Flow - Cold Streams for State and Events

Article · 2:00

A suspend function returns one value; a Flow returns many, over time.

A suspend function produces a single result. A Flow produces a sequence of values over time, asynchronously, and it is how Kotlin models anything that updates repeatedly: a database query that re-emits on change, location updates, search-as-you-type.

Flows are cold by default, meaning the producer block does not run until something collects, and it runs again for each collector. Collecting is itself a suspend operation, so a Flow is always consumed inside a coroutine and inherits that coroutine's cancellation - close the screen and the collection stops.

The operators mirror the collection ones - map, filter, take - but run in the coroutine world and can themselves suspend. Beyond those, the ones that matter in practice are debounce, which waits for a pause in emissions and is how you avoid a network call per keystroke; distinctUntilChanged, which drops consecutive duplicates; and flatMapLatest, which cancels the previous inner flow when a new value arrives.

StateFlow and SharedFlow are the hot variants. A StateFlow always holds a current value and emits it to every new collector, which makes it the standard way to expose UI state from a ViewModel - level five uses exactly this.

import kotlinx.coroutines.*
import kotlinx.coroutines.flow.*

// A cold flow: nothing runs until it is collected
fun temperatures(): Flow<Int> = flow {
    val readings = listOf(20, 20, 21, 23, 23, 26)
    for (reading in readings) {
        delay(100)
        emit(reading)
    }
}

fun main() = runBlocking {
    temperatures()
        .distinctUntilChanged()        // drop consecutive duplicates
        .filter { it > 20 }
        .map { "${it}C" }
        .collect { println(it) }       // 21C, 23C, 26C

    // StateFlow always has a current value
    val state = MutableStateFlow("idle")
    val watcher = launch {
        state.take(2).collect { println("state: $it") }
    }
    delay(50)
    state.value = "loading"
    watcher.join()
}

Discussion

No questions on this topic yet. Sign in to ask the first one.