New upstream version 1.56.0~beta.4+dfsg1

[rustc.git] / src / doc / book / src / ch04-01-what-is-ownership.md

## What Is Ownership?

Rust’s central feature is *ownership*. Although the feature is straightforward
to explain, it has deep implications for the rest of the language.

All programs have to manage the way they use a computer’s memory while running.
Some languages have garbage collection that constantly looks for no longer used
memory as the program runs; in other languages, the programmer must explicitly
allocate and free the memory. Rust uses a third approach: memory is managed
through a system of ownership with a set of rules that the compiler checks at
compile time. None of the ownership features slow down your program while it’s
running.

Because ownership is a new concept for many programmers, it does take some time
to get used to. The good news is that the more experienced you become with Rust
and the rules of the ownership system, the more you’ll be able to naturally
develop code that is safe and efficient. Keep at it!

When you understand ownership, you’ll have a solid foundation for understanding
the features that make Rust unique. In this chapter, you’ll learn ownership by
working through some examples that focus on a very common data structure:
strings.

> ### The Stack and the Heap
>
> In many programming languages, you don’t have to think about the stack and
> the heap very often. But in a systems programming language like Rust, whether
> a value is on the stack or the heap has more of an effect on how the language
> behaves and why you have to make certain decisions. Parts of ownership will
> be described in relation to the stack and the heap later in this chapter, so
> here is a brief explanation in preparation.
>
> Both the stack and the heap are parts of memory that are available to your
> code to use at runtime, but they are structured in different ways. The stack
> stores values in the order it gets them and removes the values in the
> opposite order. This is referred to as *last in, first out*. Think of a stack
> of plates: when you add more plates, you put them on top of the pile, and
> when you need a plate, you take one off the top. Adding or removing plates
> from the middle or bottom wouldn’t work as well! Adding data is called
> *pushing onto the stack*, and removing data is called *popping off the stack*.
>
> All data stored on the stack must have a known, fixed size. Data with an
> unknown size at compile time or a size that might change must be stored on
> the heap instead. The heap is less organized: when you put data on the heap,
> you request a certain amount of space. The memory allocator finds an empty
> spot in the heap that is big enough, marks it as being in use, and returns a
> *pointer*, which is the address of that location. This process is called
> *allocating on the heap* and is sometimes abbreviated as just *allocating*.
> Pushing values onto the stack is not considered allocating. Because the
> pointer is a known, fixed size, you can store the pointer on the stack, but
> when you want the actual data, you must follow the pointer.
>
> Think of being seated at a restaurant. When you enter, you state the number of
> people in your group, and the staff finds an empty table that fits everyone
> and leads you there. If someone in your group comes late, they can ask where
> you’ve been seated to find you.
>
> Pushing to the stack is faster than allocating on the heap because the
> allocator never has to search for a place to store new data; that
> location is always at the top of the stack. Comparatively, allocating space
> on the heap requires more work, because the allocator must first find
> a big enough space to hold the data and then perform bookkeeping to prepare
> for the next allocation.
>
> Accessing data in the heap is slower than accessing data on the stack because
> you have to follow a pointer to get there. Contemporary processors are faster
> if they jump around less in memory. Continuing the analogy, consider a server
> at a restaurant taking orders from many tables. It’s most efficient to get
> all the orders at one table before moving on to the next table. Taking an
> order from table A, then an order from table B, then one from A again, and
> then one from B again would be a much slower process. By the same token, a
> processor can do its job better if it works on data that’s close to other
> data (as it is on the stack) rather than farther away (as it can be on the
> heap). Allocating a large amount of space on the heap can also take time.
>
> When your code calls a function, the values passed into the function
> (including, potentially, pointers to data on the heap) and the function’s
> local variables get pushed onto the stack. When the function is over, those
> values get popped off the stack.
>
> Keeping track of what parts of code are using what data on the heap,
> minimizing the amount of duplicate data on the heap, and cleaning up unused
> data on the heap so you don’t run out of space are all problems that ownership
> addresses. Once you understand ownership, you won’t need to think about the
> stack and the heap very often, but knowing that managing heap data is why
> ownership exists can help explain why it works the way it does.

### Ownership Rules

First, let’s take a look at the ownership rules. Keep these rules in mind as we
work through the examples that illustrate them:

* Each value in Rust has a variable that’s called its *owner*.
* There can only be one owner at a time.
* When the owner goes out of scope, the value will be dropped.

### Variable Scope

We’ve walked through an example of a Rust program already in Chapter 2. Now
that we’re past basic syntax, we won’t include all the `fn main() {` code in
examples, so if you’re following along, you’ll have 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.

As a first example of ownership, we’ll look at the *scope* of some variables. A
scope is the range within a program for which an item is valid. Let’s say we
have a variable that looks like this:

```rust
let s = "hello";
```

The variable `s` refers to a string literal, where the value of the string is
hardcoded into the text of our program. The variable is valid from the point at
which it’s declared until the end of the current *scope*. Listing 4-1 has
comments annotating where the variable `s` is valid.

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/listing-04-01/src/main.rs:here}}
```

<span class="caption">Listing 4-1: A variable and the scope in which it is
valid</span>

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 introducing the `String` type.

### The `String` Type

To illustrate the rules of ownership, we need a data type that is more complex
than the ones we covered in the [“Data Types”][data-types]<!-- ignore -->
section of Chapter 3. The types covered previously are all a known size, can be
stored on the stack and popped off the stack when their scope is over, and can
be quickly and trivially copied to make a new, independent instance if another
part of code needs to use the same value in a different scope. But we want to
look at data that is stored on the heap and explore how Rust knows when to
clean up that data.

We’ll use `String` as the example here and concentrate on the parts of `String`
that relate to ownership. These aspects also apply to other complex data types,
whether they are provided by the standard library or created by you. We’ll
discuss `String` in more depth in Chapter 8.

We’ve already seen string literals, where a string value is hardcoded into our
program. String literals are convenient, but they aren’t suitable for every
situation in which we may want to use text. One reason is that they’re
immutable. Another is that not every string value can be known when we write
our code: for example, what if we want to take user input and store it? For
these situations, Rust has a second string type, `String`. This type manages
data allocated on the heap and as such is able to store an amount of text that
is unknown to us at compile time. You can create a `String` from a string
literal using the `from` function, like so:

```rust
let s = String::from("hello");
```

The double colon (`::`) is an operator that allows us to namespace this
particular `from` function under the `String` type rather than using some sort
of name like `string_from`. We’ll discuss this syntax more in the [“Method
Syntax”][method-syntax]<!-- ignore --> section of Chapter 5 and when we talk
about namespacing with modules in [“Paths for Referring to an Item in the
Module Tree”][paths-module-tree]<!-- ignore --> in Chapter 7.

This kind of string *can* be mutated:

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-01-can-mutate-string/src/main.rs:here}}
```

So, what’s the difference here? Why can `String` be mutated but literals
cannot? The difference is how these two types deal with memory.

### Memory and Allocation

In the case of a string literal, we know the contents at compile time, so the
text is hardcoded directly into the final executable. This is why string
literals are fast and efficient. But these properties only come from the string
literal’s immutability. Unfortunately, we can’t put a blob of memory into the
binary for each piece of text whose size is unknown at compile time and whose
size might change while running the program.

With the `String` type, in order to support a mutable, growable piece of text,
we need to allocate an amount of memory on the heap, unknown at compile time,
to hold the contents. This means:

* The memory must be requested from the memory allocator at runtime.
* We need a way of returning this memory to the allocator when we’re
  done with our `String`.

That first part is done by us: when we call `String::from`, its implementation
requests the memory it needs. This is pretty much universal in programming
languages.

However, the second part is different. In languages with a *garbage collector
(GC)*, the GC keeps track and cleans up memory that isn’t being used anymore,
and we don’t need to think about it. Without a GC, it’s our responsibility to
identify when memory is no longer being used and call code to explicitly return
it, just as we did to request it. Doing this correctly has historically been a
difficult programming problem. If we forget, we’ll waste memory. If we do it
too early, we’ll have an invalid variable. If we do it twice, that’s a bug too.
We need to pair exactly one `allocate` with exactly one `free`.

Rust takes a different path: the memory is automatically returned once the
variable that owns it goes out of scope. Here’s a version of our scope example
from Listing 4-1 using a `String` instead of a string literal:

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-02-string-scope/src/main.rs:here}}
```

There is a natural point at which we can return the memory our `String` needs
to the allocator: when `s` goes out of scope. When a variable goes out of
scope, Rust calls a special function for us. This function is called
[`drop`][drop]<!-- ignore -->, and it’s where the author of `String` can put
the code to return the memory. Rust calls `drop` automatically at the closing
curly bracket.

> Note: In C++, this pattern of deallocating resources at the end of an item’s
> lifetime is sometimes called *Resource Acquisition Is Initialization (RAII)*.
> The `drop` function in Rust will be familiar to you if you’ve used RAII
> patterns.

This pattern has a profound impact on the way Rust code is written. It may seem
simple right now, but the behavior of code can be unexpected in more
complicated situations when we want to have multiple variables use the data
we’ve allocated on the heap. Let’s explore some of those situations now.

#### Ways Variables and Data Interact: Move

Multiple variables can interact with the same data in different ways in Rust.
Let’s look at an example using an integer in Listing 4-2.

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/listing-04-02/src/main.rs:here}}
```

<span class="caption">Listing 4-2: Assigning the integer value of variable `x`
to `y`</span>

We can probably guess what this is doing: “bind the value `5` to `x`; then make
a copy of the value in `x` and bind it to `y`.” We now have two variables, `x`
and `y`, and both equal `5`. This is indeed what is happening, because integers
are simple values with a known, fixed size, and these two `5` values are pushed
onto the stack.

Now let’s look at the `String` version:

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-03-string-move/src/main.rs:here}}
```

This looks very similar to the previous code, so we might assume that the way
it works would be the same: that is, the second line would make a copy of the
value in `s1` and bind it to `s2`. But this isn’t quite what happens.

Take a look at Figure 4-1 to see what is happening to `String` under the
covers. A `String` is made up of three parts, shown on the left: a pointer to
the memory that holds the contents of the string, a length, and a capacity.
This group of data is stored on the stack. On the right is the memory on the
heap that holds the contents.

<img alt="String in memory" src="img/trpl04-01.svg" class="center" style="width: 50%;" />

<span class="caption">Figure 4-1: Representation in memory of a `String`
holding the value `"hello"` bound to `s1`</span>

The length is how much memory, in bytes, the contents of the `String` is
currently using. The capacity is the total amount of memory, in bytes, that the
`String` has received from the allocator. The difference between length
and capacity matters, but not in this context, so for now, it’s fine to ignore
the capacity.

When we assign `s1` to `s2`, the `String` data is copied, meaning we copy the
pointer, the length, and the capacity that are on the stack. We do not copy the
data on the heap that the pointer refers to. In other words, the data
representation in memory looks like Figure 4-2.

<img alt="s1 and s2 pointing to the same value" src="img/trpl04-02.svg" class="center" style="width: 50%;" />

<span class="caption">Figure 4-2: Representation in memory of the variable `s2`
that has a copy of the pointer, length, and capacity of `s1`</span>

The representation does *not* look like Figure 4-3, which is what memory would
look like if Rust instead copied the heap data as well. If Rust did this, the
operation `s2 = s1` could be very expensive in terms of runtime performance if
the data on the heap were large.

<img alt="s1 and s2 to two places" src="img/trpl04-03.svg" class="center" style="width: 50%;" />

<span class="caption">Figure 4-3: Another possibility for what `s2 = s1` might
do if Rust copied the heap data as well</span>

Earlier, we said that when a variable goes out of scope, Rust automatically
calls the `drop` function and cleans up the heap memory for that variable. But
Figure 4-2 shows both data pointers pointing to the same location. This is a
problem: when `s2` and `s1` go out of scope, they will both try to free the
same memory. This is known as a *double free* error and is one of the memory
safety bugs we mentioned previously. Freeing memory twice can lead to memory
corruption, which can potentially lead to security vulnerabilities.

To ensure memory safety, there’s one more detail to what happens in this
situation in Rust. After `let s2 = s1`, Rust considers `s1` to no longer be
valid. Therefore, Rust doesn’t need to free anything when `s1` goes out of
scope. Check out what happens when you try to use `s1` after `s2` is created;
it won’t work:

```rust,ignore,does_not_compile
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-04-cant-use-after-move/src/main.rs:here}}
```

You’ll get an error like this because Rust prevents you from using the
invalidated reference:

```console
{{#include ../listings/ch04-understanding-ownership/no-listing-04-cant-use-after-move/output.txt}}
```

If you’ve heard the terms *shallow copy* and *deep copy* while working with
other languages, the concept of copying the pointer, length, and capacity
without copying the data probably sounds like making a shallow copy. But
because Rust also invalidates the first variable, instead of being called a
shallow copy, it’s known as a *move*. In this example, we would say that
`s1` was *moved* into `s2`. So what actually happens is shown in Figure 4-4.

<img alt="s1 moved to s2" src="img/trpl04-04.svg" class="center" style="width: 50%;" />

<span class="caption">Figure 4-4: Representation in memory after `s1` has been
invalidated</span>

That solves our problem! With only `s2` valid, when it goes out of scope, it
alone will free the memory, and we’re done.

In addition, there’s a design choice that’s implied by this: Rust will never
automatically create “deep” copies of your data. Therefore, any *automatic*
copying can be assumed to be inexpensive in terms of runtime performance.

#### Ways Variables and Data Interact: Clone

If we *do* want to deeply copy the heap data of the `String`, not just the
stack data, we can use a common method called `clone`. We’ll discuss method
syntax 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:

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-05-clone/src/main.rs:here}}
```

This works just fine and explicitly produces the behavior shown in Figure 4-3,
where the heap data *does* get copied.

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.

#### Stack-Only Data: Copy

There’s another wrinkle we haven’t talked about yet. This code using integers –
part of which was shown in Listing 4-2 – works and is valid:

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/no-listing-06-copy/src/main.rs:here}}
```

But this code seems to contradict what we just learned: we don’t have a call to
`clone`, but `x` is still valid and wasn’t moved into `y`.

The reason is that types such as integers that have a known size at compile
time are stored entirely on the stack, so copies of the actual values are quick
to make. That means there’s no reason we would want to prevent `x` from being
valid after we create the variable `y`. In other words, there’s no difference
between deep and shallow copying here, so calling `clone` wouldn’t do anything
different from the usual shallow copying and we can leave it out.

Rust has a special annotation called the `Copy` trait that we can place on
types like integers that are stored on the stack (we’ll talk more about traits
in Chapter 10). If a type implements the `Copy` trait, an older variable is
still usable after assignment. Rust won’t let us annotate a type with the
`Copy` trait if the type, or any of its parts, has implemented the `Drop`
trait. If the type needs something special to happen when the value goes out of
scope and we add the `Copy` annotation to that type, we’ll get a compile-time
error. To learn about how to add the `Copy` annotation to your type to
implement the trait, see [“Derivable Traits”][derivable-traits]<!-- ignore -->
in Appendix C.

So what types implement the `Copy` trait? You can check the documentation for
the given type to be sure, but as a general rule, any group of simple scalar
values can implement `Copy`, and nothing that requires allocation or is some
form of resource can implement `Copy`. Here are some of the types that
implement `Copy`:

* All the integer types, such as `u32`.
* The Boolean type, `bool`, with values `true` and `false`.
* All the floating point types, such as `f64`.
* The character type, `char`.
* Tuples, if they only contain types that also implement `Copy`. For example,
  `(i32, i32)` implements `Copy`, but `(i32, String)` does not.

### Ownership and Functions

The semantics for passing a value to a function are similar to those for
assigning a value to a variable. Passing a variable to a function will move or
copy, just as assignment does. Listing 4-3 has an example with some annotations
showing where variables go into and out of scope.

<span class="filename">Filename: src/main.rs</span>

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/listing-04-03/src/main.rs}}
```

<span class="caption">Listing 4-3: Functions with ownership and scope
annotated</span>

If we tried to use `s` after the call to `takes_ownership`, Rust would throw a
compile-time error. These static checks protect us from mistakes. Try adding
code to `main` that uses `s` and `x` to see where you can use them and where
the ownership rules prevent you from doing so.

### Return Values and Scope

Returning values can also transfer ownership. Listing 4-4 is an example with
similar annotations to those in Listing 4-3.

<span class="filename">Filename: src/main.rs</span>

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/listing-04-04/src/main.rs}}
```

<span class="caption">Listing 4-4: Transferring ownership of return
values</span>

The ownership of a variable follows the same pattern every time: assigning a
value to another variable moves it. When a variable that includes data on the
heap goes out of scope, the value will be cleaned up by `drop` unless the data
has been moved to be owned by another variable.

Taking ownership and then returning ownership with every function is a bit
tedious. What if we want to let a function use a value but not take ownership?
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.

It’s possible to return multiple values using a tuple, as shown in Listing 4-5.

<span class="filename">Filename: src/main.rs</span>

```rust
{{#rustdoc_include ../listings/ch04-understanding-ownership/listing-04-05/src/main.rs}}
```

<span class="caption">Listing 4-5: Returning ownership of parameters</span>

But this is too much ceremony and a lot of work for a concept that should be
common. Luckily for us, Rust has a feature for this concept, called
*references*.

[data-types]: ch03-02-data-types.html#data-types
[derivable-traits]: appendix-03-derivable-traits.html
[method-syntax]: ch05-03-method-syntax.html#method-syntax
[paths-module-tree]: ch07-03-paths-for-referring-to-an-item-in-the-module-tree.html
[drop]: ../std/ops/trait.Drop.html#tymethod.drop