Boost logo

Boost :

From: Gustavo Guerra (gustavobt_at_[hidden])
Date: 2002-09-05 10:56:27


"Ronald Garcia" <garcia_at_[hidden]> wrote in message
news:3D76375C.4090101_at_cs.indiana.edu...
> Gustavo Guerra wrote:
>
> >Hi
> >Have you ever looked at pointainer from http://ootips.org/yonat/4dev/ ?
> >Could you comment on the different approaches used by your library, where
> >reference-count is inforced in the iterators, and pointainer, where
> >reference-count is inforced by container-wrappers?
> >
> >
> Greetings,
>
> Having quickly browsed the sparse documentation for pointainer, it looks
> to me like our libraries are altogether different beasts. A quick
> perusal of my documentation should clarify the difference.
>

Yes, you're right. Sorry. I got your library right, but misinterpreted
pointerator. What is was thinking pointerator was is the following, wich I
now ask you to compare to your solution:

I find it more easy to explain by an example.Your
shared_iterator_example3.cpp would become the following:

#include "boost/shared_container.hpp"
#include "boost/tuple/tuple.hpp"
#include <algorithm>
#include <iostream>
#include <vector>

typedef shared_container<std::vector<int> > ints;

std::pair<ints::iterator, ints::iterator>
return_range() {
    ints range;
    range.push_back(0);
    range.push_back(1);
    range.push_back(2);
    range.push_back(3);
    range.push_back(4);
    range.push_back(5);
    return make_pair(v.begin(); v.end());
}

int main() {
    ints::iterator i, end;
    boost::tie(i,end) = return_range();

    std::copy(i,end,std::ostream_iterator<int>(std::cout,","));
    std::cout.put('\n');
}

What do you think of this interface? It hides the shared_ptr from the user
and makes it more easy to use. Implementation-wise, it has the benefit that
you could use counted_base for intrusive reference counting, wich would be
more efficient. The corresponding iterators would include shared_ptr to the
shared container, much like your version. Of course this would be much more
harder to implement. A possible implementation would be to
shared_container<T> inherit from T and redefining all methods that return
iterators, requiring a little metaprogramming to know wich functions to
redefine depending on the type of the container.
Pros of your version: more simple and scalable. Pros of a pseudo-version
like I just described: more easy to use and maybe more efficient.

What do you think?

Regards
Gustavo Guerra


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk