Kotlin - Property initialization using "by lazy" vs. "lateinit"

regmoraes picture regmoraes · Apr 14, 2016 · Viewed 101.4k times · Source

In Kotlin if you don't want to initialize a class property inside the constructor or in the top of the class body, you have basically these two options (from the language reference):

  1. Lazy Initialization

lazy() is a function that takes a lambda and returns an instance of Lazy<T> which can serve as a delegate for implementing a lazy property: the first call to get() executes the lambda passed to lazy() and remembers the result, subsequent calls to get() simply return the remembered result.

Example

public class Hello {

   val myLazyString: String by lazy { "Hello" }

}

So the first call and the subsequential calls, wherever it is, to myLazyString will return Hello

  1. Late Initialization

Normally, properties declared as having a non-null type must be initialized in the constructor. However, fairly often this is not convenient. For example, properties can be initialized through dependency injection, or in the setup method of a unit test. In this case, you cannot supply a non-null initializer in the constructor, but you still want to avoid null checks when referencing the property inside the body of a class.

To handle this case, you can mark the property with the lateinit modifier:

public class MyTest {
   
   lateinit var subject: TestSubject

   @SetUp fun setup() { subject = TestSubject() }

   @Test fun test() { subject.method() }
}

The modifier can only be used on var properties declared inside the body of a class (not in the primary constructor), and only when the property does not have a custom getter or setter. The type of the property must be non-null, and it must not be a primitive type.

So, how to choose correctly between these two options, since both of them can solve the same problem?

Answer

hotkey picture hotkey · Apr 14, 2016

Here are the significant differences between lateinit var and by lazy { ... } delegated property:

  • lazy { ... } delegate can only be used for val properties, whereas lateinit can only be applied to vars, because it can't be compiled to a final field, thus no immutability can be guaranteed;

  • lateinit var has a backing field which stores the value, and by lazy { ... } creates a delegate object in which the value is stored once calculated, stores the reference to the delegate instance in the class object and generates the getter for the property that works with the delegate instance. So if you need the backing field present in the class, use lateinit;

  • In addition to vals, lateinit cannot be used for nullable properties or Java primitive types (this is because of null used for uninitialized value);

  • lateinit var can be initialized from anywhere the object is seen from, e.g. from inside a framework code, and multiple initialization scenarios are possible for different objects of a single class. by lazy { ... }, in turn, defines the only initializer for the property, which can be altered only by overriding the property in a subclass. If you want your property to be initialized from outside in a way probably unknown beforehand, use lateinit.

  • Initialization by lazy { ... } is thread-safe by default and guarantees that the initializer is invoked at most once (but this can be altered by using another lazy overload). In the case of lateinit var, it's up to the user's code to initialize the property correctly in multi-threaded environments.

  • A Lazy instance can be saved, passed around and even used for multiple properties. On contrary, lateinit vars do not store any additional runtime state (only null in the field for uninitialized value).

  • If you hold a reference to an instance of Lazy, isInitialized() allows you to check whether it has already been initialized (and you can obtain such instance with reflection from a delegated property). To check whether a lateinit property has been initialized, you can use property::isInitialized since Kotlin 1.2.

  • A lambda passed to by lazy { ... } may capture references from the context where it is used into its closure.. It will then store the references and release them only once the property has been initialized. This may lead to object hierarchies, such as Android activities, not being released for too long (or ever, if the property remains accessible and is never accessed), so you should be careful about what you use inside the initializer lambda.

Also, there's another way not mentioned in the question: Delegates.notNull(), which is suitable for deferred initialization of non-null properties, including those of Java primitive types.