Generics: in, out, where
Classes in Kotlin can have type parameters, just like in Java:
To create an instance of such a class, simply provide the type arguments:
But if the parameters can be inferred, for example, from the constructor arguments, you can omit the type arguments:
One of the trickiest aspects of Java's type system is the wildcard types (see Java Generics FAQ). Kotlin doesn't have these. Instead, Kotlin has declaration-site variance and type projections.
Let's think about why Java needs these mysterious wildcards. The problem is explained well in Effective Java, 3rd Edition, Item 31: Use bounded wildcards to increase API flexibility. First, generic types in Java are invariant, meaning that
List<String> is not a subtype of
List were not invariant, it would have been no better than Java's arrays, as the following code would have compiled but caused an exception at runtime:
Java prohibits such things in order to guarantee run-time safety. But this has implications. For example, consider the
addAll() method from the
Collection interface. What's the signature of this method? Intuitively, you'd write it this way:
But then, you would not be able to do the following (which is perfectly safe):
(In Java, you probably learned this the hard way, see Effective Java, 3rd Edition, Item 28: Prefer lists to arrays)
That's why the actual signature of
addAll() is the following:
The wildcard type argument
? extends E indicates that this method accepts a collection of objects of
E or a subtype of
E, not just
E itself. This means that you can safely read
E's from items (elements of this collection are instances of a subclass of E), but cannot write to it as you don't know what objects comply with that unknown subtype of
E. In return for this limitation, you get the desired behavior:
Collection<String> is a subtype of
Collection<? extends Object>. In other words, the wildcard with an extends-bound (upper bound) makes the type covariant.
The key to understanding why this works is rather simple: if you can only take items from a collection, then using a collection of
Strings and reading
Objects from it is fine. Conversely, if you can only put items into the collection, it's okay to take a collection of
Objects and put
Strings into it: in Java there is
List<? super String>, which accepts
Strings or any of its supertypes.
The latter is called contravariance, and you can only call methods that take
String as an argument on
List<? super String> (for example, you can call
set(int, String)). If you call something that returns
List<T>, you don't get a
String, but rather an
Joshua Bloch gives the name Producers to objects you only read from and Consumers to those you only write to. He recommends:
Let's suppose that there is a generic interface
Source<T> that does not have any methods that take
T as a parameter, only methods that return
Then, it would be perfectly safe to store a reference to an instance of
Source<String> in a variable of type
Source<Object> - there are no consumer-methods to call. But Java does not know this, and still prohibits it:
To fix this, you should declare objects of type
Source<? extends Object>. Doing so is meaningless, because you can call all the same methods on such a variable as before, so there's no value added by the more complex type. But the compiler does not know that.
In Kotlin, there is a way to explain this sort of thing to the compiler. This is called declaration-site variance: you can annotate the type parameter
Source to make sure that it is only returned (produced) from members of
Source<T>, and never consumed. To do this, use the
The general rule is this: when a type parameter
T of a class
C is declared
out, it may occur only in the out-position in the members of
C, but in return
C<Base> can safely be a supertype of
In other words, you can say that the class
C is covariant in the parameter
T, or that
T is a covariant type parameter. You can think of
C as being a producer of
T's, and NOT a consumer of
out modifier is called a variance annotation, and since it is provided at the type parameter declaration site, it provides declaration-site variance. This is in contrast with Java's use-site variance where wildcards in the type usages make the types covariant.
In addition to
out, Kotlin provides a complementary variance annotation:
in. It makes a type parameter contravariant, meaning it can only be consumed and never produced. A good example of a contravariant type is
The words in and out seem to be self-explanatory (as they've already been used successfully in C# for quite some time), and so the mnemonic mentioned above is not really needed. It can in fact be rephrased at a higher level of abstraction:
The Existential Transformation: Consumer in, Producer out!:-)
Use-site variance: type projections
It is very easy to declare a type parameter
out and avoid trouble with subtyping on the use site, but some classes can't actually be restricted to only return
T's! A good example of this is
This class can be neither co- nor contravariant in
T. And this imposes certain inflexibilities. Consider the following function:
This function is supposed to copy items from one array to another. Let's try to apply it in practice:
Here you run into the same familiar problem:
Array<T> is invariant in
T, and so neither
Array<Any> is a subtype of the other. Why not? Again, this is because
copy could have an unexpected behavior, for example, it may attempt to write a
from, and if you actually pass an array of
Int there, a
ClassCastException will be thrown later.
To prohibit the
copy function from writing to
from, you can do the following:
This is type projection, which means that
from is not a simple array, but is rather a restricted (projected) one. You can only call methods that return the type parameter
T, which in this case means that you can only call
get(). This is our approach to use-site variance, and it corresponds to Java's
Array<? extends Object> while being slightly simpler.
You can project a type with
in as well:
Array<in String> corresponds to Java's
Array<? super String>. This means that you can pass an array of
CharSequence or an array of
Object to the
Sometimes you want to say that you know nothing about the type argument, but you still want to use it in a safe way. The safe way here is to define such a projection of the generic type, that every concrete instantiation of that generic type will be a subtype of that projection.
Kotlin provides so-called star-projection syntax for this:
Foo<out T : TUpper>, where
Tis a covariant type parameter with the upper bound
Foo<*>is equivalent to
Foo<out TUpper>. This means that when the
Tis unknown you can safely read values of
Foo<in T>, where
Tis a contravariant type parameter,
Foo<*>is equivalent to
Foo<in Nothing>. This means there is nothing you can write to
Foo<*>in a safe way when
Foo<T : TUpper>, where
Tis an invariant type parameter with the upper bound
Foo<*>is equivalent to
Foo<out TUpper>for reading values and to
Foo<in Nothing>for writing values.
If a generic type has several type parameters, each of them can be projected independently. For example, if the type is declared as
interface Function<in T, out U> you could use the following star-projections:
Function<in Nothing, String>.
Function<Int, out Any?>.
Function<in Nothing, out Any?>.
Classes aren't the only declarations that can have type parameters. Functions can, too. Type parameters are placed before the name of the function:
To call a generic function, specify the type arguments at the call site after the name of the function:
Type arguments can be omitted if they can be inferred from the context, so the following example works as well:
The set of all possible types that can be substituted for a given type parameter may be restricted by generic constraints.
The most common type of constraint is an upper bound, which corresponds to Java's
The type specified after a colon is the upper bound, indicating that only a subtype of
Comparable<T> can be substituted for
T. For example:
The default upper bound (if there was none specified) is
Any?. Only one upper bound can be specified inside the angle brackets. If the same type parameter needs more than one upper bound, you need a separate where-clause:
The passed type must satisfy all conditions of the
where clause simultaneously. In the above example, the
T type must implement both
Definitely non-nullable types
To make interoperability with generic Java classes and interfaces easier, Kotlin supports declaring a generic type parameter as definitely non-nullable.
To declare a generic type
T as definitely non-nullable, declare the type with
& Any. For example:
T & Any.
A definitely non-nullable type must have a nullable upper bound.
The most common use case for declaring definitely non-nullable types is when you want to override a Java method that contains
@NotNull as an argument. For example, consider the
To override the
load() method in Kotlin successfully, you need
T1 to be declared as definitely non-nullable:
When working only with Kotlin, it's unlikely that you will need to declare definitely non-nullable types explicitly because Kotlin's type inference takes care of this for you.
The type safety checks that Kotlin performs for generic declaration usages are done at compile time. At runtime, the instances of generic types do not hold any information about their actual type arguments. The type information is said to be erased. For example, the instances of
Foo<Baz?> are erased to just
Generics type checks and casts
Due to the type erasure, there is no general way to check whether an instance of a generic type was created with certain type arguments at runtime, and the compiler prohibits such
is-checks such as
ints is List<Int> or
list is T (type parameter). However, you can check an instance against a star-projected type:
Similarly, when you already have the type arguments of an instance checked statically (at compile time), you can make an
is-check or a cast that involves the non-generic part of the type. Note that angle brackets are omitted in this case:
The same syntax but with the type arguments omitted can be used for casts that do not take type arguments into account:
list as ArrayList.
The type arguments of generic function calls are also only checked at compile time. Inside the function bodies, the type parameters cannot be used for type checks, and type casts to type parameters (
foo as T) are unchecked. The only exclusion is inline functions with reified type parameters, which have their actual type arguments inlined at each call site. This enables type checks and casts for the type parameters. However, the restrictions described above still apply for instances of generic types used inside checks or casts. For example, in the type check
arg is T, if
arg is an instance of a generic type itself, its type arguments are still erased.
Type casts to generic types with concrete type arguments such as
foo as List<String> cannot be checked at runtime.
These unchecked casts can be used when type safety is implied by the high-level program logic but cannot be inferred directly by the compiler. See the example below.
A warning appears for the cast in the last line. The compiler can't fully check it at runtime and provides no guarantee that the values in the map are
To avoid unchecked casts, you can redesign the program structure. In the example above, you could use the
DictionaryWriter<T> interfaces with type-safe implementations for different types. You can introduce reasonable abstractions to move unchecked casts from the call site to the implementation details. Proper use of generic variance can also help.
For generic functions, using reified type parameters makes casts like
arg as T checked, unless
arg's type has its own type arguments that are erased.
An unchecked cast warning can be suppressed by annotating the statement or the declaration where it occurs with
Underscore operator for type arguments
The underscore operator
_ can be used for type arguments. Use it to automatically infer a type of the argument when other types are explicitly specified: