Re: [Boost-users] [Smart Pointers] enable_shared_from_this::shared_from_this() behaviour query

Subject: Re: [Boost-users] [Smart Pointers] enable_shared_from_this::shared_from_this() behaviour query
From: eddyaod_at_[hidden]
Date: 2011-06-13 11:26:17

On 13 June 2011 06:29, David Blaikie <dblaikie_at_[hidden]> wrote:

> shared_from_this() throws bad_weak_ptr if it's called in a derived
>> destructor even if the object is heap allocated, does anyone know if this is
>> intended? And if so, why?
>
>
> What else could it do? if the dtor is running the refcount has reached
> zero... and, well, you know, the dtor is running.
>
> If you created a shared_ptr to that object - what would happen when that
> new shared_ptr went out of scope? The refcount would reach zero (again) &
> the object would be double deleted.
>

It's possible to create smart_pointers with null deleters though that in
itself doesn't solve my problem.

>
> What semantics were you expecting?
>

The problem I was facing involved a cyclic data structure (two way
references) and was implemented using weak_ptrs. When an object was deleted
it was necessary to notify its referers of its deletion which I was looking
to do using a weak_ptr.

However, I then realised that although weak_ptrs are great, they are not
useful at destruction time due to the fact that there is no smart_pointer
remaining with which they are validated, so I have re-implemented using
raw-pointers.

It's hit upon a major semantic difference between smart pointers and raw
pointers. I suspect that it may be possible to implement a smart pointer
system that could actually continue to correctly mask pointers during
deletion (perhaps by deleting the managed object just before the last
remaining shared_ptr internals are destroyed).

>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>