如何检测处理程序是否为 ASIO strand wrap 并通过 strand 调用它?
How to detect if handler is an ASIO strand wrap and call it through the strand?
如果有一个通用方法采用一些处理程序:
template< typename HandlerType >
void Register( HandlerType && handler )
{
m_handler( std::forward< HandlerType >( handler ) );
}
并且该处理程序将在将来的某个时候通过 io_service
调用:
void SomeEvent( )
{
// compute someParameter
m_IOService.post( std::bind( m_handler , someParameter ) );
}
如果 Register()
的调用者传递了由 strand
包裹的东西,如何检测到,如:
m_strand( m_IOService );
// ...
Register( m_strand.wrap( []( /*something*/ ){ /*...*/ } ) );
在这种情况下,应该如何更改 SomeEvent()
以便 post 处理程序通过 strand?
编辑
当我问这个问题时,我没有仔细阅读 io_service::strand::wrap
docs,更具体地说:
(...) Given a function object with the signature:
R f(A1 a1, ... An an);
If this function object is passed to the wrap function like so:
strand.wrap(f);
then the return value is a function object with the signature
void g(A1 a1, ... An an);
that, when invoked, executes code equivalent to:
strand.dispatch(boost::bind(f, a1, ... an));
我确实需要的就是这个——我可以将 m_handler
声明为适当的 std::function<>
并通过 SomeEvent()
中的 io_service
简单地 post ].
看了@Arunmu 的回答我才意识到这一点,所以我接受了。尽管如此,@Richard Hodges 的回答对 ASIO 的执行程序逻辑以及它在独立版本中的改进方式有一些好处。
如果我理解清楚你的需求,那么如果像下面这样实现你就不需要做任何额外的事情(阅读代码中的注释以获得解释):
#include <iostream>
#include <type_traits>
#include <thread>
#include <memory>
#include <asio.hpp>
template <typename Handler>
class GenHandler
{
public:
GenHandler(Handler&& h): hndler_(std::forward<Handler>(h))
{}
template <typename... Args>
void operator()(Args&&... args)
{
std::cout << "GenHandler called" << std::endl;
hndler_();
}
private:
Handler hndler_;
};
template<typename HandlerType>
GenHandler<std::decay_t<HandlerType>> create_handler(HandlerType&& h)
{
return {std::forward<HandlerType>(h)};
}
template <typename Handler>
void SomeEvent(asio::io_service& ios, Handler& h)
{
ios.post([=] ()mutable { h(); });
}
int main() {
asio::io_service ios;
asio::io_service::strand strand{ios};
auto work = std::make_unique<asio::io_service::work>(ios);
std::thread t([&]() { ios.run(); });
// This creates a regular handler which when called by the
// io_context would first execute GenHandler::operator()
// and inside of which it would call the lambda passed below.
auto hndl = create_handler([] {
std::cout << "Regular Handle" << std::endl;
});
SomeEvent(ios, hndl);
///-------- Example 2 ---------////
// This creates a handler just like above, but instead wraps a
// strand handler i.e when GenHandler::operator() gets called
// it will execute the lambda passed to the wrap in the execution context
// of the strand.
auto hndl2 = create_handler(
strand.wrap([] {
std::cout << "Strand handler-depth 2" << std::endl;
}));
// This is a regular strand wrap which is passed to the
// io_service execution context. The lambda passed in the strand::wrap
// would be excuted the execution context of the strand.
auto str_handler = strand.wrap([=]() mutable {
std::cout <<"strand\n";
hndl2();
});
SomeEvent(ios, str_handler);
work.reset();
t.join();
return 0;
}
在第二个示例中,处理程序的调用顺序如下:
io_service
传递给 strand::wrapped_handler
。因此,wrapped_handler
持有的处理程序在链内执行。
hndl2
即 GenHandler
持有另一个 strand::wrapped_handler
也被称为链内。
- 当调用
GenHandler::operator()
时,它也会执行持有的strand::wrapped_handler
。这是通过将 strand::wrapped_handler
持有的内部处理程序分派给链来完成的。
注意: 由于我不清楚的原因,strand::wrap
已被弃用。作者希望人们改用 bind_executor
。
对于 boost asio 我认为答案在这个模板函数中:
namespace boost_asio_handler_cont_helpers {
template <typename Context>
inline bool is_continuation(Context& context)
{
#if !defined(BOOST_ASIO_HAS_HANDLER_HOOKS)
return false;
#else
using boost::asio::asio_handler_is_continuation;
return asio_handler_is_continuation(
boost::asio::detail::addressof(context));
#endif
}
} // namespace boost_asio_handler_cont_helpers
如果我没有看错的话,它是用来检测是否有"context"(即strand
或io_service
)在其中执行处理程序。
reactor 服务中的代码然后根据结果切换,是否在现有上下文中执行。
在 standalone asio 中有些变化。
现在有一个函数可以检测处理程序的上下文(如果有的话)。咨询了作者后写了这段代码
相关行是:
auto ex = asio::get_associated_executor(handler, this->get_io_service().get_executor());
和..
asio::dispatch(ex, [handler = std::move(handler), future = std::move(future)]() mutable
{
// call the user-supplied handler
});
这是来自 "long running task" 执行服务的生产代码:
template<class Task, class Handler>
void async_execute(implementation& impl, Task&& task, Handler&& handler)
{
VALUE_DEBUG_TRACE(module) << method(__func__, this);
using task_type = std::decay_t<Task>;
static_assert(is_callable_t<task_type, long_running_task_context>(), "");
using result_type = std::result_of_t<task_type(long_running_task_context)>;
using promise_type = std::promise<result_type>;
using future_type = std::future<result_type>;
using handler_type = std::decay_t<Handler>;
static_assert(is_callable_t<handler_type, future_type>(), "");
using handler_result_type = std::result_of<handler_type(future_type)>;
auto ex = asio::get_associated_executor(handler, this->get_io_service().get_executor());
if (not impl)
{
post(ex, [handler = std::forward<Handler>(handler)]() mutable
{
promise_type promise;
promise.set_exception(std::make_exception_ptr(system_error(errors::null_handle)));
handler(promise.get_future());
});
return;
}
auto handler_work = make_work(ex);
auto& ios = get_io_service();
auto impl_ptr = impl.get();
auto async_handler = [this,
&ios,
impl_ptr,
handler_work, ex,
handler = std::forward<Handler>(handler)]
(detail::long_running_task_op::identifier ident,
auto future) mutable
{
assert(impl_ptr);
VALUE_DEBUG_TRACE(module) << method("async_execute::async_handler", this, ident);
asio::dispatch(ex, [handler = std::move(handler), future = std::move(future)]() mutable
{
VALUE_DEBUG_TRACE(module) << method("async_execute::completion_handler");
handler(std::move(future));
});
assert(impl_ptr);
impl_ptr->remove_op(ident);
};
using async_handler_type = decltype(async_handler);
static_assert(is_callable_t<async_handler_type, detail::long_running_task_op::identifier, future_type>(), "");
auto op = detail::long_running_task_op(std::forward<Task>(task), std::move(async_handler));
auto ident = op.get_identifier();
impl->add_op(ident);
auto lock = lock_type(this->_queue_mutex);
_ops.emplace(ident, op);
lock.unlock();
this->post_execute();
}
如果有一个通用方法采用一些处理程序:
template< typename HandlerType >
void Register( HandlerType && handler )
{
m_handler( std::forward< HandlerType >( handler ) );
}
并且该处理程序将在将来的某个时候通过 io_service
调用:
void SomeEvent( )
{
// compute someParameter
m_IOService.post( std::bind( m_handler , someParameter ) );
}
如果 Register()
的调用者传递了由 strand
包裹的东西,如何检测到,如:
m_strand( m_IOService );
// ...
Register( m_strand.wrap( []( /*something*/ ){ /*...*/ } ) );
在这种情况下,应该如何更改 SomeEvent()
以便 post 处理程序通过 strand?
编辑
当我问这个问题时,我没有仔细阅读 io_service::strand::wrap
docs,更具体地说:
(...) Given a function object with the signature:
R f(A1 a1, ... An an);
If this function object is passed to the wrap function like so:
strand.wrap(f);
then the return value is a function object with the signature
void g(A1 a1, ... An an);
that, when invoked, executes code equivalent to:
strand.dispatch(boost::bind(f, a1, ... an));
我确实需要的就是这个——我可以将 m_handler
声明为适当的 std::function<>
并通过 SomeEvent()
中的 io_service
简单地 post ].
看了@Arunmu 的回答我才意识到这一点,所以我接受了。尽管如此,@Richard Hodges 的回答对 ASIO 的执行程序逻辑以及它在独立版本中的改进方式有一些好处。
如果我理解清楚你的需求,那么如果像下面这样实现你就不需要做任何额外的事情(阅读代码中的注释以获得解释):
#include <iostream>
#include <type_traits>
#include <thread>
#include <memory>
#include <asio.hpp>
template <typename Handler>
class GenHandler
{
public:
GenHandler(Handler&& h): hndler_(std::forward<Handler>(h))
{}
template <typename... Args>
void operator()(Args&&... args)
{
std::cout << "GenHandler called" << std::endl;
hndler_();
}
private:
Handler hndler_;
};
template<typename HandlerType>
GenHandler<std::decay_t<HandlerType>> create_handler(HandlerType&& h)
{
return {std::forward<HandlerType>(h)};
}
template <typename Handler>
void SomeEvent(asio::io_service& ios, Handler& h)
{
ios.post([=] ()mutable { h(); });
}
int main() {
asio::io_service ios;
asio::io_service::strand strand{ios};
auto work = std::make_unique<asio::io_service::work>(ios);
std::thread t([&]() { ios.run(); });
// This creates a regular handler which when called by the
// io_context would first execute GenHandler::operator()
// and inside of which it would call the lambda passed below.
auto hndl = create_handler([] {
std::cout << "Regular Handle" << std::endl;
});
SomeEvent(ios, hndl);
///-------- Example 2 ---------////
// This creates a handler just like above, but instead wraps a
// strand handler i.e when GenHandler::operator() gets called
// it will execute the lambda passed to the wrap in the execution context
// of the strand.
auto hndl2 = create_handler(
strand.wrap([] {
std::cout << "Strand handler-depth 2" << std::endl;
}));
// This is a regular strand wrap which is passed to the
// io_service execution context. The lambda passed in the strand::wrap
// would be excuted the execution context of the strand.
auto str_handler = strand.wrap([=]() mutable {
std::cout <<"strand\n";
hndl2();
});
SomeEvent(ios, str_handler);
work.reset();
t.join();
return 0;
}
在第二个示例中,处理程序的调用顺序如下:
io_service
传递给strand::wrapped_handler
。因此,wrapped_handler
持有的处理程序在链内执行。hndl2
即GenHandler
持有另一个strand::wrapped_handler
也被称为链内。- 当调用
GenHandler::operator()
时,它也会执行持有的strand::wrapped_handler
。这是通过将strand::wrapped_handler
持有的内部处理程序分派给链来完成的。
注意: 由于我不清楚的原因,strand::wrap
已被弃用。作者希望人们改用 bind_executor
。
对于 boost asio 我认为答案在这个模板函数中:
namespace boost_asio_handler_cont_helpers {
template <typename Context>
inline bool is_continuation(Context& context)
{
#if !defined(BOOST_ASIO_HAS_HANDLER_HOOKS)
return false;
#else
using boost::asio::asio_handler_is_continuation;
return asio_handler_is_continuation(
boost::asio::detail::addressof(context));
#endif
}
} // namespace boost_asio_handler_cont_helpers
如果我没有看错的话,它是用来检测是否有"context"(即strand
或io_service
)在其中执行处理程序。
reactor 服务中的代码然后根据结果切换,是否在现有上下文中执行。
在 standalone asio 中有些变化。
现在有一个函数可以检测处理程序的上下文(如果有的话)。咨询了作者后写了这段代码
相关行是:
auto ex = asio::get_associated_executor(handler, this->get_io_service().get_executor());
和..
asio::dispatch(ex, [handler = std::move(handler), future = std::move(future)]() mutable
{
// call the user-supplied handler
});
这是来自 "long running task" 执行服务的生产代码:
template<class Task, class Handler>
void async_execute(implementation& impl, Task&& task, Handler&& handler)
{
VALUE_DEBUG_TRACE(module) << method(__func__, this);
using task_type = std::decay_t<Task>;
static_assert(is_callable_t<task_type, long_running_task_context>(), "");
using result_type = std::result_of_t<task_type(long_running_task_context)>;
using promise_type = std::promise<result_type>;
using future_type = std::future<result_type>;
using handler_type = std::decay_t<Handler>;
static_assert(is_callable_t<handler_type, future_type>(), "");
using handler_result_type = std::result_of<handler_type(future_type)>;
auto ex = asio::get_associated_executor(handler, this->get_io_service().get_executor());
if (not impl)
{
post(ex, [handler = std::forward<Handler>(handler)]() mutable
{
promise_type promise;
promise.set_exception(std::make_exception_ptr(system_error(errors::null_handle)));
handler(promise.get_future());
});
return;
}
auto handler_work = make_work(ex);
auto& ios = get_io_service();
auto impl_ptr = impl.get();
auto async_handler = [this,
&ios,
impl_ptr,
handler_work, ex,
handler = std::forward<Handler>(handler)]
(detail::long_running_task_op::identifier ident,
auto future) mutable
{
assert(impl_ptr);
VALUE_DEBUG_TRACE(module) << method("async_execute::async_handler", this, ident);
asio::dispatch(ex, [handler = std::move(handler), future = std::move(future)]() mutable
{
VALUE_DEBUG_TRACE(module) << method("async_execute::completion_handler");
handler(std::move(future));
});
assert(impl_ptr);
impl_ptr->remove_op(ident);
};
using async_handler_type = decltype(async_handler);
static_assert(is_callable_t<async_handler_type, detail::long_running_task_op::identifier, future_type>(), "");
auto op = detail::long_running_task_op(std::forward<Task>(task), std::move(async_handler));
auto ident = op.get_identifier();
impl->add_op(ident);
auto lock = lock_type(this->_queue_mutex);
_ops.emplace(ident, op);
lock.unlock();
this->post_execute();
}