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

From: Frank Mori Hess (fmhess_at_[hidden])
Date: 2007-12-06 17:47:25

On Thursday 06 December 2007 11:14, Anthony Williams wrote:
> That's something worth bearing in mind. There's nothing to stop
> condition::wait being specialized for a lock on a recursive mutex to
> do that.

Is it worth requiring objects modeling the various mutex concepts to 
provide some information on the mutex's properties?  Like a "static const 
bool my_mutex_class::recursive" which is true or false depending on 
whether my_mutex_class is a recursive mutex.  You could also have similar 
static members describing whether the mutex support try locks or timed 
locks.

Or, the property information could be split out into a separate class, like

template<typename Mutex> class mutex_properties;

sort of like how numeric_limits<> works.

The reason I ask is I'm working on some classes that have a Mutex template 
parameter, and it would be useful to be able to query some of the 
properties of the mutex.  For example, I've got a mutex wrapper class that 
automatically checks for cycles in the locking order.  But the checking 
code varies depending on whether the Mutex parameter is recursive or not 
(it should detect an error when trying to recursively lock a non-recursive 
mutex, but not when recursively locking a recursive mutex).  I'd like to 
just have one acyclic_mutex<> wrapper class that works with any kind of 
underlying mutex, instead of a acyclic_mutex, try_acyclic_mutex, 
recursive_acyclic_mutex, etc.

I suppose I could just implement mutex_properties<> in my library, 
providing specializations for all the mutex classes in boost.thread and 
for my mutex wrappers.  Then I'd put an explicit requirement in my library 
that if they want to use some other mutex type as the template parameter, 
they also need to implement a mutex_properties<> specialization.  A 
default mutex_properties<> implementation that corresponds to the minimal 
Mutex concept might also be convenient.

Hmm, so I guess my question now is, is it worth providing a 
mutex_properties<> class in boost.thread?

-- 
Frank