Re: [boost] Is there any interest in a library that provides container

From: Dave Harris (brangdon_at_[hidden])
Date: 2005-10-27 13:19:51

In-Reply-To: <djpuvk$q3d$1_at_[hidden]>
simon_at_[hidden] (Simon Buchan) wrote (abridged):
> > Generally when I've wished for that I've also wished for a few other 
> > virtual functions. For example, a hook which gets called whenever an 
> > item is added to the vector [...]
>
> That's what your class would add, presumably.

Yes, but as I understand it, I can't usefully add them by inheriting from 
the proposed class. This is because, eg, virtual_vector::push_back is not 
virtual. Users could do:

    my_vec<int> *p1 = new my_vec;
    p1->push_back( 0 ); // OK, uses my hook.
    virtual_vector *p2 = p1;
    p2->push_back( 0 ); // Silent error, does not use my hook.

In my view to be useful a virtual_vector would need lots of virtual 
methods, not just the destructor.

> But when you *do* want to write a vector (or whatever) interface 
> compliant class, it's a bloody pain to do it for each one you do.

I can only report my experience, which is that I've never wanted to do 
that.

> His impl has the benefit that you just publicly inherit from this class, 
> and add the extra stuff in, overriding any methods that you want to do 
> differently. (I assume the forwarders will be virtual?)

I didn't think he did intend that the forwarders be virtual. If it is, 
then I withdraw my objection. And I am curious as to whether he has:

    void virtual_vector::clear() {
        m_vector.clear();
    }

or:
    void virtual_vector::clear() {
        erase( begin(), end() );
    }

That is, whether I have to override both erase and clear or just erase. 
And whether things like begin() and end() are also virtual, and whether 
the iterators they return have virtual methods too. It sounds like it 
could be a performance hit.

> > In general inheriting from anything in the std library has the 
> > drawback that it injects unknown junk names into your class. 
> > Inheriting from a boost class is hardly better.
>
> Actually, his posted impl doesn't do that.

He isn't doing, but he proposes that users do. The reason for giving 
boost::virtual_vector a virtual destructor is so that users can inherit 
from it.

-- Dave Harris, Nottingham, UK.