d结构并发-CFANZ编程社区

原文工具:

core.Thread
std.concurrency.spawn
std.parallelism.task / taskPool
vibe.d.runTask
etc.

代码:

auto fun() {
    return spawn(...);
}

void gun() {
    auto task = fun();
    // ...
    join(task);
}
//
auto fun() {
    return new ...;
}

void gun() {
    auto task = fun();
    // ...
    delete task;
}

这不是,结构化的并发.

所有权和生命期

每个异步计算都需要有个所有者 所有者需要比它拥有的所有异步计算生命更长
错误必须向上传播到所有者链

异步计算应该支持取消异步计算需要一种方法来:
表示正常完成(返回值)
表示错误(抛出异常)
表示取消

标准C++异步模型,基于三个关键抽象:调度器,发送器和接收器,及一组可定制的异步算法.

发送者

/// 发送T类型单值的发送器
struct ValueSender(T) {
    alias Value = T;
    T value;
    static struct OperationalState(Receiver) {
        Receiver receiver;
        T value;
        void start() @safe nothrow {
            receiver.setValue(value); // 用返回值完成
        }
    }
    auto connect(Receiver)(return Receiver receiver) @safe scope return {
        // 确保 NVRO
        auto op = OperationalState!(Receiver)(receiver, value);
        return op;
    }
}

发送者与接受者,通过另一线程:

auto just(T t) @safe {
    return ValueSender!T(t);
}

void main() @safe {
    auto w = just(42).via(ThreadSender).syncWait();
    assert(w.value == 42);
}

调度器:

void main() @safe {
    auto pool = stdTaskPool(2);

    auto w = just(42)
        .on(pool.getScheduler)
        .syncWait();

    assert(w.value == 42);
}
//当所有
auto fooSender() @safe { /* ... */ }
auto barSender() @safe { /* ... */ }

void foobar() @safe {
    whenAll(fooSender, barSender)
        .syncWait(); // what ought to happen if either `fooSender` or `barSender` fails?
}

重试算法:

auto fooSender() @safe { /* ... */ }
auto barSender() @safe { /* ... */ }

auto foobar() @safe {
    return whenAll(fooSender.retry(MaxTimes(3)), barSender);
}

void fun() @safe {
    foobar().syncWait(); // What ought to happen if `barSender` fails?
}

竞争算法:

auto fooSender() @safe { /* ... */ }
auto barSender() @safe { /* ... */ }

auto foobar() @safe {
    return whenAll(fooSender.retry(MaxTimes(3)), barSender);
}

void fun() @safe {
    foobar().timeout(10.secs).syncWait();
}

auto timeout(Sender)(Sender s, Duration d) @safe {
    return race(s, delay(duration));
}

流:

void main() @safe {
    intervalStream(1.secs)
        .collect(() => writeln("Hello World!"))
        .syncWait();
}
//
void main() @safe {
    auto w = intervalStream(1.secs, true)
        .scan((int i) => i + 1, 0)
        .filter((int i) => i % 2)
        .take(5)
        .toList()
        .syncWait();

    assert(w.value == [0, 2, 4, 6, 8]);
}

多生产单消费队列示例:

final class MPSCQueue(Node) {
    // [...]

    @safe nothrow @nogc shared
    bool push(Node* n) { */ ... */ }

    @safe nothrow @nogc 
    Node* pop() { /* ... */ }

    alias Producer = shared(MPSCQueueProducer!Node);

    @trusted nothrow @nogc 
    Producer producer() {
      return Producer(cast(shared)this);
    }

    // [...]
}

但,生产者是共享的.

struct MPSCQueueProducer(Node) {
    private shared MPSCQueue!(Node)*  q;

    @safe nothrow @nogc shared
    void push(Node* node) {
        q.push(node);
    }
}

rest示例:

struct Application {
    @Path("/") shared @safe
    auto getXyz() { /* ... */ }
}

void main() @safe {
    auto pool = stdTaskPool(16);
    auto app = shared Application();
    auto listener = HttpListener("0.0.0.0", 8080);
    listener
        .handleRequests(app)
        .withScheduler(pool.getScheduler())
        .syncWait().assumeOk;
}

序化器:

auto mySerializer = new Serializer(1);

auto fun() @safe {
    return sequence(
        step1(),
        step2().on(mySerializer),
        step3()
    );
}