Share this

See a typo? Have a suggestion? Edit this page on Github

While reviewing the second-to-last chapter in the Begin Rust book, a slightly more advanced topic came up. This is a topic that has intrigued me for a while, in particular because it demonstrates some fundamental differences in how Rust and Haskell approach mutability. The topic is too advanced for the book, but I wanted to provide an external source to reference for people who are curious. So here it is!

Let’s build this up bit by bit. Does the following program compile?

fn main() {
    let x = 5;
    x += 1;
    println!("x == {}", x);
}

Answer: no! x is an immutable variable, and therefore you can’t use += 1 on it. Fixing this is easy: add mut:

fn main() {
    let mut x = 5;
    x += 1;
    println!("x == {}", x);
}

But since adding 1 and printing is such an important operation in my application (yes, that’s sarcasm), I’ve decided to extract it to its own function. Tell me, does this code compile?

fn add_and_print(x: i32) {
    x += 1;
    println!("x == {}", x);
}

No, and for the same reason as the first example: x is immutable. Fixing that is easy enough here too:

fn add_and_print(mut x: i32) {
    x += 1;
    println!("x == {}", x);
}

Now we have x as the only parameter to add_and_print. Its type is i32, and it is mutable. OK, cool. Finally we’re in a position to use this function from main. Tell me, will this program compile and run? And as some really great foreshadowing, will the program generate any warnings?

fn main() {
    let mut x = 5;
    add_and_print(x);
}

fn add_and_print(mut x: i32) {
    x += 1;
    println!("x == {}", x);
}

Answer: it compiles, runs, and generates the output x == 6. However, it does have a warning:

warning: variable does not need to be mutable
 --> src/main.rs:2:9
  |
2 |     let mut x = 5;
  |         ----^
  |         |
  |         help: remove this `mut`
  |

Initially, at least for me, this was really surprising. add_and_print needs to receive a mutable i32 as its first parameter. We provide it with a mutable i32. And then the compiler says that the mut in main is unnecessary. What gives?!?

There’s a mistake in my explanation above. The function add_and_print, perhaps very confusingly, does not take a mutable i32 as a parameter. “But it says mut!!!” you argue. That’s true. However, that detail is internal to the function, not part of its type signature. That sounds confusing, so let me explain.

There’s a pattern that appears all over Rust which I call the rule of three. This covers the fact that in many cases, we end up having three “versions” of things:

  • An immutable borrow
  • A mutable borrow
  • A move

That can apply to function parameters, for example:

fn immutable_borrow(x: &String)
fn mutable_borrow(x: &mut String)
fn move(x: String)

Notice that the difference is entirely after the colon. These three things are different types, and the types are what make up the signature of the function.

However, the stuff before the colon, in this case the x, does not affect the signature of the function. What name you choose for the variable that captures the parameter is irrelevant to the signature of the function. Once you pass in a value of the right type to the function, the function gets to decide what to call it.

But this rule applies to more than just the variable name. It applies to the variable mutability as well. You see, mutability in Rust is a feature of the variable, not the value. When I say let mut x = 5, I’m saying “I’m creating a variable called x, it’s pointing a value 5, and I’m allowed to use x to mutate that value.” If I drop the mut, I’m no longer allowed to mutate that value via the variable x.

A gut reaction you might have is, if you can’t mutate the variable, the only thing you can do is read it. This is where my Haskeller spidey-senses lead me to at least. But it’s not true in Rust. There’s one other thing you’re allowed to do: move the value into another scope. add_and_print accepts the value by move, and even if x is an immutable variable, I’m still allowed to move the value that it references.

Once I’ve moved the value, it’s entirely up to add_and_print to decide what to do with it. And it’s allowed to call it mutable, even though the original variable was immutable. That’s because the value itself is passed in, not the variable. And the value is perfectly comfortable being mutated.

So a warning-free version of the program is:

fn main() {
    let x = 5;
    add_and_print(x);
}

fn add_and_print(mut x: i32) {
    x += 1;
    println!("x == {}", x);
}

In fact, moving-into-mutable is something that can happen even without a function call. For example, you can “upgrade” from an immutable variable to a mutable one inside a single function:

fn main() {
    let x = 5;
    let mut y = x;
    y += 1;
    println!("y == {}", y);
}

“But wait,” you complain, “you can’t just ‘upgrade’ an immutable reference to a mutable reference!” Or maybe a dozen other similar complaints about my explanation. What I’m slightly glossing over here is that, when it comes to references, the mutability is baked into the value. That’s because with something like x: &i32, x doesn’t have a number, it has a reference to a number. And references themselves are types that do in fact enforce mutable vs immutable rules. Therefore, you can’t simply upgrade an immutable reference to a mutable one. This code is broken:

fn main() {
    let mut x: i32 = 5;
    let y: &i32 = &x;
    let z: &mut i32 = y; // b0rken!

    *z += 1;

    println!("x == {}", x);
}

So, pulling it back together:

  • You can have values, immutable references to values, and mutable references to values
  • Values are neither immutable or mutable
  • Instead, variables are immutable or mutable
  • When you move a value into a new variable (via a let or a function call), you can change the mutability of the variable
  • The mutability and names of variables in function signatures do not affect the signature of the function
  • The mutability of a reference is built into the type itself, so you can’t “upgrade” an immutable reference to a mutable one

Blog archive