From: Yitzhak Sapir (yitzhaks_at_[hidden])
Date: 2002-08-07 10:34:17
> -----Original Message-----
> From: William E. Kempf [mailto:williamkempf_at_[hidden]]
> Since call_once relies on static initialization the type has
> to be POD,
> which means it would be more like this:
>
> class once
> {
> public:
> typedef ... flag_type;
> flag_type flag;
> call_once(function0<void>&);
> };
>
> boost::once once = BOOST_ONCE_INIT;
> once.call_once(&foo);
>
> So you change the syntax slightly, but that's about all.
>
Just an idea from following the discussion: Instead of calling the class "once", you could create an interface like this:
boost::static_mutex_init -- equivalent of once_flag_type
boost::static_mutex is a regular class and is implemented like a mutex except it uses call_once mechanisms to make sure the mutex is initialized once in the constructor.
A possibly more extended variation is:
boost::static_mutex_init -- equivalent of once_flag_type
boost::static_mutex_impl is a regular class that inherits from counted_base and is implemented like a mutex except it uses call_once mechanisms to make sure the mutex is initialized once in the constructor. The constructor receives the static_mutex_init by reference.
boost::static_mutex is a wrapper for a shared_ptr to static_mutex_impl, and reimplements all the mutex functions as forwarding calls to static_mutex_impl. Its scoped_locks defined types keep copies of the shared pointer.
boost::static_mutex is copyable and all clients should copy it to make sure it is available after main() returns.
The effect is that you might provide a slightly better interface or one closer to the one you said you'd ideally like to have.