boost.threads [was RE: [boost] Re: sockets /network programming Requirements]

From: williamkempf_at_[hidden]
Date: 2001-08-06 09:11:44

--- In boost_at_y..., John Max Skaller <skaller_at_m...> wrote:
> williamkempf_at_h... wrote:
> 	[I'd even be worried about 'once' stuff: this seems
> like the moral equivalent of 'singletons' which I believe
> are also often symptomatic of a design error: is there a 
> compelling example for it 'once' things?]

Once methods are not singletons.  The classic reason for their need 
is to synchronize the creation of global data.  A simple example, I 
wrote a gc_ptr<> class that had to track objects on the GC heap and 
had a map to do so.  In order to insure proper creation I followed 
the standard idiom.

gc_map& get_map()
{
   static gc_map map;
   return map;
}

The above is not thread safe, however.  The constructor for the 
static "map" can be entered twice if get_map() is called from two 
threads at (conceptually) the same time.  Once methods allow us to 
insure objects are created only once:

static gc_map* _map;
static boost::once_flag _map_once = BOOST_INIT_ONCE;

void init_map()
{
   static gc_map map;
   _map = map;
}

gc_map& get_map()
{
   boost::call_once(&init_map, _map_once);
   return *_map;
}

A better solution is for the compiler to handle this for us with the 
advent of a new keyword:

gc_map& get_map()
{
   static synch gc_map map;
   return map;
}

This definately simplifies the code, but the result is the same.

Bill Kempf