在Kotlin协程被取消后无法捕获异常抛出

Question

使用kotlinx.coroutines库时，如果协程取消后抛出异常，我无法捕获异常。这会导致应用程序崩溃。

fun foo() {
  val job = launch(UI) {
     try {
        Log.d("TAG", "Start coroutine")
        run(CommonPool) {
           Log.d("TAG", "Start bg task")
           // Intentionally make bg task running for a long time
           SystemClock.sleep(2000)
           Log.d("TAG", "Throw bg task exception")
           throw RuntimeException("Bg task exception")
        }
     } catch (e: Exception) {
        Log.e("TAG", "Handle coroutine exception", e)
     }
  }

  launch(UI) {
     delay(1000)
     Log.d("TAG", "Cancel job = ${job.cancel()}")
  }

}

在Android上运行这些函数会产生以下日志输出。

07-26 15:09:10.038 31518-31518/co.foo.bar D/MainActivity: Start coroutine
07-26 15:09:10.044 31518-31547/co.foo.bar D/MainActivity: Start bg task
07-26 15:09:11.046 31518-31518/co.foo.bar D/MainActivity: Cancel job = true
07-26 15:09:11.047 31518-31518/co.foo.bar E/MainActivity: Handled coroutine exception
                           java.util.concurrent.CancellationException: Job was cancelled
                           at kotlinx.coroutines.experimental.JobSupport$CompletedExceptionally.getException(Job.kt:921)
                           at kotlinx.coroutines.experimental.RunCompletion.afterCompletion(Builders.kt:198)
                           ...
                           at com.android.internal.os.ZygoteInit.main(ZygoteInit.java:776)
07-26 15:09:12.046 31518-31547/co.foo.bar D/MainActivity: Throwing bg task exception

--------- beginning of crash
07-26 15:09:12.046 31518-31547/co.foo.bar E/AndroidRuntime: FATAL EXCEPTION: ForkJoinPool.commonPool-worker-1
Process: co.foo.bar, PID: 31518
                           java.lang.RuntimeException: Bg task exception
                           at co.foo.barsample.MainActivity$onCreate$1$job$1$1.doResume(MainActivity.kt:36)
                           at kotlin.coroutines.experimental.jvm.internal.CoroutineImpl.resume(CoroutineImpl.kt:54)
                           at kotlinx.coroutines.experimental.DispatchTask.run(CoroutineDispatcher.kt:120)
                           at java.util.concurrent.ForkJoinTask$RunnableExecuteAction.exec(ForkJoinTask.java:1383)
                           at java.util.concurrent.ForkJoinTask.doExec(ForkJoinTask.java:256)
                           at java.util.concurrent.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:1123)
                           at java.util.concurrent.ForkJoinPool.scan(ForkJoinPool.java:1961)
                           at java.util.concurrent.ForkJoinPool.runWorker(ForkJoinPool.java:1909)
                           at java.util.concurrent.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:128)
07-26 15:09:12.050 1705-2190/system_process W/ActivityManager:   Force finishing activity co.foo.bar/co.foo.barsample.MainActivity

似乎调用cancel()会抛出CancellationException，并且成功捕获。但是后续的RuntimeException没有被捕获。我想取消作业后的以下异常应该被库忽略？或者如何在不抛出CancellationException异常的情况下静默取消作业？

- Oleksii Masnyi

---

你能否在 launch(){} 周围加上 try-catch？ - voddan

@voddan 不，launch 立即执行。但我找到了一个解决方案。 - Oleksii Masnyi

Answer 1

使用CoroutineExceptionHandler作为额外的协程上下文来处理launch或run中的异常。

run(CommonPool + CoroutineExceptionHandler({ _, e ->
   Log.e("TAG", "CoroutineExceptionHandler", e)
})) {
    ...
}

或者

launch(UI + CoroutineExceptionHandler({ _, e ->
   Log.e("TAG", "CoroutineExceptionHandler", e)
})) {
   ...
}

- Oleksii Masnyi

---

谢谢提供信息。我们是否应该在代码中的每个“launch”或“run”使用中都这样做？在生产代码中，我们通常希望异常上升，以便Crashlytics或其他记录器可以捕获它们。有没有更简洁的方法来实现这一点？ - Pablo

我想问一下，我们是否可以将“launch”函数替换为异步函数以同样的方式捕获异常？我已经尝试过并且失败了。至少我可以使用“withContext”函数来使用你的异常处理方法。 - Eric Ho