src/doc/book/src/ch17-02-trait-objects.md

   1 ## Using Trait Objects That Allow for Values of Different Types
   2
   3 In Chapter 8, we mentioned that one limitation of vectors is that they can
   4 store elements of only one type. We created a workaround in Listing 8-9 where
   5 we defined a `SpreadsheetCell` enum that had variants to hold integers, floats,
   6 and text. This meant we could store different types of data in each cell and
   7 still have a vector that represented a row of cells. This is a perfectly good
   8 solution when our interchangeable items are a fixed set of types that we know
   9 when our code is compiled.
  10
  11 However, sometimes we want our library user to be able to extend the set of
  12 types that are valid in a particular situation. To show how we might achieve
  13 this, we’ll create an example graphical user interface (GUI) tool that iterates
  14 through a list of items, calling a `draw` method on each one to draw it to the
  15 screen—a common technique for GUI tools. We’ll create a library crate called
  16 `gui` that contains the structure of a GUI library. This crate might include
  17 some types for people to use, such as `Button` or `TextField`. In addition,
  18 `gui` users will want to create their own types that can be drawn: for
  19 instance, one programmer might add an `Image` and another might add a
  20 `SelectBox`.
  21
  22 We won’t implement a fully fledged GUI library for this example but will show
  23 how the pieces would fit together. At the time of writing the library, we can’t
  24 know and define all the types other programmers might want to create. But we do
  25 know that `gui` needs to keep track of many values of different types, and it
  26 needs to call a `draw` method on each of these differently typed values. It
  27 doesn’t need to know exactly what will happen when we call the `draw` method,
  28 just that the value will have that method available for us to call.
  29
  30 To do this in a language with inheritance, we might define a class named
  31 `Component` that has a method named `draw` on it. The other classes, such as
  32 `Button`, `Image`, and `SelectBox`, would inherit from `Component` and thus
  33 inherit the `draw` method. They could each override the `draw` method to define
  34 their custom behavior, but the framework could treat all of the types as if
  35 they were `Component` instances and call `draw` on them. But because Rust
  36 doesn’t have inheritance, we need another way to structure the `gui` library to
  37 allow users to extend it with new types.
  38
  39 ### Defining a Trait for Common Behavior
  40
  41 To implement the behavior we want `gui` to have, we’ll define a trait named
  42 `Draw` that will have one method named `draw`. Then we can define a vector that
  43 takes a *trait object*. A trait object points to both an instance of a type
  44 implementing our specified trait and a table used to look up trait methods on
  45 that type at runtime. We create a trait object by specifying some sort of
  46 pointer, such as a `&` reference or a `Box<T>` smart pointer, then the `dyn`
  47 keyword, and then specifying the relevant trait. (We’ll talk about the reason
  48 trait objects must use a pointer in Chapter 19 in the section [“Dynamically
  49 Sized Types and the `Sized` Trait.”][dynamically-sized]<!-- ignore -->) We can
  50 use trait objects in place of a generic or concrete type. Wherever we use a
  51 trait object, Rust’s type system will ensure at compile time that any value
  52 used in that context will implement the trait object’s trait. Consequently, we
  53 don’t need to know all the possible types at compile time.
  54
  55 We’ve mentioned that, in Rust, we refrain from calling structs and enums
  56 “objects” to distinguish them from other languages’ objects. In a struct or
  57 enum, the data in the struct fields and the behavior in `impl` blocks are
  58 separated, whereas in other languages, the data and behavior combined into one
  59 concept is often labeled an object. However, trait objects *are* more like
  60 objects in other languages in the sense that they combine data and behavior.
  61 But trait objects differ from traditional objects in that we can’t add data to
  62 a trait object. Trait objects aren’t as generally useful as objects in other
  63 languages: their specific purpose is to allow abstraction across common
  64 behavior.
  65
  66 Listing 17-3 shows how to define a trait named `Draw` with one method named
  67 `draw`:
  68
  69 <span class="filename">Filename: src/lib.rs</span>
  70
  71 ```rust,noplayground
  72 {{#rustdoc_include ../listings/ch17-oop/listing-17-03/src/lib.rs}}
  73 ```
  74
  75 <span class="caption">Listing 17-3: Definition of the `Draw` trait</span>
  76
  77 This syntax should look familiar from our discussions on how to define traits
  78 in Chapter 10. Next comes some new syntax: Listing 17-4 defines a struct named
  79 `Screen` that holds a vector named `components`. This vector is of type
  80 `Box<dyn Draw>`, which is a trait object; it’s a stand-in for any type inside
  81 a `Box` that implements the `Draw` trait.
  82
  83 <span class="filename">Filename: src/lib.rs</span>
  84
  85 ```rust,noplayground
  86 {{#rustdoc_include ../listings/ch17-oop/listing-17-04/src/lib.rs:here}}
  87 ```
  88
  89 <span class="caption">Listing 17-4: Definition of the `Screen` struct with a
  90 `components` field holding a vector of trait objects that implement the `Draw`
  91 trait</span>
  92
  93 On the `Screen` struct, we’ll define a method named `run` that will call the
  94 `draw` method on each of its `components`, as shown in Listing 17-5:
  95
  96 <span class="filename">Filename: src/lib.rs</span>
  97
  98 ```rust,noplayground
  99 {{#rustdoc_include ../listings/ch17-oop/listing-17-05/src/lib.rs:here}}
 100 ```
 101
 102 <span class="caption">Listing 17-5: A `run` method on `Screen` that calls the
 103 `draw` method on each component</span>
 104
 105 This works differently from defining a struct that uses a generic type
 106 parameter with trait bounds. A generic type parameter can only be substituted
 107 with one concrete type at a time, whereas trait objects allow for multiple
 108 concrete types to fill in for the trait object at runtime. For example, we
 109 could have defined the `Screen` struct using a generic type and a trait bound
 110 as in Listing 17-6:
 111
 112 <span class="filename">Filename: src/lib.rs</span>
 113
 114 ```rust,noplayground
 115 {{#rustdoc_include ../listings/ch17-oop/listing-17-06/src/lib.rs:here}}
 116 ```
 117
 118 <span class="caption">Listing 17-6: An alternate implementation of the `Screen`
 119 struct and its `run` method using generics and trait bounds</span>
 120
 121 This restricts us to a `Screen` instance that has a list of components all of
 122 type `Button` or all of type `TextField`. If you’ll only ever have homogeneous
 123 collections, using generics and trait bounds is preferable because the
 124 definitions will be monomorphized at compile time to use the concrete types.
 125
 126 On the other hand, with the method using trait objects, one `Screen` instance
 127 can hold a `Vec<T>` that contains a `Box<Button>` as well as a
 128 `Box<TextField>`. Let’s look at how this works, and then we’ll talk about the
 129 runtime performance implications.
 130
 131 ### Implementing the Trait
 132
 133 Now we’ll add some types that implement the `Draw` trait. We’ll provide the
 134 `Button` type. Again, actually implementing a GUI library is beyond the scope
 135 of this book, so the `draw` method won’t have any useful implementation in its
 136 body. To imagine what the implementation might look like, a `Button` struct
 137 might have fields for `width`, `height`, and `label`, as shown in Listing 17-7:
 138
 139 <span class="filename">Filename: src/lib.rs</span>
 140
 141 ```rust,noplayground
 142 {{#rustdoc_include ../listings/ch17-oop/listing-17-07/src/lib.rs:here}}
 143 ```
 144
 145 <span class="caption">Listing 17-7: A `Button` struct that implements the
 146 `Draw` trait</span>
 147
 148 The `width`, `height`, and `label` fields on `Button` will differ from the
 149 fields on other components; for example, a `TextField` type might have those
 150 same fields plus a `placeholder` field. Each of the types we want to draw on
 151 the screen will implement the `Draw` trait but will use different code in the
 152 `draw` method to define how to draw that particular type, as `Button` has here
 153 (without the actual GUI code, as mentioned). The `Button` type, for instance,
 154 might have an additional `impl` block containing methods related to what
 155 happens when a user clicks the button. These kinds of methods won’t apply to
 156 types like `TextField`.
 157
 158 If someone using our library decides to implement a `SelectBox` struct that has
 159 `width`, `height`, and `options` fields, they implement the `Draw` trait on the
 160 `SelectBox` type as well, as shown in Listing 17-8:
 161
 162 <span class="filename">Filename: src/main.rs</span>
 163
 164 ```rust,ignore
 165 {{#rustdoc_include ../listings/ch17-oop/listing-17-08/src/main.rs:here}}
 166 ```
 167
 168 <span class="caption">Listing 17-8: Another crate using `gui` and implementing
 169 the `Draw` trait on a `SelectBox` struct</span>
 170
 171 Our library’s user can now write their `main` function to create a `Screen`
 172 instance. To the `Screen` instance, they can add a `SelectBox` and a `Button`
 173 by putting each in a `Box<T>` to become a trait object. They can then call the
 174 `run` method on the `Screen` instance, which will call `draw` on each of the
 175 components. Listing 17-9 shows this implementation:
 176
 177 <span class="filename">Filename: src/main.rs</span>
 178
 179 ```rust,ignore
 180 {{#rustdoc_include ../listings/ch17-oop/listing-17-09/src/main.rs:here}}
 181 ```
 182
 183 <span class="caption">Listing 17-9: Using trait objects to store values of
 184 different types that implement the same trait</span>
 185
 186 When we wrote the library, we didn’t know that someone might add the
 187 `SelectBox` type, but our `Screen` implementation was able to operate on the
 188 new type and draw it because `SelectBox` implements the `Draw` trait, which
 189 means it implements the `draw` method.
 190
 191 This concept—of being concerned only with the messages a value responds to
 192 rather than the value’s concrete type—is similar to the concept of *duck
 193 typing* in dynamically typed languages: if it walks like a duck and quacks
 194 like a duck, then it must be a duck! In the implementation of `run` on `Screen`
 195 in Listing 17-5, `run` doesn’t need to know what the concrete type of each
 196 component is. It doesn’t check whether a component is an instance of a `Button`
 197 or a `SelectBox`, it just calls the `draw` method on the component. By
 198 specifying `Box<dyn Draw>` as the type of the values in the `components`
 199 vector, we’ve defined `Screen` to need values that we can call the `draw`
 200 method on.
 201
 202 The advantage of using trait objects and Rust’s type system to write code
 203 similar to code using duck typing is that we never have to check whether a
 204 value implements a particular method at runtime or worry about getting errors
 205 if a value doesn’t implement a method but we call it anyway. Rust won’t compile
 206 our code if the values don’t implement the traits that the trait objects need.
 207
 208 For example, Listing 17-10 shows what happens if we try to create a `Screen`
 209 with a `String` as a component:
 210
 211 <span class="filename">Filename: src/main.rs</span>
 212
 213 ```rust,ignore,does_not_compile
 214 {{#rustdoc_include ../listings/ch17-oop/listing-17-10/src/main.rs}}
 215 ```
 216
 217 <span class="caption">Listing 17-10: Attempting to use a type that doesn’t
 218 implement the trait object’s trait</span>
 219
 220 We’ll get this error because `String` doesn’t implement the `Draw` trait:
 221
 222 ```console
 223 {{#include ../listings/ch17-oop/listing-17-10/output.txt}}
 224 ```
 225
 226 This error lets us know that either we’re passing something to `Screen` we
 227 didn’t mean to pass and so should pass a different type or we should implement
 228 `Draw` on `String` so that `Screen` is able to call `draw` on it.
 229
 230 ### Trait Objects Perform Dynamic Dispatch
 231
 232 Recall in the [“Performance of Code Using
 233 Generics”][performance-of-code-using-generics]<!-- ignore --> section in
 234 Chapter 10 our discussion on the monomorphization process performed by the
 235 compiler when we use trait bounds on generics: the compiler generates
 236 nongeneric implementations of functions and methods for each concrete type that
 237 we use in place of a generic type parameter. The code that results from
 238 monomorphization is doing *static dispatch*, which is when the compiler knows
 239 what method you’re calling at compile time. This is opposed to *dynamic
 240 dispatch*, which is when the compiler can’t tell at compile time which method
 241 you’re calling. In dynamic dispatch cases, the compiler emits code that at
 242 runtime will figure out which method to call.
 243
 244 When we use trait objects, Rust must use dynamic dispatch. The compiler doesn’t
 245 know all the types that might be used with the code that’s using trait objects,
 246 so it doesn’t know which method implemented on which type to call. Instead, at
 247 runtime, Rust uses the pointers inside the trait object to know which method to
 248 call. This lookup incurs a runtime cost that doesn’t occur with static
 249 dispatch. Dynamic dispatch also prevents the compiler from choosing to inline a
 250 method’s code, which in turn prevents some optimizations. However, we did get
 251 extra flexibility in the code that we wrote in Listing 17-5 and were able to
 252 support in Listing 17-9, so it’s a trade-off to consider.
 253
 254 [performance-of-code-using-generics]:
 255 ch10-01-syntax.html#performance-of-code-using-generics
 256 [dynamically-sized]: ch19-04-advanced-types.html#dynamically-sized-types-and-the-sized-trait