On 7/20/07, Stefan Seefeld <seefeld@sympatico.ca> wrote:
Michael Caisse wrote:
How about using boost::shared_ptr. Something like this in your class instead:
boost::shared_ptr< boost::recursive_mutex > myMutex;
Then you would protect both copies of the resource that is part of the class by a single (shared) mutex. That doesn't sound right.
Exactly, I don't want to copy or share the mutex in any way. I want each object to have its own mutex. I just wanted to avoid having to write copy ctor for no reason. But I just realized that my whole argument is definitely moot. In all my classes with mutexes that I want to be copyable, I need to write a copy ctor and operator= anyways to make them thread-safe. For instance, I'll have a function, say foo(), that initializes a shared pointer of some kind. This function is already synchronized using scoped_lock(m_mutex). To be thread-safe, I really need my copy ctor to be synchronized (with rhs.m_mutex) and my operator= to be synchronized too (with both m_mutex and rhs.m_mutex). Phillip