[Boost-users] Re: Q: Stopping threads (Win32) or... the mysterious case of the thread with split personality

From: Jeffrey Holle (jeff.holle_at_[hidden])
Date: 2004-12-05 14:36:36

In general you do not want any thread pooling some variable.
A background thead almost always has to pend, that is wait suspended, in 
a useful multi-threaded application.

The trick is finding the way to pend on multiple events.
Any decent threading library will provide a way.

Dan Dimerman wrote:
> Hello Jeffrey and thank you for your response,
> 
> As you point out, and is my general understanding, the thread indeed does 
> kills itself by exiting its main loop. My question pointed more towards what 
> you call "legit inter-thread" communication: I thought a shared variable 
> would be enough, given there is only one state transition (true to false), 
> just two threads and one is consistently the writer and the other 
> consistently the reader. What I see (debugger) is that from inside one 
> thread (main) the value of this variable is false, and from the other thread 
> (worker) the value is different !
> 
> "Jeffrey Holle" <jeff.holle_at_[hidden]> wrote in message 
> news:covlls$he0$1_at_sea.gmane.org...
> 
>>First, a thread should kill itself.
>>Second, you should use legit inter-thread communication.
>>The request to commit suicide should be one of the things the thread pends 
>>on.
>>Don't know boost.threads, but thats the general correct way to write a 
>>multi-threaded application.
>>
>>Dan Dimerman wrote:
>>
>>>Hello,
>>>
>>>I would like to ask the group if you know of an efficient and clean way 
>>>to stop a thread.
>>>What I've been using till now was a while loop in the function 
>>>operator()() given to the boost::thread when created, whose condition was 
>>>a boolean data member. This boolean was the means to stop the thread, 
>>>such that when it was turned off, the child thread would exit the while 
>>>loop and join the main thread.
>>>I am seeing now that this data member is being held twice (one copy for 
>>>the main thread and one copy for the child thread) and there's no synch 
>>>between the two copies; so when the main thread turns it off / false, the 
>>>child thread goes on (since its copy is still on / true) and hell breaks 
>>>loose.
>>>
>>>I am using VC7 in debug mode, so I started looking for some kind of 
>>>flag/optimization I might be inadvertently using that makes data members 
>>>into registers (thus explaining the copies), but couldn't find anything 
>>>of the sort.
>>>
>>>Any help will be greatly appreciated...
>>>
>>>Dan
>>>
>>>
>>>---
>>>Outgoing mail is certified Virus Free.
>>>Checked by AVG anti-virus system (http://www.grisoft.com).
>>>Version: 6.0.805 / Virus Database: 547 - Release Date: 03-Dec-04
> 
> 
> 
> ---
> Outgoing mail is certified Virus Free.
> Checked by AVG anti-virus system (http://www.grisoft.com).
> Version: 6.0.805 / Virus Database: 547 - Release Date: 03-Dec-04