Declaration-site variance can be thought of as declaration of use-site variance once and for all the use-sites.

  class Consumer<in T> { fun consume(t: T) { ... } }

  fun charSequencesConsumer() : Consumer<CharSequence>() = ...

  val stringConsumer : Consumer<String> = charSequenceConsumer() // OK since in-projection
  val anyConsumer : Consumer<Any> = charSequenceConsumer() // Error, Any cannot be passed
  
  val outConsumer : Consumer<out CharSequence> = ... // Error, T is `in`-parameter

Widespread examples of declaration-site variance are List<out T>, which is immutable so that T only appears as the return value type, and Comparator<in T>, which only receives T as argument.

Use-site variance is similar to Java wildcards:

Out-projection:

  val takeList : MutableList<out SomeType> = ... // Java: List<? extends SomeType>

  val takenValue : SomeType = takeList[0] // OK, since upper bound is SomeType

  takeList.add(takenValue) // Error, lower bound for generic is not specified

In-projection:

  val putList : MutableList<in SomeType> = ... // Java: List<? super SomeType>
  
  val valueToPut : SomeType = ...
  putList.add(valueToPut) // OK, since lower bound is SomeType

  putList[0] // This expression has type Any, since no upper bound is specified

Star-projection

  val starList : MutableList<*> = ... // Java: List<?>

  starList[0] // This expression has type Any, since no upper bound is specified
  starList.add(someValue) // Error, lower bound for generic is not specified

See also:

• Variant Generics interoperability when calling Kotlin from Java.