Re: [boost] [intrusive_ptr] atomic ref counting?

From: Eric Niebler (eric_at_[hidden])
Date: 2006-05-08 21:13:01

Resending. It's been a while, and I don't want this issue to get lost. 
Peter?

Eric Niebler wrote:
> Peter Dimov wrote:
> 
>>Eric Niebler wrote:
>>
>>
>>>Peter Dimov wrote:
>>>
>>>
>>>>I can't help but think that even if I did add the above, you
>>>>wouldn't use it because of the virtual destructor,
>>>
>>>You're right, because in xpressive alone I use counted_base in no less
>>>than 3 places where the virtual destructor would be needless overhead!
>>>Clearly, it can't be all that uncommon.
>>
>>It may be common. The point is that we need to settle on a single version of 
>>boost::counted_base. The virtual destructor version does have its benefits: 
>>it's easier to understand and use (CRTP can be a bit hard to swallow) and it 
>>allows intrusive_ptr<counted_base> to be used as the intrusive version of 
>>shared_ptr<void> and void* (with the additional benefit of a working 
>>dynamic_cast.)
>>
>>counted_base is somewhat bicycle sheddish and I've considered adding one as 
>>documentation rather than code. But one problem with that is atomic_count 
>>being in detail. :-)
> 
> 
> 
> I'm confident that we can settle on a single design, and we should. I'd 
> rather include your counted_base with a virtual than leaving it in the 
> docs where people will ignore it and it won't get used.
> 
> How about this:
> 
> template<typename Derived>
> struct basic_counted_base
> {
>      ... my counted_base implementation here ...
>      ... with no virtual destructor ...
> };
> 
> struct counted_base
>    : basic_counted_base<counted_base>
> {
>      virtual ~counted_base() {}
> };
> 
> That way we get the best of both worlds.
> 
> 
> 
>>>The following program exposes the problem for me with vc7.1, vc8 and
>>>gcc 3.4.4. If you comment out the friend functions and uncomment the
>>>global free functions, the problem goes away. I don't understand it --
>>>is some name look-up subtlety at play here?
>>
>>I don't understand it either. Comeau/EDG compiles it. A non-template 
>>counted_base doesn't have this problem, FWIW.
> 
> 
> Confirmed. Also, the problem doesn't happen when impl in my example is a 
> non-template. Looks like compiler bugs, but it's strange that both VC 
> and gcc reject it with similar error messages. <shrug>
> 
> 

-- 
Eric Niebler
Boost Consulting
www.boost-consulting.com