Re: ASIO: acceptor::async_accept throws "Operation cancelled" after restart io_service

From: Gavin Lambert (boost_at_[hidden])
Date: 2021-04-19 21:47:34

On 20/04/2021 5:27 am, Joost Kraaijeveld wrote:
> In short my program:
> - creates an io_service
> - calls acceptor::async_accept
> - calls io_service.run
> - handle several connections OK
> - calls io_service.stop
> - the acceptor goes out of scope (i.e. the destructor of the owning
> object is ran)
> - calls io_service.restart
> - calls acceptor::async_accept which then throws "Operation cancelled"
> 
> Should it be possible to do this? When/under what circumstances does
> async_accept throw this exception?

Yes, that's normal.  Async operations are typically automatically 
cancelled when the related I/O object is destroyed, but the callback is 
still queued.  But it can't actually run until you have a thread running 
the I/O service.  This is part of why it's usually recommended to pass a 
shared_ptr to your owning object into callbacks, to keep them alive 
longer than the I/O object itself.

(There is *always* a callback at the end of every async operation.  Even 
if just to say that it was cancelled.)

io_service.stop() simply stops the thread pool, it doesn't do anything 
else.  When you start it again then any previously queued callbacks will 
be able to execute.