Thread、Future、Promise、Packaged_task、Async之间有什么关系?

一个promise的set_value()只能调用一次,如果调用多次,就会throw exception,如果外部没catch exception,程序就会crash。
首页 新闻资讯 行业资讯 Thread、Future、Promise、Packaged_task、Async之间有什么关系?

并发编程一般指多线程编程,C++11之后关于多线程编程有几个高级API:

  • std::thread

  • std::future

  • std::shared_future

  • std::promise

  • std::packaged_task

  • std::async

可能很多人都搞不清楚它们之前有什么联系,可以直接看这张图:

图片

从这张图我们可以大体看出来:

  • packaged_task ≈ promise + function

  • async ≈ thread + packaged_task

  • 通过promise的get_future()可拿到future

  • 通过future的share()可拿到shared_future


promise和future是线程之间的同步通道,类似于条件变量的封装,看它的使用:

复制

#include <future>#include <iostream>#include <thread>int main() {
 std::promise<bool> prom;
 std::future<bool> f = prom.get_future();
 prom.set_value(true);
 std::cout << f.get() << std::endl;}
  • 1.

  • 2.

  • 3.

  • 4.

  • 5.

  • 6.

  • 7.

  • 8.

  • 9.

  • 10.


首先创建一个promise,通过promise可以拿到future,future有wait()和get()等方法,这种方法会阻塞当前线程,直到future的源promise调用了set_value,future的wait()只有阻塞功能,而get()方法不仅有阻塞功能,还能拿到set_value()设置的值。我举个多线程的示例:

复制

#include <future>#include <iostream>#include <thread>int main() {
 std::promise<int> prom;
 auto f = prom.get_future();
 std::thread t(
   [](std::promise<int> p) {std::this_thread::sleep_for(std::chrono::seconds(2));p.set_value(100);
   },
   std::move(prom));
 std::cout << f.get() << std::endl;
 if (t.joinable()) t.join();}
  • 1.

  • 2.

  • 3.

  • 4.

  • 5.

  • 6.

  • 7.

  • 8.

  • 9.

  • 10.

  • 11.

  • 12.

  • 13.

  • 14.

  • 15.

  • 16.


这段代码执行后会在两秒后输出100。这个结果就验证了上面啰嗦的promise的future的get()的阻塞和获取结果的能力。


注意:一个promise的set_value()只能调用一次,如果调用多次,就会throw exception,如果外部没catch exception,程序就会crash。


promise的阻塞功能还是蛮好用的,我在工程中就经常用到它。


介绍完promise,再来看看packaged_task:

复制

#include <future>#include <iostream>#include <thread>int main() {
 std::packaged_task<int(int, int)> task([](int a, int b) { return a + b; });
 auto f = task.get_future();
 std::thread t(std::move(task), 1, 2);
 std::cout << f.get() << std::endl;
 if (t.joinable()) t.join();}
  • 1.

  • 2.

  • 3.

  • 4.

  • 5.

  • 6.

  • 7.

  • 8.

  • 9.

  • 10.


可以拿这段代码和上面那段promise的代码对比看看,可以得出结论:packaged_task ≈ promise + function

promise只能set_value,不太好执行复杂的逻辑,有执行函数+阻塞的需求时,就可以考虑使用packaged_task。

可以思考一下,如果要你封装一个packaged_task,你会怎么做?


再看async:

复制

#include <future>#include <iostream>#include <thread>int main() {
 auto f = std::async(
   std::launch::async, [](int a, int b) { return a + b; }, 1, 2);
 std::cout << f.get() << std::endl;}
  • 1.

  • 2.

  • 3.

  • 4.

  • 5.

  • 6.

  • 7.

  • 8.

这里可以看到,使用了async后,连thread都不需要创建了,这也就验证了上面图中的结论:async ≈ thread + packaged_task


这里请注意:async中的第一个参数我使用的是std::launch::async,只有当参数为std::launch::async时,函数才会异步执行。

参数还可以是std::launch::deferred,参数为这个时,函数不会异步执行,只有当对应的future调用了get时,函数才会执行,而且是在当前线程执行。

关于async有几个坑,我之前写过一篇文章,可以看这个:async的两个坑。


介绍完async,再介绍下shared_future。

普通的future有个特点,它不能拷贝,只能移动,这就意味着只能有一个线程一个实例可以通过get()拿到对应的结果。

如果想要多个线程多个实例拿到结果,就可以使用shared_future,那怎么拿到shared_future,可以通过普通future的shared()方法。

复制

#include <future>#include <iostream>#include <thread>int main() {
 std::promise<int> prom;
 auto fu = prom.get_future();
 auto shared_fu = fu.share();
 auto f1 = std::async(std::launch::async, [shared_fu]() { std::cout << shared_fu.get() << std::endl; });
 auto f2 = std::async(std::launch::async, [shared_fu]() { std::cout << shared_fu.get() << std::endl; });
 prom.set_value(102);
 f1.get();
 f2.get();}
  • 1.

  • 2.

  • 3.

  • 4.

  • 5.

  • 6.

  • 7.

  • 8.

  • 9.

  • 10.

  • 11.

  • 12.

  • 13.

看到这里,大家应该明白thread、future、promise、packaged_task、async之间的关系了吧。


30    2022-07-06 08:36:34    thread promise