当 kotlin 代码块作为 lambda 传递给另一个函数时，其行为会有所不同[重复]

import kotlinx.coroutines.*

fun main() = runBlocking {
    println("Main program starts: ${Thread.currentThread().name}")
    withContext(Dispatchers.IO.limitedParallelism(1)) {
        println("Switched to limited parallelism context: ${Thread.currentThread().name}")
        repeat(2) { i ->
            launch {
                println("Coroutine $i starts on thread ${Thread.currentThread().name}")
                delay(1000L)
            }
        }
    }
    println("Main program ends: ${Thread.currentThread().name}")
}

上述 Kotlin 代码打印以下结果

Main program starts: main
Switched to limited parallelism context: DefaultDispatcher-worker-1
Coroutine 0 starts on thread DefaultDispatcher-worker-1
Coroutine 1 starts on thread DefaultDispatcher-worker-1
Main program ends: main

但是如果我编写一个自定义函数，withContext(Dispatchers.IO.limitedParallelism(1))那么里面的代码的行为withContext就会改变。

import kotlinx.coroutines.*

fun main() = runBlocking {
    println("Main program starts: ${Thread.currentThread().name}")
    atomically {
        println("Switched to limited parallelism context: ${Thread.currentThread().name}")
        repeat(2) { i ->
            launch {
                println("Coroutine $i starts on thread ${Thread.currentThread().name}")
                delay(1000L)
            }
        }
    }
    println("Main program ends: ${Thread.currentThread().name}")
}

internal suspend inline fun <R> atomically(crossinline f: () -> R): R =
    withContext(Dispatchers.IO.limitedParallelism(1)) { f() }

上述 Kotlin 代码打印以下结果

Main program starts: main
Switched to limited parallelism context: DefaultDispatcher-worker-1
Coroutine 0 starts on thread main
Coroutine 1 starts on thread main
Main program ends: main

如您所见， 中的两个子协程withContext并未在 上启动，DefaultDispatcher而是在 上运行main。我想了解为什么会发生这种行为变化。