Question

有人可以解释为什么即使到达流的末尾，以下AsObservable方法也会创建无限循环吗？

public static class StreamExt {
    public static IObservable<byte> AsObservable(this Stream stream, int bufferSize) {
        return Observable
            .FromAsync(cancel => stream.ReadBytes(bufferSize, cancel))
            .Repeat()
            .TakeWhile(bytes => bytes != null) // EndOfStream
            .SelectMany(bytes => bytes);
    }

    private static async Task<byte[]> ReadBytes(this Stream stream, int bufferSize, CancellationToken cancel) {
        var buf = new byte[bufferSize];
        var bytesRead = await stream
            .ReadAsync(buf, 0, bufferSize, cancel)
            .ConfigureAwait(false);

        if (bytesRead < 1) return null; // EndOfStream
        var result_size = Math.Min(bytesRead, bufferSize);
        Array.Resize(ref buf, result_size);
        return buf;
    }
}

快速测试显示它产生无限循环：

class Program {
    static void Main(string[] args) {
        using (var stream = new MemoryStream(new byte[] { 1, 2, 3 })) {
            var testResult = stream
                .AsObservable(1024)
                .ToEnumerable()
                .ToArray();
            Console.WriteLine(testResult.Length);
        }
    }
}

当然我可以添加.SubscribeOn(TaskPoolScheduler.Default)但是，无限循环保持活动状态（阻止任务池调度程序+从Stream无限读取）。

[更新2017-05-09]

Shlomo发布了一个更好的例子来重现这个问题：

int i = 0;
var testResult = Observable.FromAsync(() => Task.FromResult(i++))
    .Repeat()
    .TakeWhile(l => l < 3);
testResult.Subscribe(b => Console.WriteLine(b), e => { }, () => Console.WriteLine("OnCompleted"));
Console.WriteLine("This is never printed.");

Answer 1

您可以使用以下方式确认正确生成OnCompleted：

using (var stream = new MemoryStream(new byte[] { 1, 2, 3 }))
{
    var testResult = stream
        .AsObservable(1024)
        ;
    testResult.Subscribe(b => Console.WriteLine(b), e => {}, () => Console.WriteLine("OnCompleted"));
}

看起来.FromAsync + .Repeat组合存在问题。以下代码的行为类似：

int i = 0;
var testResult = Observable.FromAsync(() => Task.FromResult(i++))
    .Repeat()
    .TakeWhile(l => l < 3)
    ;
testResult.Subscribe(b => Console.WriteLine(b), e => { }, () => Console.WriteLine("OnCompleted"));
Console.WriteLine("This is never printed.");

...而这段代码正确终止：

var testResult = Observable.Generate(0, i => true, i => i + 1, i => i)
    .Repeat()
    .TakeWhile(l => l < 3)
    ;
testResult.Subscribe(b => Console.WriteLine(b), e => { }, () => Console.WriteLine("OnCompleted"));
Console.WriteLine("This is printed.");

Answer 2

对于那些最终需要答案的人，不仅仅是一个解释：问题似乎是FromAsync的默认调度程序，如this self-answered question所示。如果您调整为“当前线程”，则调度程序Repeat().TakeWhile(...)的行为将更加可预测。例如。（摘录自问题）：

.FromAsync(cancel => stream.ReadBytes(bufferSize, cancel), 
    System.Reactive.Concurrency.Scheduler.CurrentThread)
.Repeat()
.TakeWhile(bytes => bytes != null) // EndOfStream

Answer 3

Shane Neuville发布了包含此行为解释的链接：

Observable.FromAsync + Repeat + TakeWhile的组合创建无限循环

3 个答案: