Re: [Boost-users] [thread] Condition::notify_one, wait

From: Anthony Williams (anthony_w.geo_at_[hidden])
Date: 2008-04-10 09:24:20

Anteru <newsgroups_at_[hidden]> writes:

> Anthony Williams schrieb:
>> OK. I can't reproduce the problem here, but looking at the code with fresh
>> eyes, there might be a bug in the code that tries to avoid spurious wakes.
>> Here's something to try: comment out the
>> 
>>                  while(!woken)
>> 
>> on line 175 of boost/thread/win32/condition_variable.hpp
>> 
>> This will increase the incidence of spurious wakes, but might fix the
>> problem. If the problem is reproducible without the change, and this change
>> fixes it, then I need to investigate the new condition variable code more
>> thoroughly.
>
> With the line, I can reproduce it, although not always (maybe 2 out of 3 
> runs show it). Without the line, I had 15+ runs without any problem 
> (with heavily varying system load, from totally idle to fully loaded, 
> debug and release tested). Either your fix reduces the probability of 
> the error dramatically or it even fixes it ;)
>
> If there is something more I can test, let me know. Thanks,

I found a usage pattern that triggers the bug, and added a test to trunk. I've
also added a "proper" fix to trunk: the simple fix above would yield many
spurious wakes. I'd be grateful if you could try the version of boost.thread
from svn trunk.

Anthony
-- 
Anthony Williams            | Just Software Solutions Ltd
Custom Software Development | http://www.justsoftwaresolutions.co.uk
Registered in England, Company Number 5478976.
Registered Office: 15 Carrallack Mews, St Just, Cornwall, TR19 7UL