Boost::asio 和嵌套/递归 service.run()/run_one()
io_service.run_one() 的嵌套调用是否可能不是 boost::asio 的有效使用。
例如，我有一个处理程序，它在套接字上收到某些内容后执行。 io_service 正在另一个线程中通过 io_service.run_one() 运行。因此，我认为 io_service 将在为接收处理程序提供服务后停止。
然后在接收处理程序中，我再次执行 io_service.run_one() 以便现在通过套接字发送回一些内容。这意味着存在 io_service.run_one() 方法的嵌套/递归调用。
这种行为实际上是行不通的。我的意思是有时 run_one() 方法中有一个永远不会返回的块。我也找不到任何有关此特定案例的文档处理。
那么，是否允许递归/嵌套 io_service 执行，或者这不是一个好的行为？
PS io_service.reset() 在每次 run_one() 执行之前使用。
Is it possible that nested calls of io_service.run_one() is not a valid use of boost::asio.
For example I have a handler which is executed after something has been received on a socket. The io_service is running in another thread with io_service.run_one(). Hence in my opinion the io_service will stop after serving the receive handler.
Then in the receive handler I am executing io_service.run_one() again in order to send now something back over the socket. This means, that there is a nested/recursive call of io_service.run_one() method.
This behavior doesn't really work. I mean sometimes there is a block in the run_one() method which never returns. I couldn't also find any documentation handling about this particular case.
So, is it allowed to have recursive/nested io_service execution at all or is this not a well behavior?
P.S. io_service.reset() is used in front of every run_one() execution.
如果你对这篇内容有疑问，欢迎到本站社区发帖提问 参与讨论，获取更多帮助，或者扫码二维码加入 Web 技术交流群。
绑定邮箱获取回复消息
由于您还没有绑定你的真实邮箱，如果其他用户或者作者回复了您的评论，将不能在第一时间通知您！
发布评论
评论（1）
据我所知，这既不是预期用途，也没有在 asio 中记录，所以看起来一个安全的假设是：不，它无效。
现在，它暂时可以在某些或所有平台上运行。但这并不一定是一个好主意。
另外，如果您不想在处理程序内部阻塞，调用
run_one()，您可能需要考虑poll_one()。不管怎样，这样做听起来确实很可疑。假设您假设正在发送一条特定的消息，以便您可以立即执行一些操作？我只能想到非常人为的示例，其中可以安全地假设两个消息在 asio 消息队列中是背靠背的。
As far as I know, this is neither intended use nor documented in asio, so it would seem like a safe assumption is: no, it's not valid.
Now, it may work, on some or all platforms, for now. That doesn't necessarily make it a good idea though.
Also, if you intend to not block inside your handlers, calling
run_one(), you might want to considerpoll_one().Either way, it definitely sounds suspicious to do this. Presumably you're assuming that a specific message is being dispatched, so that you can do some work immediately following it? I can only think of very contrived examples where it would be a safe assumption that two messages are back-to-back in the asio message queue.