Kotlin Flow 三 StateFlow 和 SharedFlow

StateFlow

StateFlow 和 LiveData 差不多，都是可观察的数据容器。在 StateFlow 中任何数据的发送，它的每一个接收器都能接收到。在 StateFlow 和 SharedFlow 中收集器也可以被称为订阅者，不过这个订阅者会挂起当前协程，而且永远不会结束。

private val state = MutableStateFlow(1)

suspend fun simpleStateFlow() {
    coroutineScope {
        launch {
            delay(1000)
            state.collect {
                println("before state value $it")
            }
        }
        launch {
            for (i in 1..100) {
                state.emit(i)
                delay(100)
            }
        }

        launch {
            state.collect {
                println("state value $it")
            }
        }
    }
}

需要注意的是 collect 是一个挂起函数，所以一旦调用 collect 协程就会被挂起，所以上述的例子中在一个协程中发送数据，在两个协程中接收数据。

和 LiveData 不同的在于， LiveData 不需要初始值，但 StateFlow 需要。

LiveData 会与 Activity 绑定，当 View 进入 STOPED 状态时， LiveData.observer() 会自动取消注册，而从 StateFlow 或任意其他数据流收集数据的操作并不会停止。如需实现相同的行为，需要从 Lifecycle.repeatOnLifecycle 块收集数据流。

StateFlow 是热流，并不是冷流。并且 StateFlow 的 collect 收不到调用之前发射的数据。

val state = MutableStateFlow(1)
coroutineScope {
    launch {
        for (i in 0..10) {
            state.emit(i)
            delay(1000)
        }
    }
                                        
    launch {
        delay(2000)
        state.collect {
            println("receive state $it")
        }
    }
}

可以看到最终的结果是：

receive state 2
receive state 3
receive state 4
receive state 5
receive state 6
receive state 7
receive state 8
receive state 9
receive state 10

因为在接受之前 delay 了 2s，所以最后是从 2 开始接收的。

把普通的 Flow 转化成 StateFlow 。

val flow = flow {
    for (i in 0..4) {
        emit(i)
        delay(100)
    }
}
coroutineScope {
    val stateFlow = flow.stateIn(this)
    launch {
        stateFlow.collect {
            println("receive flow.stateIn value $it")
        }
    }
}

我们同样可以像 LiveData 一样直接获取它的值。

stateFlow.value

StateFlow 分为 StateFlow 和 MutableStateFlow 。就像 LiveData 和 MutableLiveData 一样。 StateFlow 只能接收数据，不能发送数据，而 MutableStateFlow 即可以发送也可以接收。

private suspend fun simpleStateFlowAndMutableStateFlow() {
    val mutableStateFlow = MutableStateFlow(1)
    coroutineScope {
        launch {
            collectData(mutableStateFlow.asStateFlow())
        }
        launch {
            (1..10).forEach {
                delay(100)
                mutableStateFlow.emit(it)
            }
        }
    }
}

如上代码所述，可以将 MutableStateFlow 通过 asStateFlow 转换成 StateFlow 。

StateFlow 中给我们提供了一个协程安全的并发修改 StateFlow 中的值的方法 compareAndSet 。该方法能够保证原子的修改 StateFlow 的值。该方法是通过 CAS 来修改值。

public fun compareAndSet(expect: T, update: T): Boolean

将当前的值和期待的值进行比较，如果相等则更新当前的值，并返回 true，如果不相等则返回 false。这里的比较并修改是原子性的。

SharedFlow

SharedFlow 和 StateFlow 相比，他有缓冲区区，并可以定义缓冲区的溢出规则，已经可以定义给一个新的接收器发送多少数据的缓存值。

SharedFlow 同样有与之对应的 MutableSharedFlow 。 MutableSharedFlow 的参数如下：

• replay 给一个新的订阅者发送的缓冲区的数量。
• extraBufferCapacity 除了 replay 的数量之外的缓冲区的大小。
• onBufferOverflow 缓冲区溢出规则
  • SUSPEND 挂起
  • DROP_OLDEST 移除旧的值
  • DROP_LATEST 移除新的值

SharedFlow 的缓冲区大于是 replay + extraBufferCapacity 。

注意相比于 MutableStateFlow ， MutableSharedFlow 不需要初始值。

suspend fun simpleSharedFlow() {
    val sharedFlow = MutableSharedFlow(
        replay = 5,
        extraBufferCapacity = 3,
    )
    coroutineScope {
        launch {
            sharedFlow.collect {
                println("collect1 received shared flow $it")
            }
        }
        launch {
            (1..10).forEach {
                sharedFlow.emit(it)
                delay(100)
            }
        }
        // wait a minute
        delay(1000)
        launch {
            sharedFlow.collect {
                println("collect2 received shared flow $it")
            }
        }
    }
}

同样的，我们可以把普通的 Flow 转换成 SharedFlow。

suspend fun simpleConvertToSharedFlow(started: SharingStarted) {
    var start = 0L
    // create normal flow
    val flow = (1..10).asFlow()
        .onStart { start = currTime() }
        .onEach {
            println("Emit $it ${currTime() - start}ms")
            delay(100)
        }
    // convert to shared flow
    // need coroutine scope
    coroutineScope {
        val sharedFlow = flow.shareIn(this, started, replay = 2)
        delay(400)
        launch {
            println("current time ")
            sharedFlow.collect {
                println("received convert shared flow $it at ${currTime() - start}ms")
            }
        }
    }
}

这里的转换有些复杂，可以看到我们通过 shareIn 可以将普通的 flow 转换成 SharedFlow 。可以看到 sharedIn 有三个参数：

• CoroutineScope - sharing 的协程的作用域。
• SharingStarted - 启动模式
  • Eagerly 迫切的，渴望的，在转换完成后立即开始 sharing 数据，当上游的数据超过 replay 的时候，前面的数据就会被丢弃，相当于 DROP_OLDEST 。
  • Lazily 当有第一个订阅者（调用 collect）的时候开始发射数据。
  • WhileSubscribed 当第一个订阅者出现的时候立即开始，当最后一个订阅者消失的时立即停止（默认情况下），replay 数量的缓存值将永远保留（默认情况下）。这是一个函数，可以通过参数来控制当最后一个订阅者消失时的行为，以及缓存的有效期。
    • stopTimeoutMillis - 配置最后一个订阅者消失后 sharing flow 停止的延时。
    • replayExpirationMillis - 配置 sharing flow 协程的停止和重置缓冲区之间的间隔，单位是毫秒，默认值为 Long.MAX_VALUE 缓存永远都不重置，0 表示立即重置缓存。比较难懂可以看看下面的例子。
• replay 当订阅的时候回复的数量。

如果上面的函数中传递的是 Eagerly ，其输出如下：

Emit 1 2ms
Emit 2 109ms
Emit 3 213ms
Emit 4 313ms
current time 
received convert shared flow 2 at 412ms
received convert shared flow 3 at 412ms
Emit 5 413ms
received convert shared flow 4 at 414ms
Emit 6 518ms
received convert shared flow 5 at 519ms
Emit 7 619ms
received convert shared flow 6 at 619ms
Emit 8 720ms
received convert shared flow 7 at 720ms
Emit 9 822ms
received convert shared flow 8 at 823ms
Emit 10 926ms
received convert shared flow 9 at 926ms
received convert shared flow 10 at 1027ms

如果传入的是 Lazily ，其输入如下：

current time 
Emit 1 2ms
Emit 2 105ms
received convert shared flow 1 at 106ms
Emit 3 209ms
received convert shared flow 2 at 209ms
Emit 4 313ms
received convert shared flow 3 at 313ms
Emit 5 415ms
received convert shared flow 4 at 415ms
Emit 6 518ms
received convert shared flow 5 at 518ms
Emit 7 622ms
received convert shared flow 6 at 622ms
Emit 8 725ms
received convert shared flow 7 at 725ms
Emit 9 826ms
received convert shared flow 8 at 826ms
Emit 10 932ms
received convert shared flow 9 at 932ms
received convert shared flow 10 at 1032ms

很明显能够看出两者的区别。

下面看看 WhileSubscribed ，这种方式非常灵活。

fun currTime() = System.currentTimeMillis()

suspend fun simpleConvertToSharedFlow(started: SharingStarted) {
    var start = 0L
    // create normal flow
    val flow = (1..10).asFlow()
        .onStart { start = currTime() }
        .onEach {
            println("Emit $it ${currTime() - start}ms")
            delay(100)
        }
    // convert to shared flow
    // need coroutine scope
    coroutineScope {
        val sharedFlow = flow.shareIn(this, started, replay = 2)
        val job = launch {
            println("current time ")
            sharedFlow.collect {
                println("received convert shared flow $it at ${currTime() - start}ms")
            }
        }

        launch {
            delay(1000L)
            job.cancel()
            delay(110L)
            sharedFlow.collect {
                println("received again shared flow $it")
            }
            println("shared flow has stop")
        }
    }
}

@OptIn(ExperimentalTime::class)
suspend fun main() {
//    simpleSharedFlow()
    simpleConvertToSharedFlow(
        SharingStarted.WhileSubscribed(
            stopTimeout = 100L.toDuration(DurationUnit.MILLISECONDS),
            replayExpiration = 200L.toDuration(DurationUnit.MILLISECONDS)
        )
    )
}

这里配置当最后一个订阅者消失时 delay 100ms 后停止 sharing flow，在 sharing flow 停止后 200ms 后让缓存失效。这里可以通过调整 job.cancel 后的 delay 函数的时长来看看效果。当时间为 110ms 时，会重新接受到缓存 9 和 10，并重新开始 sharing flow，如果参数调整为 320ms 时，缓存会失效，会直接重新开始 sharing flow。

110ms 的结果：

current time 
Emit 1 1ms
Emit 2 107ms
received convert shared flow 1 at 108ms
Emit 3 211ms
received convert shared flow 2 at 211ms
Emit 4 315ms
received convert shared flow 3 at 315ms
Emit 5 417ms
received convert shared flow 4 at 417ms
Emit 6 521ms
received convert shared flow 5 at 521ms
Emit 7 623ms
received convert shared flow 6 at 624ms
Emit 8 727ms
received convert shared flow 7 at 727ms
Emit 9 829ms
received convert shared flow 8 at 829ms
Emit 10 933ms
received convert shared flow 9 at 933ms
received again shared flow 9
received again shared flow 10
Emit 1 0ms
Emit 2 105ms
received again shared flow 1
Emit 3 210ms
received again shared flow 2
Emit 4 314ms
received again shared flow 3
Emit 5 415ms
received again shared flow 4
Emit 6 519ms
received again shared flow 5
Emit 7 620ms
received again shared flow 6
Emit 8 721ms
received again shared flow 7
Emit 9 826ms
received again shared flow 8
Emit 10 927ms
received again shared flow 9
received again shared flow 10

320ms 的结果：

current time 
Emit 1 1ms
Emit 2 106ms
received convert shared flow 1 at 106ms
Emit 3 210ms
received convert shared flow 2 at 210ms
Emit 4 314ms
received convert shared flow 3 at 314ms
Emit 5 414ms
received convert shared flow 4 at 414ms
Emit 6 517ms
received convert shared flow 5 at 517ms
Emit 7 623ms
received convert shared flow 6 at 623ms
Emit 8 726ms
received convert shared flow 7 at 727ms
Emit 9 827ms
received convert shared flow 8 at 827ms
Emit 10 931ms
received convert shared flow 9 at 931ms
Emit 1 0ms
Emit 2 105ms
received again shared flow 1
Emit 3 209ms
received again shared flow 2
Emit 4 315ms
received again shared flow 3
Emit 5 418ms
received again shared flow 4
Emit 6 523ms
received again shared flow 5
Emit 7 627ms
received again shared flow 6
Emit 8 732ms
received again shared flow 7
Emit 9 833ms
received again shared flow 8
Emit 10 937ms
received again shared flow 9
received again shared flow 10

我们看下面这段源码就会很快明白：

override fun command(subscriptionCount: StateFlow): Flow = subscriptionCount
    .transformLatest { count ->
        if (count > 0) {
            emit(SharingCommand.START)
        } else {
            delay(stopTimeout)
            if (replayExpiration > 0) {
                emit(SharingCommand.STOP)
                delay(replayExpiration)
            }
            emit(SharingCommand.STOP_AND_RESET_REPLAY_CACHE)
        }
    }
    .dropWhile { it != SharingCommand.START } // don't emit any STOP/RESET_BUFFER to start with, only START
    .distinctUntilChanged() // just in case somebody forgets it, don't leak our multiple sending of START