...one of the most highly
regarded and expertly designed C++ library projects in the
world.
— Herb Sutter and Andrei
Alexandrescu, C++
Coding Standards
Following Boost.Asio's convention,
all network operations have asynchronous versions with the same name prefixed
by async_
. The last parameter
to async operations is a CompletionToken,
which dictates how the asynchronous operation will be managed and the function's
return type. These async_
functions
are called async initiating functions.
Every async initiating function has an associated handler type, which dictates how the asynchronous operation communicates its result back to the caller. This handler type always has one of the two following forms:
void(error_code)
.
Used in operations that do not have a proper result, e.g. connection::async_connect
.
void(error_code,
T)
.
Used in operations that have a result, e.g. connection::async_prepare_statement
(in this case, T
is statement
).
All asynchronous functions are overloaded to accept an optional diagnostics
output parameter. It is populated with any server-provided error information
before calling the completion handler.
As mentioned in this section, only a single async operation per connection can be outstanding at a given point in time. If you need to perform queries in parallel, open more connections to the server.
Any completion token you may use with Boost.Asio can also be used with this library. Here are some of the most common:
Callbacks. You can pass in a callable
(function pointer or function object) with the same signature as the handler
signature specified for the operation. The callable will be called when
the operation completes. The initiating function will return void
.
This example demonstrates how to use async functions with callbacks.
Futures. In this case, you pass in the
constant boost::asio::use_future
as completion token. The initiating function will return one of the following:
std::future<void>
,
if the completion handler has the form given by 1).
std::future<T>
,
if the completion handler has the form given by 2).
You can wait for the future by calling future::get
.
If an error occurs, future::get
will throw an exception. Note that the exception is thrown by Asio itself,
and will always be of type boost::system::system_error
,
even if diagnostics were available.
This example demonstrates using futures.
Stackful coroutines. In this case, you
pass in a boost::asio::yield_context
.
To obtain one of these, you should use boost::asio::spawn
to create a new coroutine. The initiating function will return:
void
, if the completion
handler has the form given by 1).
T
, if the completion
handler has the form given by 2).
If you use boost::asio::yield_context::operator[]
,
the operation will set the given error_code
when it fails. Otherwise, the function will throw a exception. Note that
this exception is thrown by Asio itself, and thus will always be of type
boost::system::system_error
. To obtain an error_with_diagnostics
we suggest using error codes and the throw_on_error
function.
You need to link against Boost.Context to use these coroutines.
This example uses stackful coroutines.
C++20 coroutines. In this case, you pass
in the constant boost::asio::use_awaitable
as completion token. The initiating function will return:
boost::asio::awaitable<void>
,
if the completion handler has the form given by 1).
boost::asio::awaitable<T>
,
if the completion handler has the form given by 2).
You can then use co_await
on this return value. If the operation fails, co_await
will throw an exception. Note that this exception is thrown by Asio itself,
and thus will always be of type boost::system::system_error
.
To obtain an error_with_diagnostics
we suggest using the boost::asio::as_tuple
completion token, which will make co_await
report failures using error codes, and the throw_on_error
function.
This example demonstrates C++20 coroutines.
All async operations in this library support per-operation
cancellation. All operations support only the terminal
boost::asio::cancellation_type
.
This means that, if an async operation is cancelled, the connection
object is left in an unspecified state, after which you should close or destroy
the connection. In particular, it is not safe
to retry the cancelled operation.
Supporting cancellation allows you to implement timeouts without explicit support from the library. This example demonstrates how to implement this pattern.
Note that cancellation happens at the Boost.Asio level, and not at the MySQL operation level. This means that, when cancelling an operation, the current network read or write will be cancelled. The operation may have already reached the server and be executed. As stated above, after an operation is cancelled, the connection is left in an unspecified state, and you should close or destroy it.