Re: [boost] Re: Fw: Why scoped_ptr dost provide custom deallocation?

From: Greg Colvin (Gregory.Colvin_at_[hidden])
Date: 2003-02-14 13:35:58

It wouldn't be too difficult to write a custom deallocator
for shared_ptr that supported release of ownership.  Given
the dangers I'd rather see it done that way than by adding
a release() function to shared_ptr.

And I don't see a problem with adding constructors to
scoped_ptr that support custom deallocation, especially
since C++ has no 'finally' construct.

At 04:36 AM 2/14/2003, Alisdair Meredith wrote:
>Peter Dimov wrote:
>
>> > And what if you use the deallocator as a template parameter?
>
>> > Anyway: Do you know any smart-pointer class, which supports custom
>> > deallocator and can transfer ownership? (auto_ptr does not support
>> > custom deallocator and either smart_ptr or shared_ptr does not support
>> > release() method).
>
>> Both good questions. Do we have an answer?
>
>These both sound like questions to be solved by the policy-pointer David
>Held is formalising as we speak [based strongly on Andrei Alexandrescu's
>smart_pointer class in the Loki library/Modern C++ Design book]
>
>I would be worried to see extra template parameters appearing in
>shared_pointer, as the big advantage it overs over the full-blown policy
>implemetation is its simplicity to the user.
>
>If that simplicity is no longer enough, then maybe you do need the full
>policy implementation.  I too worry a little about increasing the size
>of my pointers with an unused deallocator, but nothing in my profiling
>has suggested this is a real problem in practice, simply an
>implementation detail I am concerned about because I know it is there. 
>This is one of the trade-offs of trying to use a general-purpose
>library.
>
>Well, that's my take on it anyway <g>
>
>-- 
>AlisdairM
>
>_______________________________________________
>Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost