Flow

interface Flow<out T> (source)

A cold asynchronous data stream that sequentially emits values and completes normally or with an exception.

Intermediate operators on the flow such as map, filter, take, zip, etc are functions that are applied to the upstream flow or flows and return a downstream flow where further operators can be applied to. Intermediate operations do not execute any code in the flow and are not suspending functions themselves. They only set up a chain of operations for future execution and quickly return. This is known as a cold flow property.

Terminal operators on the flow are either suspending functions such as collect, single, reduce, toList, etc. or launchIn operator that starts collection of the flow in the given scope. They are applied to the upstream flow and trigger execution of all operations. Execution of the flow is also called collecting the flow and is always performed in a suspending manner without actual blocking. Terminal operators complete normally or exceptionally depending on successful or failed execution of all the flow operations in the upstream. The most basic terminal operator is collect, for example:

try {
    flow.collect { value ->
        println("Received $value")
    }
} catch (e: Exception) {
    println("The flow has thrown an exception: $e")
}

By default, flows are sequential and all flow operations are executed sequentially in the same coroutine, with an exception for a few operations specifically designed to introduce concurrency into flow execution such as buffer and flatMapMerge. See their documentation for details.

The Flow interface does not carry information whether a flow truly is a cold stream that can be collected repeatedly and triggers execution of the same code every time it is collected, or if it is a hot stream that emits different values from the same running source on each collection. However, conventionally flows represent cold streams. Transitions between hot and cold streams are supported via channels and the corresponding API: channelFlow, produceIn, broadcastIn.

Flow builders

There are the following basic ways to create a flow:

Flow constraints

All implementations of the Flow interface must adhere to two key properties described in detail below:

These properties ensure the ability to perform local reasoning about the code with flows and modularize the code in such a way that upstream flow emitters can be developed separately from downstream flow collectors. A user of a flow does not need to be aware of implementation details of the upstream flows it uses.

Context preservation

The flow has a context preservation property: it encapsulates its own execution context and never propagates or leaks it downstream, thus making reasoning about the execution context of particular transformations or terminal operations trivial.

There is only one way to change the context of a flow: the flowOn operator that changes the upstream context (“everything above the flowOn operator”). For additional information refer to its documentation.

This reasoning can be demonstrated in practice:

val flowA = flowOf(1, 2, 3)
    .map { it + 1 } // Will be executed in ctxA
    .flowOn(ctxA) // Changes the upstream context: flowOf and map

// Now we have a context-preserving flow: it is executed somewhere but this information is encapsulated in the flow itself

val filtered = flowA // ctxA is encapsulated in flowA
   .filter { it == 3 } // Pure operator without a context yet

withContext(Dispatchers.Main) {
    // All non-encapsulated operators will be executed in Main: filter and single
    val result = filtered.single()
    myUi.text = result
}

From the implementation point of view, it means that all flow implementations should only emit from the same coroutine. This constraint is efficiently enforced by the default flow builder. The flow builder should be used if flow implementation does not start any coroutines. Its implementation prevents most of the development mistakes:

val myFlow = flow {
   // GlobalScope.launch { // is prohibited
   // launch(Dispatchers.IO) { // is prohibited
   // withContext(CoroutineName("myFlow")) // is prohibited
   emit(1) // OK
   coroutineScope {
       emit(2) // OK -- still the same coroutine
   }
}

Use channelFlow if the collection and emission of a flow are to be separated into multiple coroutines. It encapsulates all the context preservation work and allows you to focus on your domain-specific problem, rather than invariant implementation details. It is possible to use any combination of coroutine builders from within channelFlow.

If you are looking for performance and are sure that no concurrent emits and context jumps will happen, the flow builder can be used alongside a coroutineScope or supervisorScope instead:

Exception transparency

Flow implementations never catch or handle exceptions that occur in downstream flows. From the implementation standpoint it means that calls to emit and emitAll shall never be wrapped into try { ... } catch { ... } blocks. Exception handling in flows shall be performed with catch operator and it is designed to only catch exceptions coming from upstream flows while passing all downstream exceptions. Similarly, terminal operators like collect throw any unhandled exceptions that occur in their code or in upstream flows, for example:

flow { emitData() }
    .map { computeOne(it) }
    .catch { ... } // catches exceptions in emitData and computeOne
    .map { computeTwo(it) }
    .collect { process(it) } // throws exceptions from process and computeTwo

The same reasoning can be applied to the onCompletion operator that is a declarative replacement for the finally block.

Failure to adhere to the exception transparency requirement can lead to strange behaviors which make it hard to reason about the code because an exception in the collect { ... } could be somehow “caught” by an upstream flow, limiting the ability of local reasoning about the code.

Currently, the flow infrastructure does not enforce exception transparency contracts, however, it might be enforced in the future either at run time or at compile time.

Reactive streams

Flow is Reactive Streams compliant, you can safely interop it with reactive streams using Flow.asPublisher and Publisher.asFlow from kotlinx-coroutines-reactive module.

Functions

collect

abstract suspend fun collect(
    collector: FlowCollector<T>
): Unit

Accepts the given collector and emits values into it. This method should never be implemented or used directly.

Extension Functions

broadcastIn

fun <T> Flow<T>.broadcastIn(
    scope: CoroutineScope,
    start: CoroutineStart = CoroutineStart.LAZY
): BroadcastChannel<T>

Creates a broadcast coroutine that collects the given flow.

buffer

fun <T> Flow<T>.buffer(capacity: Int = BUFFERED): Flow<T>

Buffers flow emissions via channel of a specified capacity and runs collector in a separate coroutine.

catch

fun <T> Flow<T>.catch(
    action: suspend FlowCollector<T>.(cause: Throwable) -> Unit
): Flow<T>

Catches exceptions in the flow completion and calls a specified action with the caught exception. This operator is transparent to exceptions that occur in downstream flow and does not catch exceptions that are thrown to cancel the flow.

collect

suspend fun Flow<*>.collect(): Unit

Terminal flow operator that collects the given flow but ignores all emitted values. If any exception occurs during collect or in the provided flow, this exception is rethrown from this method.

suspend fun <T> Flow<T>.collect(
    action: suspend (value: T) -> Unit
): Unit

Terminal flow operator that collects the given flow with a provided action. If any exception occurs during collect or in the provided flow, this exception is rethrown from this method.

collectIndexed

suspend fun <T> Flow<T>.collectIndexed(
    action: suspend (index: Int, value: T) -> Unit
): Unit

Terminal flow operator that collects the given flow with a provided action that takes the index of an element (zero-based) and the element. If any exception occurs during collect or in the provided flow, this exception is rethrown from this method.

collectLatest

suspend fun <T> Flow<T>.collectLatest(
    action: suspend (value: T) -> Unit
): Unit

Terminal flow operator that collects the given flow with a provided action. The crucial difference from collect is that when the original flow emits a new value, action block for previous value is cancelled.

combine

fun <T1, T2, R> Flow<T1>.combine(
    flow: Flow<T2>,
    transform: suspend (a: T1, b: T2) -> R
): Flow<R>

Returns a Flow whose values are generated with transform function by combining the most recently emitted values by each flow.

combineLatest

fun <T1, T2, R> Flow<T1>.combineLatest(
    other: Flow<T2>,
    transform: suspend (T1, T2) -> R
): Flow<R>
fun <T1, T2, T3, R> Flow<T1>.combineLatest(
    other: Flow<T2>,
    other2: Flow<T3>,
    transform: suspend (T1, T2, T3) -> R
): <ERROR CLASS>
fun <T1, T2, T3, T4, R> Flow<T1>.combineLatest(
    other: Flow<T2>,
    other2: Flow<T3>,
    other3: Flow<T4>,
    transform: suspend (T1, T2, T3, T4) -> R
): <ERROR CLASS>
fun <T1, T2, T3, T4, T5, R> Flow<T1>.combineLatest(
    other: Flow<T2>,
    other2: Flow<T3>,
    other3: Flow<T4>,
    other4: Flow<T5>,
    transform: suspend (T1, T2, T3, T4, T5) -> R
): Flow<R>

combineTransform

fun <T1, T2, R> Flow<T1>.combineTransform(
    flow: Flow<T2>,
    transform: suspend FlowCollector<R>.(a: T1, b: T2) -> Unit
): Flow<R>

Returns a Flow whose values are generated by transform function that process the most recently emitted values by each flow.

conflate

fun <T> Flow<T>.conflate(): Flow<T>

Conflates flow emissions via conflated channel and runs collector in a separate coroutine. The effect of this is that emitter is never suspended due to a slow collector, but collector always gets the most recent value emitted.

count

suspend fun <T> Flow<T>.count(): Int

Returns the number of elements in this flow.

suspend fun <T> Flow<T>.count(
    predicate: suspend (T) -> Boolean
): Int

Returns the number of elements matching the given predicate.

debounce

fun <T> Flow<T>.debounce(timeoutMillis: Long): Flow<T>

Returns a flow that mirrors the original flow, but filters out values that are followed by the newer values within the given timeout. The latest value is always emitted.

distinctUntilChanged

fun <T> Flow<T>.distinctUntilChanged(): Flow<T>

Returns flow where all subsequent repetitions of the same value are filtered out.

fun <T> Flow<T>.distinctUntilChanged(
    areEquivalent: (old: T, new: T) -> Boolean
): Flow<T>

Returns flow where all subsequent repetitions of the same value are filtered out, when compared with each other via the provided areEquivalent function.

distinctUntilChangedBy

fun <T, K> Flow<T>.distinctUntilChangedBy(
    keySelector: (T) -> K
): Flow<T>

Returns flow where all subsequent repetitions of the same key are filtered out, where key is extracted with keySelector function.

drop

fun <T> Flow<T>.drop(count: Int): Flow<T>

Returns a flow that ignores first count elements. Throws IllegalArgumentException if count is negative.

dropWhile

fun <T> Flow<T>.dropWhile(
    predicate: suspend (T) -> Boolean
): Flow<T>

Returns a flow containing all elements except first elements that satisfy the given predicate.

filter

fun <T> Flow<T>.filter(
    predicate: suspend (T) -> Boolean
): Flow<T>

Returns a flow containing only values of the original flow that matches the given predicate.

filterIsInstance

fun <R> Flow<*>.filterIsInstance(): Flow<R>

Returns a flow containing only values that are instances of specified type R.

filterNot

fun <T> Flow<T>.filterNot(
    predicate: suspend (T) -> Boolean
): Flow<T>

Returns a flow containing only values of the original flow that do not match the given predicate.

filterNotNull

fun <T : Any> Flow<T?>.filterNotNull(): Flow<T>

Returns a flow containing only values of the original flow that are not null.

first

suspend fun <T> Flow<T>.first(): T

The terminal operator that returns the first element emitted by the flow and then cancels flow’s collection. Throws NoSuchElementException if the flow was empty.

suspend fun <T> Flow<T>.first(
    predicate: suspend (T) -> Boolean
): T

The terminal operator that returns the first element emitted by the flow matching the given predicate and then cancels flow’s collection. Throws NoSuchElementException if the flow has not contained elements matching the predicate.

flatMapConcat

fun <T, R> Flow<T>.flatMapConcat(
    transform: suspend (value: T) -> Flow<R>
): Flow<R>

Transforms elements emitted by the original flow by applying transform, that returns another flow, and then concatenating and flattening these flows.

flatMapLatest

fun <T, R> Flow<T>.flatMapLatest(
    transform: suspend (value: T) -> Flow<R>
): Flow<R>

Returns a flow that switches to a new flow produced by transform function every time the original flow emits a value. When the original flow emits a new value, the previous flow produced by transform block is cancelled.

flatMapMerge

fun <T, R> Flow<T>.flatMapMerge(
    concurrency: Int = DEFAULT_CONCURRENCY,
    transform: suspend (value: T) -> Flow<R>
): Flow<R>

Transforms elements emitted by the original flow by applying transform, that returns another flow, and then merging and flattening these flows.

flattenConcat

fun <T> Flow<Flow<T>>.flattenConcat(): Flow<T>

Flattens the given flow of flows into a single flow in a sequentially manner, without interleaving nested flows. This method is conceptually identical to flattenMerge(concurrency = 1) but has faster implementation.

flattenMerge

fun <T> Flow<Flow<T>>.flattenMerge(
    concurrency: Int = DEFAULT_CONCURRENCY
): Flow<T>

Flattens the given flow of flows into a single flow with a concurrency limit on the number of concurrently collected flows.

flowOn

fun <T> Flow<T>.flowOn(context: CoroutineContext): Flow<T>

Changes the context where this flow is executed to the given context. This operator is composable and affects only preceding operators that do not have its own context. This operator is context preserving: context does not leak into the downstream flow.

flowWith

fun <T, R> Flow<T>.flowWith(
    flowContext: CoroutineContext,
    bufferSize: Int = BUFFERED,
    builder: Flow<T>.() -> Flow<R>
): Flow<R>

The operator that changes the context where all transformations applied to the given flow within a builder are executed. This operator is context preserving and does not affect the context of the preceding and subsequent operations.

fold

suspend fun <T, R> Flow<T>.fold(
    initial: R,
    operation: suspend (acc: R, value: T) -> R
): R

Accumulates value starting with initial value and applying operation current accumulator value and each element

launchIn

fun <T> Flow<T>.launchIn(scope: CoroutineScope): Job

Terminal flow operator that launches the collection of the given flow in the scope. It is a shorthand for scope.launch { flow.collect() }.

map

fun <T, R> Flow<T>.map(
    transform: suspend (value: T) -> R
): Flow<R>

Returns a flow containing the results of applying the given transform function to each value of the original flow.

mapLatest

fun <T, R> Flow<T>.mapLatest(
    transform: suspend (value: T) -> R
): Flow<R>

Returns a flow that emits elements from the original flow transformed by transform function. When the original flow emits a new value, computation of the transform block for previous value is cancelled.

mapNotNull

fun <T, R : Any> Flow<T>.mapNotNull(
    transform: suspend (value: T) -> R?
): Flow<R>

Returns a flow that contains only non-null results of applying the given transform function to each value of the original flow.

onCompletion

fun <T> Flow<T>.onCompletion(
    action: suspend FlowCollector<T>.(cause: Throwable?) -> Unit
): Flow<T>

Invokes the given action when the given flow is completed or cancelled, using the exception from the upstream (if any) as cause parameter of action.

onEach

fun <T> Flow<T>.onEach(action: suspend (T) -> Unit): Flow<T>

Returns a flow which performs the given action on each value of the original flow.

onStart

fun <T> Flow<T>.onStart(
    action: suspend FlowCollector<T>.() -> Unit
): Flow<T>

Invokes the given action when the this flow starts to be collected.

produceIn

fun <T> Flow<T>.produceIn(
    scope: CoroutineScope
): ReceiveChannel<T>

Creates a produce coroutine that collects the given flow.

reduce

suspend fun <S, T : S> Flow<T>.reduce(
    operation: suspend (accumulator: S, value: T) -> S
): S

Accumulates value starting with the first element and applying operation to current accumulator value and each element. Throws UnsupportedOperationException if flow was empty.

retry

fun <T> Flow<T>.retry(
    retries: Long = Long.MAX_VALUE,
    predicate: suspend (cause: Throwable) -> Boolean = { true }
): Flow<T>

Retries collection of the given flow up to retries times when an exception that matches the given predicate occurs in the upstream flow. This operator is transparent to exceptions that occur in downstream flow and does not retry on exceptions that are thrown to cancel the flow.

fun <T> Flow<T>.retry(
    retries: Int = Int.MAX_VALUE,
    predicate: (Throwable) -> Boolean = { true }
): Flow<T>

retryWhen

fun <T> Flow<T>.retryWhen(
    predicate: suspend FlowCollector<T>.(cause: Throwable, attempt: Long) -> Boolean
): Flow<T>

Retries collection of the given flow when an exception occurs in the upstream flow and the predicate returns true. The predicate also receives an attempt number as parameter, starting from zero on the initial call. This operator is transparent to exceptions that occur in downstream flow and does not retry on exceptions that are thrown to cancel the flow.

sample

fun <T> Flow<T>.sample(periodMillis: Long): Flow<T>

Returns a flow that emits only the latest value emitted by the original flow during the given sampling period.

scan

fun <T, R> Flow<T>.scan(
    initial: R,
    operation: suspend (accumulator: R, value: T) -> R
): Flow<R>

Folds the given flow with operation, emitting every intermediate result, including initial value. Note that initial value should be immutable (or should not be mutated) as it is shared between different collectors. For example:

scanReduce

fun <T> Flow<T>.scanReduce(
    operation: suspend (accumulator: T, value: T) -> T
): Flow<T>

Reduces the given flow with operation, emitting every intermediate result, including initial value. The first element is taken as initial value for operation accumulator. This operator has a sibling with initial value – scan.

single

suspend fun <T> Flow<T>.single(): T

The terminal operator, that awaits for one and only one value to be published. Throws NoSuchElementException for empty flow and IllegalStateException for flow that contains more than one element.

singleOrNull

suspend fun <T : Any> Flow<T>.singleOrNull(): T?

The terminal operator, that awaits for one and only one value to be published. Throws IllegalStateException for flow that contains more than one element.

switchMap

fun <T, R> Flow<T>.switchMap(
    transform: suspend (value: T) -> Flow<R>
): Flow<R>

take

fun <T> Flow<T>.take(count: Int): Flow<T>

Returns a flow that contains first count elements. When count elements are consumed, the original flow is cancelled. Throws IllegalArgumentException if count is not positive.

takeWhile

fun <T> Flow<T>.takeWhile(
    predicate: suspend (T) -> Boolean
): Flow<T>

Returns a flow that contains first elements satisfying the given predicate.

toCollection

suspend fun <T, C : MutableCollection<in T>> Flow<T>.toCollection(
    destination: C
): C

Collects given flow into a destination

toList

suspend fun <T> Flow<T>.toList(
    destination: MutableList<T> = ArrayList()
): List<T>

Collects given flow into a destination

toSet

suspend fun <T> Flow<T>.toSet(
    destination: MutableSet<T> = LinkedHashSet()
): Set<T>

Collects given flow into a destination

transform

fun <T, R> Flow<T>.transform(
    transform: suspend FlowCollector<R>.(value: T) -> Unit
): Flow<R>

Applies transform function to each value of the given flow.

transformLatest

fun <T, R> Flow<T>.transformLatest(
    transform: suspend FlowCollector<R>.(value: T) -> Unit
): Flow<R>

Returns a flow that produces element by transform function every time the original flow emits a value. When the original flow emits a new value, the previous transform block is cancelled, thus the name transformLatest.

withIndex

fun <T> Flow<T>.withIndex(): Flow<IndexedValue<T>>

Returns a flow that wraps each element into IndexedValue, containing value and its index (starting from zero).

zip

fun <T1, T2, R> Flow<T1>.zip(
    other: Flow<T2>,
    transform: suspend (T1, T2) -> R
): Flow<R>

Zips values from the current flow (this) with other flow using provided transform function applied to each pair of values. The resulting flow completes as soon as one of the flows completes and cancel is called on the remaining flow.

Inheritors

AbstractFlow

abstract class AbstractFlow<T> : Flow<T>

Base class for stateful implementations of Flow. It tracks all the properties required for context preservation and throws an IllegalStateException if any of the properties are violated.