Re: [Threads-devel] boost::condition and Lock concept

From: Frank Mori Hess (fmhess_at_[hidden])
Date: 2007-12-06 10:46:38

On Thursday 06 December 2007 03:02, Anthony Williams wrote:
> Quoting Frank Mori Hess <frank.hess_at_[hidden]>:
> > Is it possible to make condition_variable_any work with recursive
> > mutexes?  It
> > seems like an additional interface on the lock would be required,
> > since a recursive mutex may still be locked by the current thread even
> > after calling unlock().
>
> I think that would be dangerous. If you don't know how many times
> you've locked the recursive mutex when you call wait() (in order to
> unlock all but one prior to the call), then you might be unlocking the
> mutex when a layer above you is expecting it to still be locked and
> has incomplete changes that are protected by the lock.

Okay, I suggest a sentence be added to the condition docs to explicitly 
state it is unsafe (will probably deadlock) to wait on a recursively 
locked mutex.  

I wonder if a sanity check could be added to condition::wait() to test for 
this case.  It would require a generalization of owns_lock(), which 
instead of returning a bool, returns an integer indicating how many times 
the current thread has recursively locked the mutex.  Then wait() could 
check it returns exactly 1.

-- 
Frank