Integration with Swift/Objective-C ARC

Kotlin and Objective-C use different memory management strategies. Kotlin has a tracing garbage collector, while Objective-C relies on automatic reference counting (ARC).

The integration between these strategies is usually seamless and generally requires no additional work. However, there are some specifics you should keep in mind:

Threads

Deinitializers

Deinitialization on the Swift/Objective-C objects and the objects they refer to is called on the main thread if these objects are passed to Kotlin on the main thread, for example:

// Kotlin
class KotlinExample {
    fun action(arg: Any) {
        println(arg)
    }
}

// Swift
class SwiftExample {
    init() {
        print("init on \(Thread.current)")
    }

    deinit {
        print("deinit on \(Thread.current)")
    }
}

func test() {
    KotlinExample().action(arg: SwiftExample())
}

The resulting output:

init on <_NSMainThread: 0x600003bc0000>{number = 1, name = main}
shared.SwiftExample
deinit on <_NSMainThread: 0x600003bc0000>{number = 1, name = main}

Deinitialization on the Swift/Objective-C objects is called on a special GC thread instead of the main one if:

Swift/Objective-C objects are passed to Kotlin on a thread other than main.
The main dispatch queue isn't processed.

If you want to call deinitialization on a special GC thread explicitly, set kotlin.native.binary.objcDisposeOnMain=false in your gradle.properties. This option enables deinitialization on a special GC thread, even if Swift/Objective-C objects were passed to Kotlin on the main thread.

A special GC thread complies with the Objective-C runtime, meaning that it has a run loop and drain autorelease pools.

Completion handlers

When calling Kotlin suspending functions from Swift, completion handlers might be called on threads other than the main one, for example:

// Kotlin
// coroutineScope, launch, and delay are from kotlinx.coroutines
suspend fun asyncFunctionExample() = coroutineScope {
    launch {
        delay(1000L)
        println("World!")
    }
    println("Hello")
}

// Swift
func test() {
    print("Running test on \(Thread.current)")
    PlatformKt.asyncFunctionExample(completionHandler: { _ in
        print("Running completion handler on \(Thread.current)")
    })
}

The resulting output:

Running test on <_NSMainThread: 0x600001b100c0>{number = 1, name = main}
Hello
World!
Running completion handler on <NSThread: 0x600001b45bc0>{number = 7, name = (null)}

Garbage collection and lifecycle

Object reclamation

An object is reclaimed only during the garbage collection. This applies to Swift/Objective-C objects that cross interop boundaries into Kotlin/Native, for example:

// Kotlin
class KotlinExample {
    fun action(arg: Any) {
        println(arg)
    }
}

// Swift
class SwiftExample {
    deinit {
        print("SwiftExample deinit")
    }
}

func test() {
    swiftTest()
    kotlinTest()
}

func swiftTest() {
    print(SwiftExample())
    print("swiftTestFinished")
}

func kotlinTest() {
    KotlinExample().action(arg: SwiftExample())
    print("kotlinTest finished")
}

The resulting output:

shared.SwiftExample
SwiftExample deinit
swiftTestFinished
shared.SwiftExample
kotlinTest finished
SwiftExample deinit

Objective-C objects lifecycle

The Objective-C objects might live longer than they should, which sometimes might cause performance issues. For example, when a long-running loop creates several temporary objects that cross the Swift/Objective-C interop boundary on each iteration.

In the GC logs, there's a number of stable refs in the root set. If this number keeps growing, it may indicate that the Swift/Objective-C objects are not freed up when they should. In this case, try the autoreleasepool block around loop bodies that do interop calls:

// Kotlin
fun growingMemoryUsage() {
    repeat(Int.MAX_VALUE) {
        NSLog("$it\n")
    }
}

fun steadyMemoryUsage() {
    repeat(Int.MAX_VALUE) {
        autoreleasepool {
            NSLog("$it\n")
        }
    }
}

Garbage collection of Swift and Kotlin objects' chains

Consider the following example:

// Kotlin
interface Storage {
    fun store(arg: Any)
}

class KotlinStorage(var field: Any? = null) : Storage {
    override fun store(arg: Any) {
        field = arg
    }
}

class KotlinExample {
    fun action(firstSwiftStorage: Storage, secondSwiftStorage: Storage) {
        // Here, we create the following chain:
        // firstKotlinStorage -> firstSwiftStorage -> secondKotlinStorage -> secondSwiftStorage.
        val firstKotlinStorage = KotlinStorage()
        firstKotlinStorage.store(firstSwiftStorage)
        val secondKotlinStorage = KotlinStorage()
        firstSwiftStorage.store(secondKotlinStorage)
        secondKotlinStorage.store(secondSwiftStorage)
    }
}

// Swift
class SwiftStorage : Storage {

    let name: String

    var field: Any? = nil

    init(_ name: String) {
        self.name = name
    }

    func store(arg: Any) {
        field = arg
    }

    deinit {
        print("deinit SwiftStorage \(name)")
    }
}

func test() {
    KotlinExample().action(
        firstSwiftStorage: SwiftStorage("first"),
        secondSwiftStorage: SwiftStorage("second")
    )
}

It takes some time between "deinit SwiftStorage first" and "deinit SwiftStorage second" messages to appear in the log. The reason is that firstKotlinStorage and secondKotlinStorage are collected in different GC cycles. Here's the sequence of events:

KotlinExample.action finishes. firstKotlinStorage is considered "dead" because nothing references it, while secondKotlinStorage is not because it is referenced by firstSwiftStorage.
First GC cycle starts, and firstKotlinStorage is collected.
There are no references to firstSwiftStorage, so it is "dead" as well, and deinit is called.
Second GC cycle starts. secondKotlinStorage is collected because firstSwiftStorage is no longer referencing it.
secondSwiftStorage is finally reclaimed.

It requires two GC cycles to collect these four objects because deinitialization of Swift and Objective-C objects happens after the GC cycle. The limitation stems from deinit, which can call arbitrary code, including the Kotlin code that cannot be run during the GC pause.

Retain cycles

In a retain cycle, a number of objects refer each other using strong references cyclically:

Kotlin's tracing GC and Objective-C's ARC handle retain cycles differently. When objects become unreachable, Kotlin's GC can properly reclaim such cycles, while Objective-C's ARC cannot. Therefore, retain cycles of Kotlin objects can be reclaimed, while retain cycles of Swift/Objective-C objects cannot.

Consider the case when a retain cycle contains both Objective-C and Kotlin objects:

Retain cycles with Objective-C and Kotlin objects

This involves combining Kotlin's and Objective-C's memory management models that cannot handle (reclaim) retain cycles together. That means if at least one Objective-C object is present, the retain cycle of a whole graph of objects cannot be reclaimed, and it's impossible to break the cycle from the Kotlin side.

Unfortunately, no special instruments are currently available to automatically detect retain cycles in Kotlin/Native code. To avoid retain cycles, use weak or unowned references.

Support for background state and App Extensions

The current memory manager does not track application state by default and does not integrate with App Extensions out of the box.

It means that the memory manager doesn't adjust GC behavior accordingly, which might be harmful in some cases. To change this behavior, add the following Experimental binary option to your gradle.properties:

kotlin.native.binary.appStateTracking=enabled

It turns off a timer-based invocation of the garbage collector when the application is in the background, so GC is called only when memory consumption becomes too high.

What's next

Learn more about Swift/Objective-C interoperability.

Last modified: 26 十一月 2024