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 initi
| 46 | // To enable this, when implementing an asynchronous operation we must package |
| 47 | // the initiation step as a function object. |
| 48 | struct async_write_message_initiation |
| 49 | { |
| 50 | // The initiation function object's call operator is passed the concrete |
| 51 | // completion handler produced by the completion token. This completion |
| 52 | // handler matches the asynchronous operation's completion handler signature, |
| 53 | // which in this example is: |
| 54 | // |
| 55 | // void(boost::system::error_code error) |
| 56 | // |
| 57 | // The initiation function object also receives any additional arguments |
| 58 | // required to start the operation. (Note: We could have instead passed these |
| 59 | // arguments as members in the initiaton function object. However, we should |
| 60 | // prefer to propagate them as function call arguments as this allows the |
| 61 | // completion token to optimise how they are passed. For example, a lazy |
| 62 | // future which defers initiation would need to make a decay-copy of the |
| 63 | // arguments, but when using a simple callback the arguments can be trivially |
| 64 | // forwarded straight through.) |
| 65 | template <typename CompletionHandler> |
| 66 | void operator()(CompletionHandler&& completion_handler, |
| 67 | tcp::socket& socket, const char* message) const |
| 68 | { |
| 69 | // The post operation has a completion handler signature of: |
| 70 | // |
| 71 | // void() |
| 72 | // |
| 73 | // and the async_write operation has a completion handler signature of: |
| 74 | // |
| 75 | // void(boost::system::error_code error, std::size n) |
| 76 | // |
| 77 | // Both of these operations' completion handler signatures differ from our |
| 78 | // operation's completion handler signature. We will adapt our completion |
| 79 | // handler to these signatures by using std::bind, which drops the |
| 80 | // additional arguments. |
| 81 | // |
| 82 | // However, it is essential to the correctness of our composed operation |
| 83 | // that we preserve the executor of the user-supplied completion handler. |
| 84 | // The std::bind function will not do this for us, so we must do this by |
| 85 | // first obtaining the completion handler's associated executor (defaulting |
| 86 | // to the I/O executor - in this case the executor of the socket - if the |
| 87 | // completion handler does not have its own) ... |
| 88 | auto executor = boost::asio::get_associated_executor( |
| 89 | completion_handler, socket.get_executor()); |
| 90 | |
| 91 | // ... and then binding this executor to our adapted completion handler |
| 92 | // using the boost::asio::bind_executor function. |
| 93 | std::size_t length = std::strlen(message); |
| 94 | if (length == 0) |
| 95 | { |
| 96 | boost::asio::post( |
| 97 | boost::asio::bind_executor(executor, |
| 98 | std::bind(std::forward<CompletionHandler>(completion_handler), |
| 99 | boost::asio::error::invalid_argument))); |
| 100 | } |
| 101 | else |
| 102 | { |
| 103 | boost::asio::async_write(socket, |
| 104 | boost::asio::buffer(message, length), |
| 105 | boost::asio::bind_executor(executor, |
no outgoing calls
no test coverage detected