Re: [Boost-users] Thread_group semantics

Subject: Re: [Boost-users] Thread_group semantics
From: Gavin Lambert (gavinl_at_[hidden])
Date: 2014-01-13 19:51:25

On 14/01/2014 13:26, Quoth Kenneth Adam Miller:
> Would it work better if it was threads.join_all() then ioServices.stop()?

No, that would make it block forever.

You must do one of the following:

1. create an io_service.
2. create an io_service::work.
3. create threads for your thread_group.
4. at any time after this, post work to the service.
5a. only on shutdown, call stop() and join_all(), in that order.  Note 
that this may result in some posted work not getting called, and 
therefore some resources not cleaned up.
5b. OR instead of calling stop(), destroy the io_service::work you 
created earlier (and if you're using ASIO async operations, you must 
also cancel or close anything outstanding that you don't want to wait 
for), and then call join_all().  this will block until all posted work 
is completed.

OR

1. create an io_service.
2. post work to the service.  (can only be done here, or from inside a 
callback called by the service)
3. create threads for your thread_group.
4. call join_all(), which will block until all pending work is done.

The former lets you make a threadpool you can keep around for the life 
of your app, and post work as desired.  The latter lets you make a 
threadpool to execute a single blocking task with multiple concurrent 
subtasks.

(Other variations are possible of course but the order of the above is 
important.)