On 9/30/2013 6:21 PM, Quoth Bo Jensen:
I see I have no io_service::stop(), but I read in a stack exchange reply that was not needed in this case, is that true ?
Mostly. If you don't stop() the io_service then it will still waste some time entering each posted-but-not-yet-run job and running up until the first interruption point (assuming that the interruption point will still be "live" after the interruption exception has been thrown once, which I'm not entirely sure about -- I haven't really played with interruption points too much myself). If your number of jobs is less than or equal to your number of worker threads then this will be invisible except in a rare race if the job completes very quickly. Even with a larger number of jobs it might be hard to spot if some of their work is asynchronous itself, thereby allowing all of the original jobs to get run early on. Either way though as long as you join all the worker threads before you allow the service to be destroyed then it ought to be safe -- if you hit the above case then it'll just take longer to join than you might be expecting.
Also I don't call join on the thread group, but on each thread individually, which I assume must be OK (done in JoinAllWorker()).
The thread_group::join_all() is just an individual thread join in a loop, so it should be fine as long as you're doing an unconditional join and not a timed_join. Otherwise I can't think of anything else helpful, sorry! :( You could try asking on the asio-specific mailing list. They're likely going to ask you to try to make a smaller test-case that you can share in full though.