New upstream version 1.55.0+dfsg1

[rustc.git] / src / doc / book / src / ch06-01-defining-an-enum.md

## Defining an Enum

Let’s look at a situation we might want to express in code and see why enums
are useful and more appropriate than structs in this case. Say we need to work
with IP addresses. Currently, two major standards are used for IP addresses:
version four and version six. These are the only possibilities for an IP
address that our program will come across: we can *enumerate* all possible
variants, which is where enumeration gets its name.

Any IP address can be either a version four or a version six address, but not
both at the same time. That property of IP addresses makes the enum data
structure appropriate, because enum values can only be one of its variants.
Both version four and version six addresses are still fundamentally IP
addresses, so they should be treated as the same type when the code is handling
situations that apply to any kind of IP address.

We can express this concept in code by defining an `IpAddrKind` enumeration and
listing the possible kinds an IP address can be, `V4` and `V6`. These are the
variants of the enum:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-01-defining-enums/src/main.rs:def}}
```

`IpAddrKind` is now a custom data type that we can use elsewhere in our code.

### Enum Values

We can create instances of each of the two variants of `IpAddrKind` like this:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-01-defining-enums/src/main.rs:instance}}
```

Note that the variants of the enum are namespaced under its identifier, and we
use a double colon to separate the two. The reason this is useful is that now
both values `IpAddrKind::V4` and `IpAddrKind::V6` are of the same type:
`IpAddrKind`. We can then, for instance, define a function that takes any
`IpAddrKind`:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-01-defining-enums/src/main.rs:fn}}
```

And we can call this function with either variant:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-01-defining-enums/src/main.rs:fn_call}}
```

Using enums has even more advantages. Thinking more about our IP address type,
at the moment we don’t have a way to store the actual IP address *data*; we
only know what *kind* it is. Given that you just learned about structs in
Chapter 5, you might tackle this problem as shown in Listing 6-1.

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/listing-06-01/src/main.rs:here}}
```

<span class="caption">Listing 6-1: Storing the data and `IpAddrKind` variant of
an IP address using a `struct`</span>

Here, we’ve defined a struct `IpAddr` that has two fields: a `kind` field that
is of type `IpAddrKind` (the enum we defined previously) and an `address` field
of type `String`. We have two instances of this struct. The first, `home`, has
the value `IpAddrKind::V4` as its `kind` with associated address data of
`127.0.0.1`. The second instance, `loopback`, has the other variant of
`IpAddrKind` as its `kind` value, `V6`, and has address `::1` associated with
it. We’ve used a struct to bundle the `kind` and `address` values together, so
now the variant is associated with the value.

We can represent the same concept in a more concise way using just an enum,
rather than an enum inside a struct, by putting data directly into each enum
variant. This new definition of the `IpAddr` enum says that both `V4` and `V6`
variants will have associated `String` values:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-02-enum-with-data/src/main.rs:here}}
```

We attach data to each variant of the enum directly, so there is no need for an
extra struct.

There’s another advantage to using an enum rather than a struct: each variant
can have different types and amounts of associated data. Version four type IP
addresses will always have four numeric components that will have values
between 0 and 255. If we wanted to store `V4` addresses as four `u8` values but
still express `V6` addresses as one `String` value, we wouldn’t be able to with
a struct. Enums handle this case with ease:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-03-variants-with-different-data/src/main.rs:here}}
```

We’ve shown several different ways to define data structures to store version
four and version six IP addresses. However, as it turns out, wanting to store
IP addresses and encode which kind they are is so common that [the standard
library has a definition we can use!][IpAddr]<!-- ignore --> Let’s look at how
the standard library defines `IpAddr`: it has the exact enum and variants that
we’ve defined and used, but it embeds the address data inside the variants in
the form of two different structs, which are defined differently for each
variant:

[IpAddr]: ../std/net/enum.IpAddr.html

```rust
struct Ipv4Addr {
    // --snip--
}

struct Ipv6Addr {
    // --snip--
}

enum IpAddr {
    V4(Ipv4Addr),
    V6(Ipv6Addr),
}
```

This code illustrates that you can put any kind of data inside an enum variant:
strings, numeric types, or structs, for example. You can even include another
enum! Also, standard library types are often not much more complicated than
what you might come up with.

Note that even though the standard library contains a definition for `IpAddr`,
we can still create and use our own definition without conflict because we
haven’t brought the standard library’s definition into our scope. We’ll talk
more about bringing types into scope in Chapter 7.

Let’s look at another example of an enum in Listing 6-2: this one has a wide
variety of types embedded in its variants.

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/listing-06-02/src/main.rs:here}}
```

<span class="caption">Listing 6-2: A `Message` enum whose variants each store
different amounts and types of values</span>

This enum has four variants with different types:

* `Quit` has no data associated with it at all.
* `Move` has named fields like a struct does.
* `Write` includes a single `String`.
* `ChangeColor` includes three `i32` values.

Defining an enum with variants such as the ones in Listing 6-2 is similar to
defining different kinds of struct definitions, except the enum doesn’t use the
`struct` keyword and all the variants are grouped together under the `Message`
type. The following structs could hold the same data that the preceding enum
variants hold:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-04-structs-similar-to-message-enum/src/main.rs:here}}
```

But if we used the different structs, which each have their own type, we
couldn’t as easily define a function to take any of these kinds of messages as
we could with the `Message` enum defined in Listing 6-2, which is a single type.

There is one more similarity between enums and structs: just as we’re able to
define methods on structs using `impl`, we’re also able to define methods on
enums. Here’s a method named `call` that we could define on our `Message` enum:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-05-methods-on-enums/src/main.rs:here}}
```

The body of the method would use `self` to get the value that we called the
method on. In this example, we’ve created a variable `m` that has the value
`Message::Write(String::from("hello"))`, and that is what `self` will be in the
body of the `call` method when `m.call()` runs.

Let’s look at another enum in the standard library that is very common and
useful: `Option`.

### The `Option` Enum and Its Advantages Over Null Values

In the previous section, we looked at how the `IpAddr` enum let us use Rust’s
type system to encode more information than just the data into our program.
This section explores a case study of `Option`, which is another enum defined
by the standard library. The `Option` type is used in many places because it
encodes the very common scenario in which a value could be something or it
could be nothing. Expressing this concept in terms of the type system means the
compiler can check whether you’ve handled all the cases you should be handling;
this functionality can prevent bugs that are extremely common in other
programming languages.

Programming language design is often thought of in terms of which features you
include, but the features you exclude are important too. Rust doesn’t have the
null feature that many other languages have. *Null* is a value that means there
is no value there. In languages with null, variables can always be in one of
two states: null or not-null.

In his 2009 presentation “Null References: The Billion Dollar Mistake,” Tony
Hoare, the inventor of null, has this to say:

> I call it my billion-dollar mistake. At that time, I was designing the first
> comprehensive type system for references in an object-oriented language. My
> goal was to ensure that all use of references should be absolutely safe, with
> checking performed automatically by the compiler. But I couldn’t resist the
> temptation to put in a null reference, simply because it was so easy to
> implement. This has led to innumerable errors, vulnerabilities, and system
> crashes, which have probably caused a billion dollars of pain and damage in
> the last forty years.

The problem with null values is that if you try to use a null value as a
not-null value, you’ll get an error of some kind. Because this null or not-null
property is pervasive, it’s extremely easy to make this kind of error.

However, the concept that null is trying to express is still a useful one: a
null is a value that is currently invalid or absent for some reason.

The problem isn’t really with the concept but with the particular
implementation. As such, Rust does not have nulls, but it does have an enum
that can encode the concept of a value being present or absent. This enum is
`Option<T>`, and it is [defined by the standard library][option]<!-- ignore -->
as follows:

[option]: ../std/option/enum.Option.html

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

The `Option<T>` enum is so useful that it’s even included in the prelude; you
don’t need to bring it into scope explicitly. In addition, so are its variants:
you can use `Some` and `None` directly without the `Option::` prefix. The
`Option<T>` enum is still just a regular enum, and `Some(T)` and `None` are
still variants of type `Option<T>`.

The `<T>` syntax is a feature of Rust we haven’t talked about yet. It’s a
generic type parameter, and we’ll cover generics in more detail in Chapter 10.
For now, all you need to know is that `<T>` means the `Some` variant of the
`Option` enum can hold one piece of data of any type. Here are some examples of
using `Option` values to hold number types and string types:

```rust
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-06-option-examples/src/main.rs:here}}
```

If we use `None` rather than `Some`, we need to tell Rust what type of
`Option<T>` we have, because the compiler can’t infer the type that the `Some`
variant will hold by looking only at a `None` value.

When we have a `Some` value, we know that a value is present and the value is
held within the `Some`. When we have a `None` value, in some sense, it means
the same thing as null: we don’t have a valid value. So why is having
`Option<T>` any better than having null?

In short, because `Option<T>` and `T` (where `T` can be any type) are different
types, the compiler won’t let us use an `Option<T>` value as if it were
definitely a valid value. For example, this code won’t compile because it’s
trying to add an `i8` to an `Option<i8>`:

```rust,ignore,does_not_compile
{{#rustdoc_include ../listings/ch06-enums-and-pattern-matching/no-listing-07-cant-use-option-directly/src/main.rs:here}}
```

If we run this code, we get an error message like this:

```console
{{#include ../listings/ch06-enums-and-pattern-matching/no-listing-07-cant-use-option-directly/output.txt}}
```

Intense! In effect, this error message means that Rust doesn’t understand how
to add an `i8` and an `Option<i8>`, because they’re different types. When we
have a value of a type like `i8` in Rust, the compiler will ensure that we
always have a valid value. We can proceed confidently without having to check
for null before using that value. Only when we have an `Option<i8>` (or
whatever type of value we’re working with) do we have to worry about possibly
not having a value, and the compiler will make sure we handle that case before
using the value.

In other words, you have to convert an `Option<T>` to a `T` before you can
perform `T` operations with it. Generally, this helps catch one of the most
common issues with null: assuming that something isn’t null when it actually
is.

Not having to worry about incorrectly assuming a not-null value helps you to be
more confident in your code. In order to have a value that can possibly be
null, you must explicitly opt in by making the type of that value `Option<T>`.
Then, when you use that value, you are required to explicitly handle the case
when the value is null. Everywhere that a value has a type that isn’t an
`Option<T>`, you *can* safely assume that the value isn’t null. This was a
deliberate design decision for Rust to limit null’s pervasiveness and increase
the safety of Rust code.

So, how do you get the `T` value out of a `Some` variant when you have a value
of type `Option<T>` so you can use that value? The `Option<T>` enum has a large
number of methods that are useful in a variety of situations; you can check
them out in [its documentation][docs]<!-- ignore -->. Becoming familiar with
the methods on `Option<T>` will be extremely useful in your journey with Rust.

[docs]: ../std/option/enum.Option.html

In general, in order to use an `Option<T>` value, you want to have code that
will handle each variant. You want some code that will run only when you have a
`Some(T)` value, and this code is allowed to use the inner `T`. You want some
other code to run if you have a `None` value, and that code doesn’t have a `T`
value available. The `match` expression is a control flow construct that does
just this when used with enums: it will run different code depending on which
variant of the enum it has, and that code can use the data inside the matching
value.