如何将可变参数传递给 std::thread？

Question

我想通过包装 C++11 中的 std::thread 类来使用我自己的 Thread 实现，这样我就可以处理我想要的异常。

这是我的包装类：

#include <Types.hpp>
#include <thread>
#include <exception>
#include <functional>

class Thread
{
    private:

        std::exception_ptr exceptionPtr;
        std::thread thread;

    public:

        using Id = std::thread::id;

        using NativeHandleType = std::thread::native_handle_type;

        Thread() noexcept = default;
        Thread(Thread &&t) noexcept :
            exceptionPtr(std::move(t.exceptionPtr)),
            thread(std::move(t.thread))
        {
        }

        Thread &operator =(Thread &&t) noexcept
        {
            exceptionPtr = std::move(t.exceptionPtr);
            thread = std::move(t.thread);
            return *this;
        }

        template<typename Callable, typename... Args>
        Thread(Callable &&f, Args &&... args) :
            exceptionPtr(nullptr),
            thread([&](Callable &&f, Args &&... args)
            {
                try
                {
                    std::once_flag flag;
                    std::call_once(flag, f, args...);
                }
                catch (...)
                {
                    exceptionPtr = std::current_exception();
                }

            }, f, args...)
        {
            if (exceptionPtr != nullptr)
            {
                 std::rethrow_exception(exceptionPtr);
            }
        }

        bool joinable() const noexcept
        {
            return thread.joinable();
        }

        void join()
        {
            thread.join();
        }

        void detach()
        {
            thread.detach();
        }

        Id getId() const noexcept
        {
            return thread.get_id();
        }

        NativeHandleType nativeHandle()
        {
            return thread.native_handle();
        }

        static uint32_t hardwareConcurrency() noexcept
        {
            return std::thread::hardware_concurrency();
        }

        static void wait(Time t)
        {
            std::this_thread::sleep_for(t);
        }
};

如果没有参数，效果很好：

Thread([&]() {  /* do something */ }).detach();

...但是如果我尝试传递可变参数：

Thread(&GUI::refreshTask, this, refreshDelay).detach();

...编译时出现错误：

buildroot-2014.02/output/host/usr/i586-buildroot-linux-uclibc/include/c++/4.8.2/functional: 在 'struct std::_Bind_simple)(std::chrono::duration > ); Args = {CRH::GUI const, std::chrono::duration >&}]::__lambda1(void (CRH::GUI::)(std::chrono::duration >), CRH:: GUI，std::chrono::duration >)>': buildroot-2014.02/output/host/usr/i586-buildroot-linux-uclibc/include/c++/4.8.2/thread:137:47: 来自 'std::thread::thread(_Callable&&, _Args&& ...) [with _Callable = CRH::Thread::Thread(Callable&&, Args&& ...) [with Callable = void (CRH::GUI::)(std::chrono::duration >); Args = {CRH::GUI const, std::chrono::duration >&}]::__lambda1; _Args = {void (CRH::GUI::&)(std::chrono::duration >), CRH::GUI const&, std::chrono::duration >&}]' /home/cyril/Documents/crh-2016/src/robot2/../core/Thread.hpp:72:30: 'CRH::Thread::Thread(Callable&&, Args&& ...) [with Callable =无效 (CRH::GUI::)(std::chrono::duration >); Args = {CRH::GUI const, std::chrono::duration >&}]' src/core/GUI.cpp:90:57：从这里需要 buildroot-2014.02/output/host/usr/i586-buildroot-linux-uclibc/include/c++/4.8.2/functional:1697:61: 错误：'class std::result_of)(std ::chrono::duration >); Args = {CRH::GUI const, std::chrono::duration >&}]::__lambda1(void (CRH::GUI::)(std::chrono::duration >), CRH:: GUI，std::chrono::duration >)>' typedef typename result_of<_callable>::type result_type; ^ buildroot-2014.02/output/host/usr/i586-buildroot-linux-uclibc/include/c++/4.8.2/functional:1727:9: 错误：'class std::result_of)(std ::chrono::duration >); Args = {CRH::GUI const, std::chrono::duration >&}]::__lambda1(void (CRH::GUI::)(std::chrono::duration >), CRH:: GUI，std::chrono::duration >)>' _M_invoke(_Index_tuple<_indices...>)

可能会更清楚一点……但对 GCC 来说要求太高了。

知道如何解决这个问题吗？

解决方案

#include <Types.hpp>
#include <thread>
#include <exception>
#include <functional>

class Thread
{
    private:

        std::exception_ptr exceptionPtr;
        std::thread thread;

    public:

        using Id = std::thread::id;

        using NativeHandleType = std::thread::native_handle_type;

        Thread() noexcept = default;
        Thread(Thread &&t) noexcept :
            exceptionPtr(std::move(t.exceptionPtr)),
            thread(std::move(t.thread))
        {
        }

        Thread &operator =(Thread &&t) noexcept
        {
            exceptionPtr = std::move(t.exceptionPtr);
            thread = std::move(t.thread);
            return *this;
        }

        template<typename Callable, typename... Args>
        Thread(Callable &&f, Args &&... args) :
            exceptionPtr(nullptr),
            thread([&](typename std::decay<Callable>::type &&f, typename std::decay<Args>::type &&... args)
            {
                try
                {
                    std::bind(f, args...)();
                }
                catch (...)
                {
                    exceptionPtr = std::current_exception();
                }

            }, std::forward<Callable>(f), std::forward<Args>(args)...)
        {
        }

        bool joinable() const noexcept
        {
            return thread.joinable();
        }

        void join()
        {
            thread.join();

            if (exceptionPtr != nullptr)
            {
                std::rethrow_exception(exceptionPtr);
            }
        }

        void detach()
        {
            thread.detach();
        }

        Id getId() const noexcept
        {
            return thread.get_id();
        }

        NativeHandleType nativeHandle()
        {
            return thread.native_handle();
        }

        static uint32_t hardwareConcurrency() noexcept
        {
            return std::thread::hardware_concurrency();
        }

        static void wait(Time t)
        {
            std::this_thread::sleep_for(t);
        }
};

Answer 1

Callable 和 Args 是转发引用，因此模板参数推导可以使它们成为左值引用或普通类型，具体取决于参数表达式的值类别。

这意味着当您在 lambda 声明中重用推导的类型时：

thread([&](Callable&& f, Args&&... args)

引用折叠开始发挥作用，对于左值参数refreshDelay，Args 成为左值引用。

但是，std::thread 存储它接收到的参数的衰减副本，然后它从其内部存储移动到实际的处理程序，将存储的对象转换为 xvalues。这就是错误告诉您的内容：处理程序不能使用线程尝试传入的参数调用。

相反，您可以按如下方式实现它：

template <typename Callable, typename... Args>
Thread(Callable&& f, Args&&... args)
    : exceptionPtr(nullptr)
    , thread([] (typename std::decay<Callable>::type&& f
               , typename std::decay<Args>::type&&... args)
            {
                // (...)
            }
            , std::forward<Callable>(f), std::forward<Args>(args)...)
{
    // (...)
}