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

换句话说,这里有两个重要的时间点:

  • 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() };
}

如果你试图运行这段代码,你会得到一个编译时错误:

$ 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
}

现在,当A超出作用域时,它的字典将被自动squashed,并且程序将被编译。

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.

下面是一个 clone 方法的实例。

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.

文件名: 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.

文件名: 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.