X-Received: by 10.68.226.168 with SMTP id rt8mr3129762pbc.8.1356321762873; Sun, 23 Dec 2012 20:02:42 -0800 (PST)
Received: by 10.50.42.194 with SMTP id q2mr7539974igl.11.1356321762834; Sun, 23 Dec 2012 20:02:42 -0800 (PST)
Path: csiph.com!v102.xanadu-bbs.net!xanadu-bbs.net!nntp.club.cc.cmu.edu!newsfeed.news.ucla.edu!usenet.stanford.edu!f6no14678232pbd.1!news-out.google.com!s9ni56359pbb.0!nntp.google.com!kr7no8059466pbb.0!postnews.google.com!glegroupsg2000goo.googlegroups.com!not-for-mail
Newsgroups: comp.programming.threads
Date: Sun, 23 Dec 2012 20:02:42 -0800 (PST)
In-Reply-To: <0.a47ead5e11f16b983c51.20121221144029GMT.87a9t78o4i.fsf@bsb.me.uk>
Complaints-To: groups-abuse@google.com
Injection-Info: glegroupsg2000goo.googlegroups.com; posting-host=116.252.156.45; posting-account=0gvzRgoAAAD7Lpquvmb26pYmSD9lG7LB
NNTP-Posting-Host: 116.252.156.45
References: <d54451ed-576c-4873-b031-c03509ccc8e0@googlegroups.com> <0.a47ead5e11f16b983c51.20121221144029GMT.87a9t78o4i.fsf@bsb.me.uk>
User-Agent: G2/1.0
MIME-Version: 1.0
Message-ID: <fcba6a20-e725-4b1c-af1e-4a283a37b191@googlegroups.com>
Subject: Re: Have you ever seen real world lost wake-up problems?
From: Tian Tan <silverbullettt@gmail.com>
Injection-Date: Mon, 24 Dec 2012 04:02:42 +0000
Content-Type: text/plain; charset=ISO-8859-1
Xref: csiph.com comp.programming.threads:1268

> And what CV implementation allows one to specify the thread that should
> 
> be woken up?  If there can be only one waiting thread, there's nothing
> 
> lost by using the conventional signalling mechanism, and if there can be
> 
> many waiting threads, targeting the notification is almost certainly
> 
> going to cause problems.
> 
What does it mean? Why multi-waiting-threads is going to cause problems?

 
> You might also lack some knowledge as well.  Obviously I can't tell, but
> 
> the example above looks so wrong it doesn't seem likely to be common
> 
> patter for a bug.
> 
You're right, it's too naive. CV shouldn't be used like that.


> It's not clear what the bug is you are talking about.  Anyway, I lack the
> 
> experience to estimate how common any particular synchronisation bug is.
> 
The bug which I'm talking about can be concluded like this: http://docs.oracle.com/cd/E19253-01/816-5137/sync-30/index.html

> 
> Yes.  You just need to use CVs correctly -- associate them with a real
> 
> condition and a mutex (or lock) and test the condition in a loop with
> 
> the correct locking:
> 
> 
> 
>   mutex.lock()
> 
>   while (the condition I need is not true)
> 
>      cv.wait(mutex);
> 
>   mutex.unlock();
> 
> If you find examples of incorrectly used condition variables, that would
> 
> have some value (you could submit bug reports for example) but if it is
> 
> valuable enough to spend time looking for them, I can't say.
> 
It's helpful, thank you Ben.


-- TT