asio/example/cpp14/operations/composed_6.cpp

//
// composed_6.cpp
// ~~~~~~~~~~~~~~
//
// Copyright (c) 2003-2019 Christopher M. Kohlhoff (chris at kohlhoff dot com)
//
// Distributed under the Boost Software License, Version 1.0. (See accompanying
// file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
//

#include <boost/asio/executor_work_guard.hpp>
#include <boost/asio/io_context.hpp>
#include <boost/asio/ip/tcp.hpp>
#include <boost/asio/steady_timer.hpp>
#include <boost/asio/use_future.hpp>
#include <boost/asio/write.hpp>
#include <functional>
#include <iostream>
#include <memory>
#include <sstream>
#include <string>
#include <type_traits>
#include <utility>

using boost::asio::ip::tcp;

// NOTE: This example requires the new boost::asio::async_initiate function. For
// an example that works with the Networking TS style of completion tokens,
// please see an older version of asio.

//------------------------------------------------------------------------------

// This composed operation shows composition of multiple underlying operations.
// It automatically serialises a message, using its I/O streams insertion
// operator, before sending it N times on the socket. To do this, it must
// allocate a buffer for the encoded message and ensure this buffer's validity
// until all underlying async_write operation complete. A one second delay is
// inserted prior to each write operation, using a steady_timer.

template <typename T, typename CompletionToken>
auto async_write_messages(tcp::socket& socket,
    const T& message, std::size_t repeat_count,
    CompletionToken&& token)
  // The return type of the initiating function is deduced from the combination
  // of CompletionToken type and the completion handler's signature. When the
  // completion token is a simple callback, the return type is always void.
  // In this example, when the completion token is boost::asio::yield_context
  // (used for stackful coroutines) the return type would be also be void, as
  // there is no non-error argument to the completion handler. When the
  // completion token is boost::asio::use_future it would be std::future<void>.
  //
  // In C++14 we can omit the return type as it is automatically deduced from
  // the return type of boost::asio::async_initiate.
{
  // In addition to determining the mechanism by which an asynchronous
  // operation delivers its result, a completion token also determines the time
  // when the operation commences. For example, when the completion token is a
  // simple callback the operation commences before the initiating function
  // returns. However, if the completion token's delivery mechanism uses a
  // future, we might instead want to defer initiation of the operation until
  // the returned future object is waited upon.
  //
  // To enable this, when implementing an asynchronous operation we must
  // package the initiation step as a function object. The initiation function
  // object's call operator is passed the concrete completion handler produced
  // by the completion token. This completion handler matches the asynchronous
  // operation's completion handler signature, which in this example is:
  //
  //   void(boost::system::error_code error)
  //
  // The initiation function object also receives any additional arguments
  // required to start the operation. (Note: We could have instead passed these
  // arguments in the lambda capture set. However, we should prefer to
  // propagate them as function call arguments as this allows the completion
  // token to optimise how they are passed. For example, a lazy future which
  // defers initiation would need to make a decay-copy of the arguments, but
  // when using a simple callback the arguments can be trivially forwarded
  // straight through.)
  auto initiation = [](auto&& completion_handler, tcp::socket& socket,
      std::unique_ptr<std::string> encoded_message, std::size_t repeat_count,
      std::unique_ptr<boost::asio::steady_timer> delay_timer)
  {
    // In this example, the composed operation's intermediate completion
    // handler is implemented as a hand-crafted function object.
    struct intermediate_completion_handler
    {
      // The intermediate completion handler holds a reference to the socket as
      // it is used for multiple async_write operations, as well as for
      // obtaining the I/O executor (see get_executor below).
      tcp::socket& socket_;

      // The allocated buffer for the encoded message. The std::unique_ptr
      // smart pointer is move-only, and as a consequence our intermediate
      // completion handler is also move-only.
      std::unique_ptr<std::string> encoded_message_;

      // The repeat count remaining.
      std::size_t repeat_count_;

      // A steady timer used for introducing a delay.
      std::unique_ptr<boost::asio::steady_timer> delay_timer_;

      // To manage the cycle between the multiple underlying asychronous
      // operations, our intermediate completion handler is implemented as a
      // state machine.
      enum { starting, waiting, writing } state_;

      // As our composed operation performs multiple underlying I/O operations,
      // we should maintain a work object against the I/O executor. This tells
      // the I/O executor that there is still more work to come in the future.
      boost::asio::executor_work_guard<tcp::socket::executor_type> io_work_;

      // The user-supplied completion handler, called once only on completion
      // of the entire composed operation.
      typename std::decay<decltype(completion_handler)>::type handler_;

      // By having a default value for the second argument, this function call
      // operator matches the completion signature of both the async_write and
      // steady_timer::async_wait operations.
      void operator()(const boost::system::error_code& error, std::size_t = 0)
      {
        if (!error)
        {
          switch (state_)
          {
          case starting:
          case writing:
            if (repeat_count_ > 0)
            {
              --repeat_count_;
              state_ = waiting;
              delay_timer_->expires_after(std::chrono::seconds(1));
              delay_timer_->async_wait(std::move(*this));
              return; // Composed operation not yet complete.
            }
            break; // Composed operation complete, continue below.
          case waiting:
            state_ = writing;
            boost::asio::async_write(socket_,
                boost::asio::buffer(*encoded_message_), std::move(*this));
            return; // Composed operation not yet complete.
          }
        }

        // This point is reached only on completion of the entire composed
        // operation.

        // We no longer have any future work coming for the I/O executor.
        io_work_.reset();

        // Deallocate the encoded message before calling the user-supplied
        // completion handler.
        encoded_message_.reset();

        // Call the user-supplied handler with the result of the operation.
        handler_(error);
      }

      // It is essential to the correctness of our composed operation that we
      // preserve the executor of the user-supplied completion handler. With a
      // hand-crafted function object we can do this by defining a nested type
      // executor_type and member function get_executor. These obtain the
      // completion handler's associated executor, and default to the I/O
      // executor - in this case the executor of the socket - if the completion
      // handler does not have its own.
      using executor_type = boost::asio::associated_executor_t<
          typename std::decay<decltype(completion_handler)>::type,
          tcp::socket::executor_type>;

      executor_type get_executor() const noexcept
      {
        return boost::asio::get_associated_executor(
            handler_, socket_.get_executor());
      }

      // Although not necessary for correctness, we may also preserve the
      // allocator of the user-supplied completion handler. This is achieved by
      // defining a nested type allocator_type and member function
      // get_allocator. These obtain the completion handler's associated
      // allocator, and default to std::allocator<void> if the completion
      // handler does not have its own.
      using allocator_type = boost::asio::associated_allocator_t<
          typename std::decay<decltype(completion_handler)>::type,
          std::allocator<void>>;

      allocator_type get_allocator() const noexcept
      {
        return boost::asio::get_associated_allocator(
            handler_, std::allocator<void>{});
      }
    };

    // Initiate the underlying async_write operation using our intermediate
    // completion handler.
    auto encoded_message_buffer = boost::asio::buffer(*encoded_message);
    boost::asio::async_write(socket, encoded_message_buffer,
        intermediate_completion_handler{
          socket, std::move(encoded_message),
          repeat_count, std::move(delay_timer),
          intermediate_completion_handler::starting,
          boost::asio::make_work_guard(socket.get_executor()),
          std::forward<decltype(completion_handler)>(completion_handler)});
  };

  // Encode the message and copy it into an allocated buffer. The buffer will
  // be maintained for the lifetime of the composed asynchronous operation.
  std::ostringstream os;
  os << message;
  std::unique_ptr<std::string> encoded_message(new std::string(os.str()));

  // Create a steady_timer to be used for the delay between messages.
  std::unique_ptr<boost::asio::steady_timer> delay_timer(
      new boost::asio::steady_timer(socket.get_executor()));

  // The boost::asio::async_initiate function takes:
  //
  // - our initiation function object,
  // - the completion token,
  // - the completion handler signature, and
  // - any additional arguments we need to initiate the operation.
  //
  // It then asks the completion token to create a completion handler (i.e. a
  // callback) with the specified signature, and invoke the initiation function
  // object with this completion handler as well as the additional arguments.
  // The return value of async_initiate is the result of our operation's
  // initiating function.
  //
  // Note that we wrap non-const reference arguments in std::reference_wrapper
  // to prevent incorrect decay-copies of these objects.
  return boost::asio::async_initiate<
    CompletionToken, void(boost::system::error_code)>(
      initiation, token, std::ref(socket),
      std::move(encoded_message), repeat_count,
      std::move(delay_timer));
}

//------------------------------------------------------------------------------

void test_callback()
{
  boost::asio::io_context io_context;

  tcp::acceptor acceptor(io_context, {tcp::v4(), 55555});
  tcp::socket socket = acceptor.accept();

  // Test our asynchronous operation using a lambda as a callback.
  async_write_messages(socket, "Testing callback\r\n", 5,
      [](const boost::system::error_code& error)
      {
        if (!error)
        {
          std::cout << "Messages sent\n";
        }
        else
        {
          std::cout << "Error: " << error.message() << "\n";
        }
      });

  io_context.run();
}

//------------------------------------------------------------------------------

void test_future()
{
  boost::asio::io_context io_context;

  tcp::acceptor acceptor(io_context, {tcp::v4(), 55555});
  tcp::socket socket = acceptor.accept();

  // Test our asynchronous operation using the use_future completion token.
  // This token causes the operation's initiating function to return a future,
  // which may be used to synchronously wait for the result of the operation.
  std::future<void> f = async_write_messages(
      socket, "Testing future\r\n", 5, boost::asio::use_future);

  io_context.run();

  try
  {
    // Get the result of the operation.
    f.get();
    std::cout << "Messages sent\n";
  }
  catch (const std::exception& e)
  {
    std::cout << "Error: " << e.what() << "\n";
  }
}

//------------------------------------------------------------------------------

int main()
{
  test_callback();
  test_future();
}