Subj : Re: reliability of setting thread priority
To   : comp.programming.threads
From : gottlobfrege
Date : Sat Feb 26 2005 10:26 pm

So to summarize this thread, how's this:

if (AtomicRead(&initted))
      return;

if (!AtomicExchange(&entry, true)) // no one was in before us?
{
      do_the_important_init_stuff();

	AtomicExchange(&initted, true);

	if (AtomicRead(&event)) // anyone waiting?
	{
		SetEvent(event);
	}
}
else
{
	// wait for init.
	// while waiting, create an event to wait on
	if (!AtomicRead(&event)) // anyone else waiting already?
	{
		tmpEvent = CreateEvent(NULL, true, false, NULL); // manual reset (ie
will never be reset)
		if (AtomicCompareExchange(&event, 0, tmpEvent) == 0)
		{
			// wasn't set, is now
		}
		else
		{
			// someone snuck in
			// get rid of our attempt
			CloseHandle(tmpEvent);
		}
	}

	if (!AtomicRead(&initted))
	{
		Wait(event);
	}
}

Looks to me, like
1. hopefully, and most importantly, it works :-)
2. if there is no contention, there is very very little cost
3. as this is all about initialization, I'm not too worried about
Alexander's concerns about throwing in a destructor.  (But I suspect
some error handling could be added - as an exercise for the reader)
4. I could merge the bools 'initted' and 'entry' into a single 3 state
int.

Thoughts?
Anyone still interested? :-)

Thanks,
Tony.

.