref self: S, value: T);
    fn pop(ref self: S) -> Option<T>;
    fn is_empty(self: @S) -> bool;
}

struct NullableStack<T> {
    data: Felt252Dict<Nullable<T>>,
    len: usize,
}

impl DestructNullableStack<T, +Drop<T>> of Destruct<NullableStack<T>> {
    fn destruct(self: NullableStack<T>) nopanic {
        self.data.squash();
    }
}


impl NullableStackImpl<T, +Drop<T>, +Copy<T>> of StackTrait<NullableStack<T>, T> {
    fn push(ref self: NullableStack<T>, value: T) {
        self.data.insert(self.len.into(), NullableTrait::new(value));
        self.len += 1;
    }

    fn pop(ref self: NullableStack<T>) -> Option<T> {
        if self.is_empty() {
            return Option::None;
        }
        self.len -= 1;
        Option::Some(self.data.get(self.len.into()).deref())
    }

    fn is_empty(self: @NullableStack<T>) -> bool {
        *self.len == 0
    }
}

The code uses the insert and get methods to access the values in the Felt252Dict<T>. To push an element to the top of the stack, the push function inserts the element in the dict at index len and increases the len field of the stack to keep track of the position of the stack top. To remove a value, the pop function decreases the value of len to update the position of the stack top and then retrieves the last value at position len.

The full implementation of the Stack, along with more data structures that you can use in your code, can be found in the community-maintained Alexandria library, in the "data_structures" crate.

Summary

Well done! Now you have knowledge of arrays, dictionaries and even custom data structures. While Cairo's memory model is immutable and can make it difficult to implement mutable data structures, we can fortunately use the Felt252Dict<T> type to simulate mutable data structures. This allows us to implement a wide range of data structures that are useful for many applications, effectively hiding the complexity of the underlying memory model.

Smart Pointers

A pointer is a general concept for a variable that contains a memory address. This address refers to, or “points at,” some other data. While pointers are a powerful feature, they can also be a source of bugs and security vulnerabilities. For example, a pointer can reference an unassigned memory cell, which means that attempting to access the data at that address would cause the program to crash, making it unprovable. To prevent such issues, Cairo uses Smart Pointers.

Smart pointers are data structures that act like a pointer, but also have additional metadata and capabilities. The concept of smart pointers isn’t unique to Cairo: smart pointers originated in C++ and exist in other languages like Rust as well. In the specific case of Cairo, smart pointers ensure that memory is not addressed in an unsafe way that could cause a program to be unprovable, by providing a safe way to access memory through strict type checking and ownership rules.

Though we didn’t call them as such at the time, we’ve already encountered a few smart pointers in this book, including Felt252Dict<T> and Array<T> in Chapter 3. Both these types count as smart pointers because they own a memory segment and allow you to manipulate it. They also have metadata and extra capabilities or guarantees. Arrays keep track of their current length to ensure that existing elements are not overwritten, and that new elements are only appended to the end.

The Cairo VM memory is composed by multiple segments that can store data, each identified by a unique index. When you create an array, you allocate a new segment in the memory to store the future elements. The array itself is just a pointer to that segment where the elements are stored.

The Box<T> Type to Manipulate Pointers

The principal smart pointer type in Cairo is a box, denoted as Box<T>. Manually defining boxes allow you to store data in a specific memory segment of the Cairo VM called the boxed segment. This segment is dedicated to store all boxed values, and what remains in the execution segment is only a pointer to the boxed segment. Whenever you instantiate a new pointer variable of type Box<T>, you append the data of type T to the boxed segment.

Boxes have very little performance overhead, other than writing their inner values to the boxed segment. But they don’t have many extra capabilities either. You’ll use them most often in these situations:

• When you have a type whose size can’t be known at compile time and you want to use a value of that type in a context that requires an exact size
• When you have a large amount of data and you want to transfer ownership but ensure the data won’t be copied when you do so

We’ll demonstrate the first situation in the “Enabling Recursive Types with Boxes” section. In the second case, transferring ownership of a large amount of data can take a long time because the data is copied around in memory. To improve performance in this situation, we can store the large amount of data in the boxed segment using a box type. Then, only the small amount of pointer data is copied around in memory, while the data it references stays in one place on the boxed segment.

Using a Box<T> to Store Data in the Boxed Segment

Before we discuss the boxed segment storage use cases for Box<T>, we’ll cover the syntax and how to interact with values stored within a Box<T>.

Listing 11-1 shows how to use a box to store a value in the boxed segment:

fn main() {
    let b = BoxTrait::new(5_u128);
    println!("b = {}", b.unbox())
}

Listing 11-1: Storing a u128 value in the boxed segment using a box

We define the variable b to have the value of a Box that points to the value 5, which is stored in the boxed segment. This program will print b = 5; in this case, we can access the data in the box similar to how we would if this data was simply in the execution memory. Putting a single value in a box isn’t very useful, so you won’t use boxes by themselves in this way very often. Having values like a single u128 in the execution memory, where they’re stored by default, is more appropriate in the majority of situations. Let’s look at a case where boxes allow us to define types that we wouldn’t be allowed to if we didn’t have boxes.

Enabling Recursive Types with Boxes

A value of recursive type can have another value of the same type as part of itself. Recursive types pose an issue because at compile time because Cairo needs to know how much space a type takes up. However, the nesting of values of recursive types could theoretically continue infinitely, so Cairo can’t know how much space the value needs. Because boxes have a known size, we can enable recursive types by inserting a box in the recursive type definition.

As an example of a recursive type, let’s explore the implementation of a binary tree. The binary tree type we’ll define is straightforward except for the recursion; therefore, the concepts in the example we’ll work with will be useful any time you get into more complex situations involving recursive types.

A binary tree is a tree data structure in which each node has at most two children, which are referred to as the left child and the right child. The last element of a branch is a leaf, which is a node without children.

Listing 11-2 shows an attempt to implement a binary tree of u32 values. Note that this code won’t compile yet because the BinaryTree type doesn’t have a known size, which we’ll demonstrate.

#[derive(Copy, Drop)]
enum BinaryTree {
    Leaf: u32,
    Node: (u32, BinaryTree, BinaryTree),
}

fn main() {
    let leaf1 = BinaryTree::Leaf(1);
    let leaf2 = BinaryTree::Leaf(2);
    let leaf3 = BinaryTree::Leaf(3);
    let node = BinaryTree::Node((4, leaf2, leaf3));
    let _root = BinaryTree::Node((5, leaf1, node));
}

Listing 11-2: The first attempt at implementing a binary tree of u32 values

Note: We’re implementing a binary tree that holds only u32 values for the purposes of this example. We could have implemented it using generics, as we discussed in Chapter 8, to define a binary tree that could store values of any type.

The root node contains 5 and two child nodes. The left child is a leaf containing 1. The right child is another node containing 4, which in turn has two leaf children: one containing 2 and another containing 3. This structure forms a simple binary tree with a depth of 2.

If we try to compile the code in listing 11-2, we get the following error:

$ scarb build 
   Compiling listing_recursive_types_wrong v0.1.0 (listings/ch11-advanced-features/listing_recursive_types_wrong/Scarb.toml)
error: Recursive type "(core::integer::u32, listing_recursive_types_wrong::BinaryTree, listing_recursive_types_wrong::BinaryTree)" has infinite size.
 --> listings/ch11-advanced-features/listing_recursive_types_wrong/src/lib.cairo:6:5
    Node: (u32, BinaryTree, BinaryTree),
    ^*********************************^

error: Recursive type "listing_recursive_types_wrong::BinaryTree" has infinite size.
 --> listings/ch11-advanced-features/listing_recursive_types_wrong/src/lib.cairo:10:17
    let leaf1 = BinaryTree::Leaf(1);
                ^*****************^

error: Recursive type "(core::integer::u32, listing_recursive_types_wrong::BinaryTree, listing_recursive_types_wrong::BinaryTree)" has infinite size.
 --> listings/ch11-advanced-features/listing_recursive_types_wrong/src/lib.cairo:13:33
    let node = BinaryTree::Node((4, leaf2, leaf3));
                                ^***************^

error: could not compile `listing_recursive_types_wrong` due to previous error

The error shows this type “has infinite size.” The reason is that we’ve defined BinaryTree with a variant that is recursive: it holds another value of itself directly. As a result, Cairo can’t figure out how much space it needs to store a BinaryTree value.

Hopefully, we can fix this error by using a Box<T> to store the recursive variant of BinaryTree. Because a Box<T> is a pointer, Cairo always knows how much space a Box<T> needs: a pointer’s size doesn’t change based on the amount of data it’s pointing to. This means we can put a Box<T> inside the Node variant instead of another BinaryTree value directly. The Box<T> will point to the child BinaryTree values that will be stored in their own segment, rather than inside the Node variant. Conceptually, we still have a binary tree, created with binary trees holding other binary trees, but this implementation is now more like placing the items next to one another rather than inside one another.

We can change the definition of the BinaryTree enum in Listing 11-2 and the usage of the BinaryTree in Listing 11-2 to the code in Listing 11-3, which will compile:

use core::box::{BoxTrait};

mod display;
use display::DebugBinaryTree;

#[derive(Copy, Drop)]
enum BinaryTree {
    Leaf: u32,
    Node: (u32, Box<BinaryTree>, Box<BinaryTree>),
}


fn main() {
    let leaf1 = BinaryTree::Leaf(1);
    let leaf2 = BinaryTree::Leaf(2);
    let leaf3 = BinaryTree::Leaf(3);
    let node = BinaryTree::Node((4, BoxTrait::new(leaf2), BoxTrait::new(leaf3)));
    let root = BinaryTree::Node((5, BoxTrait::new(leaf1), BoxTrait::new(node)));

    println!("{:?}", root);
}

Listing 11-3: Defining a recursive Binary Tree using Boxes

The Node variant now holds a (u32, Box<BinaryTree>, Box<BinaryTree>), indicating that the Node variant will store a u32 value, and two Box<BinaryTree> values. Now, we know that the Node variant will need a size of u32 plus the size of the two Box<BinaryTree> values. By using a box, we’ve broken the infinite, recursive chain, so the compiler can figure out the size it needs to store a BinaryTree value.

Using Boxes to Improve Performance

Passing pointers between functions allows you to reference data without copying the data itself. Using boxes can improve performance as it allows you to pass a pointer to some data from one function to another, without the need to copy the entire data in memory before performing the function call. Instead of having to write n values into memory before calling a function, only a single value is written, corresponding to the pointer to the data. If the data stored in the box is very large, the performance improvement can be significant, as you would save n-1 memory operations before each function call.

Let's take a look at the code in Listing 11-4, which shows two ways of passing data to a function: by value and by pointer.

#[derive(Drop)]
struct Cart {
    paid: bool,
    items: u256,
    buyer: ByteArray
}

fn pass_data(cart: Cart) {
    println!("{} is shopping today and bought {} items", cart.buyer, cart.items);
}

fn pass_pointer(cart: Box<Cart>) {
    let cart = cart.unbox();
    println!("{} is shopping today and bought {} items", cart.buyer, cart.items);
}

fn main() {
    let new_struct = Cart { paid: true, items: 1, buyer: "Eli" };
    pass_data(new_struct);

    let new_box = BoxTrait::new(Cart { paid: false, items: 2, buyer: "Uri" });
    pass_pointer(new_box);
}

Listing 11-4: Storing large amounts of data in a box for performance.

The main function includes 2 function calls:

• pass_data that takes a variable of type Cart.
• pass_pointer that takes a pointer of type Box<Cart>.

When passing data to a function, the entire data is copied into the last available memory cells right before the function call. Calling pass_data will copy all 3 fields of Cart to memory, while pass_pointer only requires the copy of the new_box pointer which is of size 1.

CairoVM Memory layout when using boxes

The illustration above demonstrates how the memory behaves in both cases. The first instance of Cart is stored in the execution segment, and we need to copy all its fields to memory before calling the pass_data function. The second instance of Cart is stored in the boxed segment, and the pointer to it is stored in the execution segment. When calling the pass_pointer function, only the pointer to the struct is copied to memory right before the function call. In both cases, however, instantiating the struct will store all its values in the execution segment: the boxed segment can only be filled with data taken from the execution segment.

The Nullable<T> Type for Dictionaries

Nullable<T> is another type of smart pointer that can either point to a value or be null in the absence of value. It is defined at the Sierra level. This type is mainly used in dictionaries that contain types that don't implement the zero_default method of the Felt252DictValue<T> trait (i.e., arrays and structs).

If we try to access an element that does not exist in a dictionary, the code will fail if the zero_default method cannot be called.

Chapter 3 about dictionaries thoroughly explains how to store a Span<felt252> variable inside a dictionary using the Nullable<T> type. Please refer to it for further information.

Deref Coercion

Deref coercion simplifies the way we interact with nested or wrapped data structures by allowing an instance of one type to behave like an instance of another type. This mechanism is enabled by implementing the Deref trait, which allows implicit conversion (or coercion) to a different type, providing direct access to the underlying data.

Note: For now, deref coercion allows you to access a member of a type T as if it was a type K, but will not allow you to call functions whose self argument is of the original type when holding an instance of the coerced type.

Deref coercion is implemented via the Deref and DerefMut traits. When a type T implements Deref or DerefMut to type K, instances of T can access the members of K directly.

The Deref trait in Cairo is defined as follows:

pub trait Deref<T> {
    type Target;
    fn deref(self: T) -> Self::Target;
}

pub trait DerefMut<T> {
    type Target;
    fn deref_mut(ref self: T) -> Self::Target;
}

The Target type specifies the result of dereferencing, and the deref method defines how to transform T into K.

Using Deref Coercion

To better understand how deref coercion works, let's look at a practical example. We'll create a simple generic wrapper type around a type T called Wrapper<T>, and use it to wrap a UserProfile struct.

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefWrapper<T> of Deref<Wrapper<T>> {
    type Target = T;
    fn deref(self: Wrapper<T>) -> T {
        self.value
    }
}

fn main() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Access fields directly via deref coercion
    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

The Wrapper struct wraps a single value generic of type T. To simplify access to the wrapped value, we implement the Deref trait for Wrapper<T>.

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefWrapper<T> of Deref<Wrapper<T>> {
    type Target = T;
    fn deref(self: Wrapper<T>) -> T {
        self.value
    }
}

fn main() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Access fields directly via deref coercion
    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

This implementation is quite simple. The deref method returns the wrapped value, allowing instances of Wrapper<T> to access the members of T directly.

In practice, this mechanism is totally transparent. The following example demonstrates how, holding an instance of Wrapper<UserProfile>, we can print the username and age fields of the underlying UserProfile instance.

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefWrapper<T> of Deref<Wrapper<T>> {
    type Target = T;
    fn deref(self: Wrapper<T>) -> T {
        self.value
    }
}

fn main() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Access fields directly via deref coercion
    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

Restricting Deref Coercion to Mutable Variables

While Deref works for both mutable and immutable variables, DerefMut will only be applicable to mutable variables. Contrary to what the name might suggest, DerefMut does not provide mutable access to the underlying data.

//TAG: does_not_compile

use core::ops::DerefMut;

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefMutWrapper<T, +Copy<T>> of DerefMut<Wrapper<T>> {
    type Target = T;
    fn deref_mut(ref self: Wrapper<T>) -> T {
        self.value
    }
}

fn error() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Uncommenting the next line will cause a compilation error
    println!("Username: {}", wrapped_profile.username);
}

fn main() {
    let mut wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };

    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

If you try to use DerefMut with an immutable variable, the compiler would throw an error. Here’s an example:

//TAG: does_not_compile

use core::ops::DerefMut;

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefMutWrapper<T, +Copy<T>> of DerefMut<Wrapper<T>> {
    type Target = T;
    fn deref_mut(ref self: Wrapper<T>) -> T {
        self.value
    }
}

fn error() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Uncommenting the next line will cause a compilation error
    println!("Username: {}", wrapped_profile.username);
}

fn main() {
    let mut wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };

    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

Compiling this code will result in the following error:

$ scarb build 
   Compiling no_listing_09_deref_coercion_example v0.1.0 (listings/ch11-advanced-features/no_listing_09_deref_mut_example/Scarb.toml)
error: Type "no_listing_09_deref_coercion_example::Wrapper::<no_listing_09_deref_coercion_example::UserProfile>" has no member "username"
 --> listings/ch11-advanced-features/no_listing_09_deref_mut_example/src/lib.cairo:32:46
    println!("Username: {}", wrapped_profile.username);
                                             ^******^

error: could not compile `no_listing_09_deref_coercion_example` due to previous error

For the above code to work, we need to define wrapped_profile as a mutable variable.

//TAG: does_not_compile

use core::ops::DerefMut;

#[derive(Drop, Copy)]
struct UserProfile {
    username: felt252,
    email: felt252,
    age: u16,
}

#[derive(Drop, Copy)]
struct Wrapper<T> {
    value: T,
}

impl DerefMutWrapper<T, +Copy<T>> of DerefMut<Wrapper<T>> {
    type Target = T;
    fn deref_mut(ref self: Wrapper<T>) -> T {
        self.value
    }
}

fn error() {
    let wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };
    // Uncommenting the next line will cause a compilation error
    println!("Username: {}", wrapped_profile.username);
}

fn main() {
    let mut wrapped_profile = Wrapper {
        value: UserProfile { username: 'john_doe', email: 'john@example.com', age: 30 }
    };

    println!("Username: {}", wrapped_profile.username);
    println!("Current age: {}", wrapped_profile.age);
}

Summary

By using the Deref and DerefMut traits, we can transparently convert one type into another, simplifying the access to nested or wrapped data structures. This feature is particularly useful when working with generic types or building abstractions that require seamless access to the underlying data and can help reduce boilerplate code. However, this functionality is quite limited, as you cannot call functions whose self argument is of the original type when holding an instance of the coerced type.

Associated Items

Associated Items are the items declared in traits or defined in implementations. Specifically, there are associated functions (including methods, that we already covered in Chapter 5), associated types, associated constants, and associated implementations.

Associated items are useful when they are logically related to the implementation. For example, the is_some method on Option is intrinsically related to Options, so should be associated.

Every associated item kind comes in two varieties: definitions that contain the actual implementation and declarations that declare signatures for definitions.

Associated Types

Associated types are type aliases allowing you to define abstract type placeholders within traits. Instead of specifying concrete types in the trait definition, associated types let trait implementers choose the actual types to use.

Let's consider the following Pack trait:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

The Result type in our Pack trait acts as placeholder for a type that will be filled in later. Think of associated types as leaving a blank space in your trait for each implementation to write in the specific type it needs. This approach keeps your trait definition clean and flexible. When you use the trait, you don't need to worry about specifying these types - they're already chosen for you by the implementation. In our Pack trait, the type Result is such a placeholder. The method's definition shows that it will return values of type Self::Result, but it doesn't specify what Result actually is. This is left to the implementers of the Pack trait, who will specify the concrete type for Result. When the pack method is called, it will return a value of that chosen concrete type, whatever it may be.

Let's see how associated types compare to a more traditional generic approach. Suppose we need a function foo that can pack two variables of type T. Without associated types, we might define a PackGeneric trait and an implementation to pack two u32 like this:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

With this approach, foo would be implemented as:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

Notice how foo needs to specify both T and U as generic parameters. Now, let's compare this to our Pack trait with an associated type:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

With associated types, we can define bar more concisely:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

Finally, let's see both approaches in action, demonstrating that the end result is the same:

trait Pack<T> {
    type Result;

    fn pack(self: T, other: T) -> Self::Result;
}

impl PackU32Impl of Pack<u32> {
    type Result = u64;

    fn pack(self: u32, other: u32) -> Self::Result {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn bar<T, impl PackImpl: Pack<T>>(self: T, b: T) -> PackImpl::Result {
    PackImpl::pack(self, b)
}

trait PackGeneric<T, U> {
    fn pack_generic(self: T, other: T) -> U;
}

impl PackGenericU32 of PackGeneric<u32, u64> {
    fn pack_generic(self: u32, other: u32) -> u64 {
        let shift: u64 = 0x100000000; // 2^32
        self.into() * shift + other.into()
    }
}

fn foo<T, U, +PackGeneric<T, U>>(self: T, other: T) -> U {
    self.pack_generic(other)
}

fn main() {
    let a: u32 = 1;
    let b: u32 = 1;

    let x = foo(a, b);
    let y = bar(a, b);

    // result is 2^32 + 1
    println!("x: {}", x);
    println!("y: {}", y);
}

As you can see, bar doesn't need to specify a second generic type for the packing result. This information is hidden in the implementation of the Pack trait, making the function signature cleaner and more flexible. Associated types allow us to express the same functionality with less verbosity, while still maintaining the flexibility of generic programming.

Associated Constants

Associated constants are constants associated with a type. They are declared using the const keyword in a trait and defined in its implementation. In our next example, we define a generic Shape trait that we implement for a Triangle and a Square. This trait includes an associated constant, defining the number of sides of the type that implements the trait.

trait Shape<T> {
    const SIDES: u32;
    fn describe() -> ByteArray;
}

struct Triangle {}

impl TriangleShape of Shape<Triangle> {
    const SIDES: u32 = 3;
    fn describe() -> ByteArray {
        "I am a triangle."
    }
}

struct Square {}

impl SquareShape of Shape<Square> {
    const SIDES: u32 = 4;
    fn describe() -> ByteArray {
        "I am a square."
    }
}

fn print_shape_info<T, impl ShapeImpl: Shape<T>>() {
    println!("I have {} sides. {}", ShapeImpl::SIDES, ShapeImpl::describe());
}

fn main() {
    print_shape_info::<Triangle>();
    print_shape_info::<Square>();
}

After that, we create a print_shape_info generic function, which requires that the generic argument implements the Shape trait. This function will use the associated constant to retrieve the number of sides of the geometric figure, and print it along with its description.

trait Shape<T> {
    const SIDES: u32;
    fn describe() -> ByteArray;
}

struct Triangle {}

impl TriangleShape of Shape<Triangle> {
    const SIDES: u32 = 3;
    fn describe() -> ByteArray {
        "I am a triangle."
    }
}

struct Square {}

impl SquareShape of Shape<Square> {
    const SIDES: u32 = 4;
    fn describe() -> ByteArray {
        "I am a square."
    }
}

fn print_shape_info<T, impl ShapeImpl: Shape<T>>() {
    println!("I have {} sides. {}", ShapeImpl::SIDES, ShapeImpl::describe());
}

fn main() {
    print_shape_info::<Triangle>();
    print_shape_info::<Square>();
}

Associated constants allow us to bind a constant number to the Shape trait rather than adding it to the struct or just hardcoding the value in the implementation. This approach provides several benefits:

1. It keeps the constant closely tied to the trait, improving code organization.
2. It allows for compile-time checks to ensure all implementors define the required constant.
3. It ensures two instances of the same type have the same number of sides.

Associated constants can also be used for type-specific behavior or configuration, making them a versatile tool in trait design.

We can ultimately run the print_shape_info and see the output for both Triangle and Square:

trait Shape<T> {
    const SIDES: u32;
    fn describe() -> ByteArray;
}

struct Triangle {}

impl TriangleShape of Shape<Triangle> {
    const SIDES: u32 = 3;
    fn describe() -> ByteArray {
        "I am a triangle."
    }
}

struct Square {}

impl SquareShape of Shape<Square> {
    const SIDES: u32 = 4;
    fn describe() -> ByteArray {
        "I am a square."
    }
}

fn print_shape_info<T, impl ShapeImpl: Shape<T>>() {
    println!("I have {} sides. {}", ShapeImpl::SIDES, ShapeImpl::describe());
}

fn main() {
    print_shape_info::<Triangle>();
    print_shape_info::<Square>();
}

Associated Implementations

Associated implementations allow you to declare that a trait implementation must exist for an associated type. This feature is particularly useful when you want to enforce relationships between types and implementations at the trait level. It ensures type safety and consistency across different implementations of a trait, which is important in generic programming contexts.

To understand the utility of associated implementations, let's examine the Iterator and IntoIterator traits from the Cairo core library, with their respective implementations using ArrayIter<T> as the collection type:

// Collection type that contains a simple array
#[derive(Drop)]
pub struct ArrayIter<T> {
    array: Array<T>,
}

// T is the collection type
pub trait Iterator<T> {
    type Item;
    fn next(ref self: T) -> Option<Self::Item>;
}

impl ArrayIterator<T> of Iterator<ArrayIter<T>> {
    type Item = T;
    fn next(ref self: ArrayIter<T>) -> Option<T> {
        self.array.pop_front()
    }
}

/// Turns a collection of values into an iterator
pub trait IntoIterator<T> {
    /// The iterator type that will be created
    type IntoIter;
    impl Iterator: Iterator<Self::IntoIter>;

    fn into_iter(self: T) -> Self::IntoIter;
}

impl ArrayIntoIterator<T> of IntoIterator<Array<T>> {
    type IntoIter = ArrayIter<T>;
    fn into_iter(self: Array<T>) -> ArrayIter<T> {
        ArrayIter { array: self }
    }
}

fn main() {
    let mut arr: Array<felt252> = array![1, 2, 3];

    // Converts the array into an iterator
    let mut iter = arr.into_iter();

    // Uses the iterator to print each element
    loop {
        match iter.next() {
            Option::Some(item) => println!("Item: {}", item),
            Option::None => { break; },
        };
    }
}

1. The IntoIterator trait is designed to convert a collection into an iterator.
2. The IntoIter associated type represents the specific iterator type that will be created. This allows different collections to define their own efficient iterator types.
3. The associated implementation Iterator: Iterator<Self::IntoIter> (the key feature we're discussing) declares that this IntoIter type must implement the Iterator trait.
4. This design allows for type-safe iteration without needing to specify the iterator type explicitly every time, improving code ergonomics.

The associated implementation creates a binding at the trait level, guaranteeing that:

• The into_iter method will always return a type that implements Iterator.
• This relationship is enforced for all implementations of IntoIterator, not just on a case-by-case basis.

The following main function demonstrates how this works in practice for an Array<felt252>:

// Collection type that contains a simple array
#[derive(Drop)]
pub struct ArrayIter<T> {
    array: Array<T>,
}

// T is the collection type
pub trait Iterator<T> {
    type Item;
    fn next(ref self: T) -> Option<Self::Item>;
}

impl ArrayIterator<T> of Iterator<ArrayIter<T>> {
    type Item = T;
    fn next(ref self: ArrayIter<T>) -> Option<T> {
        self.array.pop_front()
    }
}

/// Turns a collection of values into an iterator
pub trait IntoIterator<T> {
    /// The iterator type that will be created
    type IntoIter;
    impl Iterator: Iterator<Self::IntoIter>;

    fn into_iter(self: T) -> Self::IntoIter;
}

impl ArrayIntoIterator<T> of IntoIterator<Array<T>> {
    type IntoIter = ArrayIter<T>;
    fn into_iter(self: Array<T>) -> ArrayIter<T> {
        ArrayIter { array: self }
    }
}

fn main() {
    let mut arr: Array<felt252> = array![1, 2, 3];

    // Converts the array into an iterator
    let mut iter = arr.into_iter();

    // Uses the iterator to print each element
    loop {
        match iter.next() {
            Option::Some(item) => println!("Item: {}", item),
            Option::None => { break; },
        };
    }
}

Operator Overloading

Operator overloading is a feature in some programming languages that allows the redefinition of standard operators, such as addition (+), subtraction (-), multiplication (*), and division (/), to work with user-defined types. This can make the syntax of the code more intuitive, by enabling operations on user-defined types to be expressed in the same way as operations on primitive types.

In Cairo, operator overloading is achieved through the implementation of specific traits. Each operator has an associated trait, and overloading that operator involves providing an implementation of that trait for a custom type. However, it's essential to use operator overloading judiciously. Misuse can lead to confusion, making the code more difficult to maintain, for example when there is no semantic meaning to the operator being overloaded.

Consider an example where two Potions need to be combined. Potions have two data fields, mana and health. Combining two Potions should add their respective fields.

struct Potion {
    health: felt252,
    mana: felt252
}

impl PotionAdd of Add<Potion> {
    fn add(lhs: Potion, rhs: Potion) -> Potion {
        Potion { health: lhs.health + rhs.health, mana: lhs.mana + rhs.mana, }
    }
}

fn main() {
    let health_potion: Potion = Potion { health: 100, mana: 0 };
    let mana_potion: Potion = Potion { health: 0, mana: 100 };
    let super_potion: Potion = health_potion + mana_potion;
    // Both potions were combined with the `+` operator.
    assert(super_potion.health == 100, '');
    assert(super_potion.mana == 100, '');
}

In the code above, we're implementing the Add trait for the Potion type. The add function takes two arguments: lhs and rhs (left and right-hand side). The function body returns a new Potion instance, its field values being a combination of lhs and rhs.

As illustrated in the example, overloading an operator requires specification of the concrete type being overloaded. The overloaded generic trait is Add<T>, and we define a concrete implementation for the type Potion with Add<Potion>.

Hashes

At its essence, hashing is a process of converting input data (often called a message) of any length into a fixed-size value, typically referred to as a "hash." This transformation is deterministic, meaning that the same input will always produce the same hash value. Hash functions are a fundamental component in various fields, including data storage, cryptography and data integrity verification. They are very often used when developing smart contracts, especially when working with Merkle trees.

In this chapter, we will present the two hash functions implemented natively in the Cairo core library: Poseidon and Pedersen. We will discuss when and how to use them, and see examples with Cairo programs.

Hash Functions in Cairo

The Cairo core library provides two hash functions: Pedersen and Poseidon.

Pedersen hash functions are cryptographic algorithms that rely on elliptic curve cryptography. These functions perform operations on points along an elliptic curve — essentially, doing math with the locations of these points — which are easy to do in one direction and hard to undo. This one-way difficulty is based on the Elliptic Curve Discrete Logarithm Problem (ECDLP), which is a problem so hard to solve that it ensures the security of the hash function. The difficulty of reversing these operations is what makes the Pedersen hash function secure and reliable for cryptographic purposes.

Poseidon is a family of hash functions designed to be very efficient as algebraic circuits. Its design is particularly efficient for Zero-Knowledge proof systems, including STARKs (so, Cairo). Poseidon uses a method called a 'sponge construction,' which soaks up data and transforms it securely using a process known as the Hades permutation. Cairo's version of Poseidon is based on a three-element state permutation with specific parameters.

When to Use Them?

Pedersen was the first hash function used on Starknet, and is still used to compute the addresses of variables in storage (for example, LegacyMap uses Pedersen to hash the keys of a storage mapping on Starknet). However, as Poseidon is cheaper and faster than Pedersen when working with STARK proofs system, it's now the recommended hash function to use in Cairo programs.

Working with Hashes

The core library makes it easy to work with hashes. The Hash trait is implemented for all types that can be converted to felt252, including felt252 itself. For more complex types like structs, deriving Hash allows them to be hashed easily using the hash function of your choice - given that all of the struct's fields are themselves hashable. You cannot derive the Hash trait on a struct that contains un-hashable values, such as Array<T> or Felt252Dict<T>, even if T itself is hashable.

The Hash trait is accompanied by the HashStateTrait and HashStateExTrait that define the basic methods to work with hashes. They allow you to initialize a hash state that will contain the temporary values of the hash after each application of the hash function, update the hash state and finalize it when the computation is completed. HashStateTrait and HashStateExTrait are defined as follows:

/// A trait for hash state accumulators.
trait HashStateTrait<S> {
    fn update(self: S, value: felt252) -> S;
    fn finalize(self: S) -> felt252;
}

/// Extension trait for hash state accumulators.
trait HashStateExTrait<S, T> {
    /// Updates the hash state with the given value.
    fn update_with(self: S, value: T) -> S;
}

/// A trait for values that can be hashed.
trait Hash<T, S, +HashStateTrait<S>> {
    /// Updates the hash state with the given value.
    fn update_state(state: S, value: T) -> S;
}

To use hashes in your code, you must first import the relevant traits and functions. In the following example, we will demonstrate how to hash a struct using both the Pedersen and Poseidon hash functions.

The first step is to initialize the hash with either PoseidonTrait::new() -> HashState or PedersenTrait::new(base: felt252) -> HashState depending on which hash function we want to work with. Then the hash state can be updated with the update(self: HashState, value: felt252) -> HashState or update_with(self: S, value: T) -> S functions as many times as required. Then the function finalize(self: HashState) -> felt252 is called on the hash state and it returns the value of the hash as a felt252.

use core::poseidon::PoseidonTrait;
use core::hash::{HashStateTrait, HashStateExTrait};

#[derive(Drop, Hash)]
struct StructForHash {
    first: felt252,
    second: felt252,
    third: (u32, u32),
    last: bool,
}

fn main() -> felt252 {
    let struct_to_hash = StructForHash { first: 0, second: 1, third: (1, 2), last: false };

    let hash = PoseidonTrait::new().update_with(struct_to_hash).finalize();
    hash
}

Pedersen is different from Poseidon, as it starts with a base state. This base state must be of felt252 type, which forces us to either hash the struct with an arbitrary base state using the update_with method, or serialize the struct into an array to loop through all of its fields and hash its elements together.

Here is a short example of Pedersen hashing:

use core::pedersen::PedersenTrait;
use core::hash::{HashStateTrait, HashStateExTrait};

#[derive(Drop, Hash, Serde, Copy)]
struct StructForHash {
    first: felt252,
    second: felt252,
    third: (u32, u32),
    last: bool,
}

fn main() -> (felt252, felt252) {
    let struct_to_hash = StructForHash { first: 0, second: 1, third: (1, 2), last: false };

    // hash1 is the result of hashing a struct with a base state of 0
    let hash1 = PedersenTrait::new(0).update_with(struct_to_hash).finalize();

    let mut serialized_struct: Array<felt252> = ArrayTrait::new();
    Serde::serialize(@struct_to_hash, ref serialized_struct);
    let first_element = serialized_struct.pop_front().unwrap();
    let mut state = PedersenTrait::new(first_element);

    while let Option::Some(value) = serialized_struct.pop_front() {
        state = state.update(value);
    };

    // hash2 is the result of hashing only the fields of the struct
    let hash2 = state.finalize();

    (hash1, hash2)
}

Advanced Hashing: Hashing Arrays with Poseidon

Let us look at an example of hashing a struct that contains a Span<felt252>. To hash a Span<felt252> or a struct that contains a Span<felt252> you can use the built-in function poseidon_hash_span(mut span: Span<felt252>) -> felt252. Similarly, you can hash Array<felt252> by calling poseidon_hash_span on its span.

First, let us import the following traits and function:

use core::poseidon::PoseidonTrait;
use core::poseidon::poseidon_hash_span;
use core::hash::{HashStateTrait, HashStateExTrait};

Now we define the struct. As you might have noticed, we didn't derive the Hash trait. If you attempt to derive the Hash trait for this struct, it will result in an error because the structure contains a field that is not hashable.

#[derive(Drop)]
struct StructForHashArray {
    first: felt252,
    second: felt252,
    third: Array<felt252>,
}

In this example, we initialized a HashState (hash), updated it and then called the function finalize() on the HashState to get the computed hash hash_felt252. We used poseidon_hash_span on the Span of the Array<felt252> to compute its hash.

use core::poseidon::PoseidonTrait;
use core::poseidon::poseidon_hash_span;
use core::hash::{HashStateTrait, HashStateExTrait};

#[derive(Drop)]
struct StructForHashArray {
    first: felt252,
    second: felt252,
    third: Array<felt252>,
}

fn main() {
    let struct_to_hash = StructForHashArray { first: 0, second: 1, third: array![1, 2, 3, 4, 5] };

    let mut hash = PoseidonTrait::new().update(struct_to_hash.first).update(struct_to_hash.second);
    let hash_felt252 = hash.update(poseidon_hash_span(struct_to_hash.third.span())).finalize();
}

Macros

The Cairo language has some plugins that allow developers to simplify their code. They are called inline_macros and are a way of writing code that generates other code.

consteval_int! Macro

In some situations, a developer might need to declare a constant that is the result of a computation of integers. To compute a constant expression and use its result at compile time, it is required to use the consteval_int! macro.

Here is an example of consteval_int!:

const a: felt252 = consteval_int!(2 * 2 * 2);

This will be interpreted as const a: felt252 = 8; by the compiler.

selector! Macro

selector!("function_name") macro generates the entry point selector for the given function name.

print! and println! Macros

Please refer to the Printing page.

array! Macro

Please refer to the Arrays page.

panic! Macro

See Unrecoverable Errors with panic page.

assert! and assert_xx! Macros

See How to Write Tests page.

format! Macro

See Printing page.

write! and writeln! Macros

See Printing page.

get_dep_component!, get_dep_component_mut and component! Macros

Please refer to the Composability and Components chapter.

Procedural Macros

Cairo procedural macros are Rust functions that takes Cairo code as input and returns a modified Cairo code as output, enabling developers to extend Cairo's syntax and create reusable code patterns. In the previous chapter, we discussed some Cairo built-in macros like println!, format!, etc. In this chapter, we will explore how to create and use custom procedural macros in Cairo.

Types of Procedural Macros

There are three types of procedural macros in Cairo:

• Expression Macros (macro!()): These macros are used like function calls and can generate code based on their arguments.

• Attribute Macros (#[macro]): These macros can be attached to items like functions or structs to modify their behavior or implementation.

• Derive Macros (#[derive(Macro)]): These macros automatically implement traits for structs or enums.

Creating a Procedural Macro

Before creating or using procedural macros, we need to ensure that the necessary tools are installed:

• Rust Toolchain: Cairo procedural macros are implemented in Rust, so we will need the Rust toolchain setup on our machine.
• To set up Rust, visit rustup and follow the installation instructions for your operating system.

Since procedural macros are in fact Rust functions, we will need to add a Cargo.toml file to the root directory ( same level as the Scarb.toml file ). In the Cargo.toml file, we need to add a crate-type = ["cdylib"] on the [lib] target, and also add the cairo-lang-macro crate as a dependency.

It is essential that both the Scarb.toml and Cargo.toml have the same package name, or there will be an error when trying to use the macro.

Below is an example of the Scarb.toml and Cargo.toml files:

# Scarb.toml
[package]
name = "no_listing_15_pow_macro"
version = "0.1.0"
edition = "2024_07"

# See more keys and their definitions at https://docs.swmansion.com/scarb/docs/reference/manifest.html
[cairo-plugin]

[dependencies]

[dev-dependencies]
cairo_test = "2.7.0"

Listing 11-5: Example Scarb.toml file needed for building a procedural macro.

# Cargo.toml
[package]
name = "no_listing_15_pow_macro"
version = "0.1.0"
edition = "2021"

[lib]
crate-type = ["cdylib"]

[dependencies]
bigdecimal = "0.4.5"
cairo-lang-macro = "0.1"
cairo-lang-parser = "2.7.0"
cairo-lang-syntax = "2.7.0"

[workspace]

Listing 11-6: Example Cargo.toml file needed for building a procedural macro.

Also notice that you can also add other rust dependencies in your Cargo.toml file. In the example above, we added the bigdecimal, cairo-lang-parser and cairo-lang-syntax crates as a dependencies.

Listing 11-7 shows the rust code for creating an inline macro in Rust:

use bigdecimal::{num_traits::pow, BigDecimal};
use cairo_lang_macro::{inline_macro, Diagnostic, ProcMacroResult, TokenStream};
use cairo_lang_parser::utils::SimpleParserDatabase;
use cairo_lang_syntax::node::kind::SyntaxKind::Arg;

/// Compile-time power function.
///
/// Takes two arguments, `x, y`, calculates the value of `x` raised to the power of `y`.
///
/// ```
/// const MEGABYTE: u64 = pow!(2, 20);
/// assert_eq!(MEGABYTE, 1048576);
/// ```
#[inline_macro]
pub fn pow(token_stream: TokenStream) -> ProcMacroResult {
    let db = SimpleParserDatabase::default();
    let (parsed, _diag) = db.parse_virtual_with_diagnostics(token_stream);

    let macro_args: Vec<String> = parsed
        .descendants(&db)
        .filter_map(|node| {
            if let Arg = node.kind(&db) {
                Some(node.get_text(&db))
            } else {
                None
            }
        })
        .collect();

    if macro_args.len() != 2 {
        return ProcMacroResult::new(TokenStream::empty())
            .with_diagnostics(Diagnostic::error("Invalid number of arguments").into());
    }

    let base: BigDecimal = match macro_args[0].parse() {
        Ok(val) => val,
        Err(_) => {
            return ProcMacroResult::new(TokenStream::empty())
                .with_diagnostics(Diagnostic::error("Invalid base value").into());
        }
    };

    let exp: usize = match macro_args[1].parse() {
        Ok(val) => val,
        Err(_) => {
            return ProcMacroResult::new(TokenStream::empty())
                .with_diagnostics(Diagnostic::error("Invalid exponent value").into());
        }
    };

    let result: BigDecimal = pow(base, exp);

    ProcMacroResult::new(TokenStream::new(result.to_string()))
}

Listing 11-7: Code for creating inline pow procedural macro

The essential dependency for building a cairo macro cairo_lang_macro is imported here with inline_macro, Diagnostic, ProcMacroResult, TokenStream. The inline_macro is used for implementing an expression macro, ProcMacroResult is used for the function return, TokenStream as the input, and the Diagnostic is used for error handling. We also use the cairo-lang-parser crate to parse the input code. Then the pow function is defined utilizing the imports to create a macro that calculate the pow based on the TokenStream input.

How to Use Existing Procedural Macros

Note: While you need Rust installed to use procedural macros, you don't need to know Rust programming to use existing macros in your Cairo project.

Incorporating an Existing Procedural Macro Into Your Project

Similar to how you add a library dependency in your Cairo project, you can also add a procedural macro as a dependency in your Scarb.toml file.

    #[test]
    fn test_pow_macro() {
        assert_eq!(super::TWO_TEN, 144);
        assert_eq!(pow!(10, 2), 100);
        assert_eq!(pow!(20, 30), 1073741824000000000000000000000000000000_felt252);
        assert_eq!(
            pow!(2, 255),
            57896044618658097711785492504343953926634992332820282019728792003956564819968_u256
        );
    }

Listing 11-8: Using pow procedural macro

You'd notice a pow! macro, which is not a built-in Cairo macro being used in this example above. It is a custom procedural macro that calculates the power of a number as defined in the example above on creating a procedural macro.

#[derive(Add, AddAssign, Sub, SubAssign, Mul, MulAssign, Div, DivAssign, Debug, Drop, PartialEq)]
pub struct B {
    pub a: u8,
    pub b: u16
}

Listing 11-9: Using derive procedural macro

The example above shows using a derive macro on a struct B, which grants the custom struct the ability to perform addition, subtraction, multiplication, and division operations on the struct.

    let b1 = B { a: 1, b: 2 };
    let b2 = B { a: 3, b: 4 };
    let b3 = b1 + b2;

Summary

Procedural macros offer a powerful way to extend Cairo's capabilities by leveraging Rust functions to generate new Cairo code. They allow for code generation, custom syntax, and automated implementations, making them a valuable tool for Cairo developers. While they require some setup and careful consideration of performance impacts, the flexibility they provide can significantly enhance your Cairo development experience.

Inlining in Cairo

Inlining is a common code optimization technique supported by most compilers. It involves replacing a function call at the call site with the actual code of the called function, eliminating the overhead associated with the function call itself. This can improve performance by reducing the number of instructions executed, but may increase the total size of the program. When you're thinking about whether to inline a function, take into account things like how big it is, what parameters it has, how often it gets called, and how it might affect the size of your compiled code.

The inline Attribute

In Cairo, the inline attribute suggests whether or not the Sierra code corresponding to the attributed function should be directly injected in the caller function's context, rather than using a function_call libfunc to execute that code.

There are three variants of the inline attribute that one can use:

• #[inline] suggests performing an inline expansion.
• #[inline(always)] suggests that an inline expansion should always be performed.
• #[inline(never)] suggests that an inline expansion should never be performed.

Note: the inline attribute in every form is a hint, with no requirements on the language to place a copy of the attributed function in the caller. This means that the attribute may be ignored by the compiler. In practice, #[inline(always)] will cause inlining in all but the most exceptional cases.

Many of the Cairo corelib functions are inlined. User-defined functions may also be annotated with the inline attribute. Annoting functions with the #[inline(always)] attribute reduces the total number of steps required when calling these attributed functions. Indeed, injecting the Sierra code at the caller site avoids the step-cost involved in calling functions and obtaining their arguments.

However, inlining can also lead to increased code size. Whenever a function is inlined, the call site contains a copy of the function's Sierra code, potentially leading to duplication of code across the compiled code.

Therefore, inlining should be applied with caution. Using #[inline] or #[inline(always)] indiscriminately will lead to increased compile time. It is particularly useful to inline small functions, ideally with many arguments. This is because inlining large functions will increase the code length of the program, and handling many arguments will increase the number of steps to execute these functions.

The more frequently a function is called, the more beneficial inlining becomes in terms of performance. By doing so, the number of steps for the execution will be lower, while the code length will not grow that much or might even decrease in terms of total number of instructions.

Inlining is often a tradeoff between number of steps and code length. Use the inline attribute cautiously where it is appropriate.

Inlining decision process

The Cairo compiler follows the inline attribute but for functions without explicit inline directives, it will use a heuristic approach. The decision to inline or not a function will be made depending on the complexity of the attributed function and mostly rely on the threshold DEFAULT_INLINE_SMALL_FUNCTIONS_THRESHOLD.

The compiler calculates a function's "weight" using the ApproxCasmInlineWeight struct, which estimates the number of Cairo Assembly (CASM) statements the function will generate. This weight calculation provides a more nuanced view of the function's complexity than a simple statement count. If a function's weight falls below the threshold, it will be inlined.

In addition to the weight-based approach, the compiler also considers the raw statement count. Functions with fewer statements than the threshold are typically inlined, promoting the optimization of small, frequently called functions.

The inlining process also accounts for special cases. Very simple functions, such as those that only call another function or return a constant, are always inlined regardless of other factors. Conversely, functions with complex control flow structures like Match or those ending with a Panic are generally not inlined.

Inlining Example

Let's introduce a short example to illustrate the mechanisms of inlining in Cairo. Listing 11-10 shows a basic program allowing comparison between inlined and non-inlined functions.

fn main() -> felt252 {
    inlined() + not_inlined()
}

#[inline(always)]
fn inlined() -> felt252 {
    1
}

#[inline(never)]
fn not_inlined() -> felt252 {
    2
}

Listing 11-10: A small Cairo program that adds the return value of 2 functions, with one of them being inlined

Let's take a look at the corresponding Sierra code to see how inlining works under the hood:

// type declarations
type felt252 = felt252 [storable: true, drop: true, dup: true, zero_sized: false]

// libfunc declarations
libfunc function_call<user@main::main::not_inlined> = function_call<user@main::main::not_inlined>
libfunc felt252_const<1> = felt252_const<1>
libfunc store_temp<felt252> = store_temp<felt252>
libfunc felt252_add = felt252_add
libfunc felt252_const<2> = felt252_const<2>

// statements
00 function_call<user@main::main::not_inlined>() -> ([0])
01 felt252_const<1>() -> ([1])
02 store_temp<felt252>([1]) -> ([1])
03 felt252_add([1], [0]) -> ([2])
04 store_temp<felt252>([2]) -> ([2])
05 return([2])
06 felt252_const<1>() -> ([0])
07 store_temp<felt252>([0]) -> ([0])
08 return([0])
09 felt252_const<2>() -> ([0])
10 store_temp<felt252>([0]) -> ([0])
11 return([0])

// funcs
main::main::main@0() -> (felt252)
main::main::inlined@6() -> (felt252)
main::main::not_inlined@9() -> (felt252)

The Sierra file is structured in three parts:

• Type and libfunc declarations.
• Statements that constitute the program.
• Declaration of the functions of the program.

The Sierra code statements always match the order of function declarations in the Cairo program. Indeed, the declaration of the functions of the program tells us that:

• main function starts at line 0, and returns a felt252 on line 5.
• inlined function starts at line 6, and returns a felt252 on line 8.
• not_inlined function starts at line 9, and returns a felt252 on line 11.

All statements corresponding to the main function are located between lines 0 and 5:

00 function_call<user@main::main::not_inlined>() -> ([0])
01 felt252_const<1>() -> ([1])
02 store_temp<felt252>([1]) -> ([1])
03 felt252_add([1], [0]) -> ([2])
04 store_temp<felt252>([2]) -> ([2])
05 return([2])

The function_call libfunc is called on line 0 to execute the not_inlined function. This will execute the code from lines 9 to 10 and store the return value in the variable with id 0.

09	felt252_const<2>() -> ([0])
10	store_temp<felt252>([0]) -> ([0])

This code uses a single data type, felt252. It uses two library functions - felt252_const<2>, which returns the constant felt252 2, and store_temp<felt252>, which pushes a constant value to memory. The first line calls the felt252_const<2> libfunc to create a variable with id 0. Then, the second line pushes this variable to memory for later use.

After that, Sierra statements from line 1 to 2 are the actual body of the inlined function:

06	felt252_const<1>() -> ([0])
07	store_temp<felt252>([0]) -> ([0])

The only difference is that the inlined code will store the felt252_const value in a variable with id 1, because [0] refers to a variable previously assigned:

01	felt252_const<1>() -> ([1])
02	store_temp<felt252>([1]) -> ([1])

Note: in both cases (inlined or not), the return instruction of the function being called is not executed, as this would lead to prematurely end the execution of the main function. Instead, return values of inlined and not_inlined will be added and the result will be returned.

Lines 3 to 5 contain the Sierra statements that will add the values contained in variables with ids 0 and 1, store the result in memory and return it:

03	felt252_add([1], [0]) -> ([2])
04	store_temp<felt252>([2]) -> ([2])
05	return([2])

Now, let's take a look at the Casm code corresponding to this program to really understand the benefits of inlining.

Casm Code Explanations

Here is the Casm code for our previous program example:

1	call rel 3
2	ret
3	call rel 9
4	[ap + 0] = 1, ap++
5	[ap + 0] = [ap + -1] + [ap + -2], ap++
6	ret
7	[ap + 0] = 1, ap++
8	ret
9	[ap + 0] = 2, ap++
10	ret
11	ret

Don't hesitate to use cairovm.codes playground to follow along and see all the execution trace.

Each instruction and each argument for any instruction increment the Program Counter (known as PC) by 1. This means that ret on line 2 is actually the instruction at PC = 3, as the argument 3 corresponds to PC = 2.

The call and ret instructions allow implementation of a function stack:

• call instruction acts like a jump instruction, updating the PC to a given value, whether relatively to the current value using rel or absolutely using abs.
• ret instruction jumps back right after the call instruction and continues the execution of the code.

We can now decompose how these instructions are executed to understand what this code does:

• call rel 3: this instruction increments the PC by 3 and executes the instruction at this location, which is call rel 9 at PC = 4.
• call rel 9 increments the PC by 9 and executes the instruction at PC = 13, which is actually line 9.
• [ap + 0] = 2, ap++: ap stands for Allocation Pointer, which points to the first memory cell that has not been used by the program so far. This means we store the value 2 in the next free memory cell indicated by the current value of ap, after which we increment ap by 1. Then, we go to the next line which is ret.
• ret: jumps back to the line after call rel 9, so we go to line 4.
• [ap + 0] = 1, ap++ : we store the value 1 in [ap] and we apply ap++ so that [ap - 1] = 1. This means we now have [ap-1] = 1, [ap-2] = 2 and we go to the next line.
• [ap + 0] = [ap + -1] + [ap + -2], ap++: we sum the values 1 and 2 and store the result in [ap], and we apply ap++ so the result is [ap-1] = 3, [ap-2] = 1, [ap-3]=2.
• ret: jumps back to the line after call rel 3, so we go to line 2.
• ret: last instruction executed as there is no more call instruction where to jump right after. This is the actual return instruction of the Cairo main function.

To summarize:

• call rel 3 corresponds to the main function, which is obviously not inlined.
• call rel 9 triggers the call to the not_inlined function, which returns 2 and stores it at the final location [ap-3].
• The line 4 is the inlined code of the inlined function, which returns 1 and stores it at the final location [ap-2]. We clearly see that there is no call instruction in this case, because the body of the function is inserted and directly executed.
• After that, the sum is computed and we ultimately go back to the line 2 which contains the final ret instruction that returns the sum, corresponding to the return value of the main function.

It is interesting to note that in both Sierra code and Casm code, the not_inlined function will be called and executed before the body of the inlined function, even though the Cairo program executes inlined() + not_inlined().

The Casm code of our program clearly shows that there is a function call for the not_inlined function, while the inlined function is correctly inlined.

Additional Optimizations

Let's study another program that shows other benefits that inlining may sometimes provide. Listing 11-11 shows a Cairo program that calls 2 functions and doesn't return anything:

fn main() {
    inlined();
    not_inlined();
}

#[inline(always)]
fn inlined() -> felt252 {
    'inlined'
}

#[inline(never)]
fn not_inlined() -> felt252 {
    'not inlined'
}

Listing 11-11: A small Cairo program that calls inlined and not_inlined and doesn't return any value.

Here is the corresponding Sierra code:

// type declarations
type felt252 = felt252 [storable: true, drop: true, dup: true, zero_sized: false]
type Unit = Struct<ut@Tuple> [storable: true, drop: true, dup: true, zero_sized: true]

// libfunc declarations
libfunc function_call<user@main::main::not_inlined> = function_call<user@main::main::not_inlined>
libfunc drop<felt252> = drop<felt252>
libfunc struct_construct<Unit> = struct_construct<Unit>
libfunc felt252_const<29676284458984804> = felt252_const<29676284458984804>
libfunc store_temp<felt252> = store_temp<felt252>
libfunc felt252_const<133508164995039583817065828> = felt252_const<133508164995039583817065828>

// statements
00 function_call<user@main::main::not_inlined>() -> ([0])
01 drop<felt252>([0]) -> ()
02 struct_construct<Unit>() -> ([1])
03 return([1])
04 felt252_const<29676284458984804>() -> ([0])
05 store_temp<felt252>([0]) -> ([0])
06 return([0])
07 felt252_const<133508164995039583817065828>() -> ([0])
08 store_temp<felt252>([0]) -> ([0])
09 return([0])

// funcs
main::main::main@0() -> (Unit)
main::main::inlined@4() -> (felt252)
main::main::not_inlined@7() -> (felt252)

In this specific case, we can observe that the compiler has applied additional optimizations to the main function of our code : the code of the inlined function, which is annotated with the #[inline(always)] attribute, is actually not copied in the main function. Instead, the main function starts with the function_call libfunc to call the not_inlined function, entirely omitting the code of the inlined function.

Because inlined return value is never used, the compiler optimizes the main function by skipping the inlined function code. This will actually reduce the code length while reducing the number of steps required to execute main.

In contrast, line 0 uses the function_call libfunc to execute the not_inlined function normally. This means that all the code from lines 7 to 8 will be executed:

07 felt252_const<133508164995039583817065828>() -> ([0])
08 store_temp<felt252>([0]) -> ([0])

This value stored in the variable with id 0 is then dropped on line 1, as it is not used in the main function:

01 drop<felt252>([0]) -> ()

Finally, as the main function doesn't return any value, a variable of unit type () is created and returned:

02 struct_construct<Unit>() -> ([1])
03 return([1])

Summary

Inlining is a compiler optimization technique that can be very useful in various situations. Inlining a function allows to get rid of the overhead of calling a function with the function_call libfunc by injecting the Sierra code directly in the caller function's context, while potentially optimizing the Sierra code executed to reduce the number of steps. If used effectively, inlining can even reduce code length as shown in the previous example.

Nevertheless, applying the inline attribute to a function with a lot of code and few parameters might result in an increased code size, especially if the inlined function is used many times in the codebase. Use inlining only where it makes sense, and be aware that the compiler handles inlining by default. Therefore, manually applying inlining is not recommended in most situations, but can help improve and fine-tune your code's behavior.

Printing

When writing a program, it is quite common to print some data to the console, either for the normal process of the program or for debugging purpose. In this chapter, we describe the options you have to print simple and complex data types.

Printing Standard Data Types

Cairo provides two macros to print standard data types:

• println! which prints on a new line
• print! with inline printing

Both take a ByteArray string as first parameter (see Data Types), which can be a simple string to print a message or a string with placeholders to format the way values are printed.

There are two ways to use these placeholders and both can be mixed:

• empty curly brackets {} are replaced by values given as parameters to the print! macro, in the same order.
• curly brackets with variable names are directly replaced by the variable value.

Here are some examples:

fn main() {
    let a = 10;
    let b = 20;
    let c = 30;

    println!("Hello world!");
    println!("{} {} {}", a, b, c); // 10 20 30
    println!("{c} {a} {}", b); // 30 10 20
}

print! and println! macros use the Display trait under the hood, and are therefore used to print the value of types that implement it. This is the case for basic data types, but not for more complex ones. If you try to print complex data type values with these macros, e.g. for debugging purposes, you will get an error. In that case, you can either manually implement the Display trait for your type or use the Debug trait (see below).

Formatting

Cairo also provides a useful macro to handle string formatting: format!. This macro works like println!, but instead of printing the output to the screen, it returns a ByteArray with the contents. In the following example, we perform string concatenation using either the + operator or the format! macro. The version of the code using format! is much easier to read, and the code generated by the format! macro uses snapshots, so that this call doesn’t take ownership of any of its parameters.

fn main() {
    let s1: ByteArray = "tic";
    let s2: ByteArray = "tac";
    let s3: ByteArray = "toe";
    let s = s1 + "-" + s2 + "-" + s3;
    // using + operator consumes the strings, so they can't be used again!

    let s1: ByteArray = "tic";
    let s2: ByteArray = "tac";
    let s3: ByteArray = "toe";
    let s = format!("{s1}-{s2}-{s3}"); // s1, s2, s3 are not consumed by format!
    // or
    let s = format!("{}-{}-{}", s1, s2, s3);

    println!("{}", s);
}

Printing Custom Data Types

As previously explained, if you try to print the value of a custom data type with print! or println! macros, you'll get an error telling you that the Display trait is not implemented for your custom type:

error: Trait has no implementation in context: core::fmt::Display::<package_name::struct_name>

The println! macro can do many kinds of formatting, and by default, the curly brackets tell println! to use formatting known as Display - output intended for direct end user consumption. The primitive types we’ve seen so far implement Display by default because there’s only one way you’d want to show a 1 or any other primitive type to a user. But with structs, the way println! should format the output is less clear because there are more display possibilities: Do we want commas or not? Do we want to print the curly brackets? Should all the fields be shown? Due to this ambiguity, Cairo doesn’t try to guess what we want, and structs don’t have a provided implementation of Display to use with println! and the {} placeholder.

Here is the Display trait to implement:

trait Display<T> {
    fn fmt(self: @T, ref f: Formatter) -> Result<(), Error>;
}

The second parameter f is of type Formatter, which is just a struct containing a ByteArray, representing the pending result of formatting:

#[derive(Default, Drop)]
pub struct Formatter {
    /// The pending result of formatting.
    pub buffer: ByteArray,
}

Knowing this, here is an example of how to implement the Display trait for a custom Point struct:

use core::fmt::{Display, Formatter, Error};

#[derive(Copy, Drop)]
struct Point {
    x: u8,
    y: u8
}

impl PointDisplay of Display<Point> {
    fn fmt(self: @Point, ref f: Formatter) -> Result<(), Error> {
        let str: ByteArray = format!("Point ({}, {})", *self.x, *self.y);
        f.buffer.append(@str);
        Result::Ok(())
    }
}

fn main() {
    let p = Point { x: 1, y: 3 };
    println!("{}", p); // Point: (1, 3)
}

Cairo also provides the write! and writeln! macros to write formatted strings in a formatter. Here is a short example using write! macro to concatenate multiple strings on the same line and then print the result:

use core::fmt::Formatter;

fn main() {
    let mut formatter: Formatter = Default::default();
    let a = 10;
    let b = 20;
    write!(formatter, "hello");
    write!(formatter, "world");
    write!(formatter, " {a} {b}");

    println!("{}", formatter.buffer); // helloworld 10 20
}

It is also possible to implement the Display trait for the Point struct using these macros, as shown here:

use core::fmt::{Display, Formatter, Error};

#[derive(Copy, Drop)]
struct Point {
    x: u8,
    y: u8
}

impl PointDisplay of Display<Point> {
    fn fmt(self: @Point, ref f: Formatter) -> Result<(), Error> {
        let x = *self.x;
        let y = *self.y;

        writeln!(f, "Point ({x}, {y})")
    }
}

fn main() {
    let p = Point { x: 1, y: 3 };
    println!("{}", p); // Point: (1, 3)
}

Printing complex data types this way might not be ideal as it requires additional steps to use the print! and println! macros. If you need to print complex data types, especially when debugging, use the Debug trait described below instead.

Print Debug Traces

Cairo provides the Debug trait, which can be derived to print the value of variables when debugging. Simply add :? within the curly brackets {} placeholders in a print! or println! macro string.

This trait is very useful and is implemented by default for basic data types. It can also be simply derived for complex data types using the #[derive(Debug)] attribute, as long as all types they contain implement it. This eliminates the need to manually implement extra code to print complex data types.

Note that assert_xx! macros used in tests require the provided values to implement the Debug trait, as they also print the result in case of assertion failure.

For more details about the Debug trait and its usage for printing values when debugging, please refer to the Derivable Traits appendix.

Arithmetic Circuits

Arithmetic circuits are mathematical models used to represent polynomial computations. They are defined over a field (typically a finite field \(F_p\) where \(p\) is prime) and consist of:

• Input signals (values in the range \([0, p-1]\))
• Arithmetic operations (addition and multiplication gates)
• Output signals

Cairo supports emulated arithmetic circuits with modulo up to 384 bits.

This is especially useful for:

• Implementing verification for other proof systems
• Implementing cryptographic primitives
• Creating more low-level programs, with potential reduced overhead compared to standard Cairo constructs

Implementing Arithmetic Circuits in Cairo

Cairo's circuit constructs are available in the core::circuit module of the corelib.

Arithmetic circuits consist of:

• Addition modulo \(p\): AddMod builtin
• Multiplication modulo \(p\): MulMod builtin

Because of the modulo properties, we can build four basic arithmetic gates:

• Addition: AddModGate
• Subtraction: SubModGate
• Multiplication: MulModGate
• Inverse: InvModGate

Let's create a circuit that computes \(a \cdot (a + b)\) over the BN254 prime field.

We start from the empty struct CircuitElement<T>.

The inputs of our circuit are defined as CircuitInput:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

We can combine circuit inputs and gates: CircuitElement<a> and CircuitElement<b> combined with an addition gate gives CircuitElement<AddModGate<a, b>>.

We can use circuit_add, circuit_sub, circuit_mul and circuit_inverse to directly combine circuit elements. For \(a * (a + b)\), the description of our circuit is:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

Note that a, b and add are intermediate circuit elements and not specifically inputs or gates, which is why we need the distinction between the empty struct CircuitElement<T> and the circuit description specified by the type T.

The outputs of the circuits are defined as a tuple of circuit elements. It's possible to add any intermediate gates of our circuit, but we must add all gates with degree 0 (gates where the output signal is not used as input of any other gate). In our case, we will only add the last gate mul:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

We now have a complete description of our circuit and its outputs. We now need to assign a value to each input. As circuits are defined with 384-bit modulus, a single u384 value can be represented as a fixed array of four u96. We can initialize \(a\) and \(b\) to respectively \(10\) and \(20\):

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

As the number of inputs can vary, Cairo use an accumulator and the new_inputs and next functions return a variant of the AddInputResult enum.

pub enum AddInputResult<C> {
    /// All inputs have been filled.
    Done: CircuitData<C>,
    /// More inputs are needed to fill the circuit instance's data.
    More: CircuitInputAccumulator<C>,
}

We have to assign a value to every input, by calling next on each CircuitInputAccumulator variant. After the inputs initialization, by calling the done function we get the complete circuit CircuitData<C>, where C is a long type that encodes the entire circuit instance.

We then need to define what modulus our circuit is using (up to 384-bit modulus), by defining a CircuitModulus. We want to use BN254 prime field modulus:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

The last part is the evaluation of the circuit, i.e. the actual process of passing the input signals correctly through each gate described by our circuit and getting the values of each output gate. We can evaluate and get the results for a given modulus as follows:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

To retrieve the value of a specific output, we can use the get_output function on our results with the CircuitElement instance of the output gate we want. We can also retrieve any intermediate gate value as well.

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

To recap, we did the following steps:

• Define Circuit Inputs
• Describe the circuit
• Specify the outputs
• Assign values to the inputs
• Define the modulus
• Evaluate the circuit
• Get the output values

And the full code is:

use core::circuit::{
    CircuitElement, CircuitInput, circuit_add, circuit_mul, EvalCircuitTrait, CircuitOutputsTrait,
    CircuitModulus, AddInputResultTrait, CircuitInputs, u384
};

// Circuit: a * (a + b)
// witness: a = 10, b = 20
// expected output: 10 * (10 + 20) = 300
fn eval_circuit() -> (u384, u384) {
    let a = CircuitElement::<CircuitInput<0>> {};
    let b = CircuitElement::<CircuitInput<1>> {};

    let add = circuit_add(a, b);
    let mul = circuit_mul(a, add);

    let output = (mul,);

    let mut inputs = output.new_inputs();
    inputs = inputs.next([10, 0, 0, 0]);
    inputs = inputs.next([20, 0, 0, 0]);

    let instance = inputs.done();

    let bn254_modulus = TryInto::<
        _, CircuitModulus
    >::try_into([0x6871ca8d3c208c16d87cfd47, 0xb85045b68181585d97816a91, 0x30644e72e131a029, 0x0])
        .unwrap();

    let res = instance.eval(bn254_modulus).unwrap();

    let add_output = res.get_output(add);
    let circuit_output = res.get_output(mul);

    assert(add_output == u384 { limb0: 30, limb1: 0, limb2: 0, limb3: 0 }, 'add_output');
    assert(circuit_output == u384 { limb0: 300, limb1: 0, limb2: 0, limb3: 0 }, 'circuit_output');

    (add_output, circuit_output)
}

fn main() {
    eval_circuit();
}

Arithmetic Circuits in Zero-Knowledge Proof Systems

In zero-knowledge proof systems, a prover creates a proof of computational statements, which a verifier can check without performing the full computation. However, these statements must first be converted into a suitable representation for the proof system.

zk-SNARKs Approach

Some proof systems, like zk-SNARKs, use arithmetic circuits over a finite field \(F_p\). These circuits include constraints at specific gates, represented as equations:

\[ (a_1 \cdot s_1 + ... + a_n \cdot s_n) \cdot (b_1 \cdot s_1 + ... + b_n \cdot s_n) + (c_1 \cdot s_1 + ... + c_n \cdot s_n) = 0 \mod p \] Where \(s_1, ..., s_n\) are signals, and \(a_i, b_i, c_i\) are coefficients.

A witness is an assignment of signals that satisfies all constraints in a circuit. zk-SNARK proofs use these properties to prove knowledge of a witness without revealing private input signals, ensuring the prover's honesty while preserving privacy.

Some work has already been done, such as Garaga Groth16 verifier

zk-STARKs Approach

STARKs (which Cairo uses) use an Algebraic Intermediate Representation (AIR) instead of arithmetic circuits. AIR describes computations as a set of polynomial constraints.

By allowing emulated arithmetic circuits, Cairo can be used to implement zk-SNARKs proof verification inside STARK proofs.

Appendix

The following sections contain reference material you may find useful in your Cairo journey.

Appendix A - Keywords

The following list contains keywords that are reserved for current or future use by the Cairo language.

There are three keyword categories:

• strict
• loose
• reserved

There is a fourth category, which are functions from the core library. While their names are not reserved, they are not recommended to be used as names of any items to follow good practices.

Strict keywords

These keywords can only be used in their correct contexts. They cannot be used as names of any items.

• as - Rename import
• break - Exit a loop immediately
• const - Define constant items
• continue - Continue to the next loop iteration
• else - Fallback for if and if let control flow constructs
• enum - Define an enumeration
• extern - Function defined at the compiler level that can be compiled to CASM
• false - Boolean false literal
• fn - Define a function
• if - Branch based on the result of a conditional expression
• impl - Implement inherent or trait functionality
• implicits - Special kind of function parameters that are required to perform certain actions
• let - Bind a variable
• loop - Loop unconditionally
• match - Match a value to patterns
• mod - Define a module
• mut - Denote variable mutability
• nopanic - Functions marked with this notation mean that the function will never panic.
• of - Implement a trait
• pub - Denote public visibility in items, such as struct and struct fields, enums, consts, traits and impl blocks, or modules
• ref - Parameter passed implicitly returned at the end of a function
• return - Return from function
• struct - Define a structure
• trait - Define a trait
• true - Boolean true literal
• type - Define a type alias
• use - Bring symbols into scope
• while - loop conditionally based on the result of an expression