[boost] Re: Formal Review request: Pointer Container Library

From: Thorsten Ottosen (nesotto_at_[hidden])
Date: 2003-12-25 02:09:13

"Nicolas Fleury" <nidoizo_at_[hidden]> wrote in message
news:bscjq3$782$1_at_sea.gmane.org...
> Thorsten Ottosen wrote:
> > Feedback is welcome.
>
> I just want to tell you that I am very interested in your library.

That's good to hear.

> Right now, when I want a container to have ownership of pointers, I
> basically have the choice between complicate the usage (not that much
> but still)

exactly.

>and handle the ownership by hand, which can be error-prone,

you have no idea :-)

> particularly if code deleting elements is added.
> I usually choose to
> store shared_ptrs, but it complicates for nothing the creation of all
> added elements in context when the elements are not actually *shared*.
> Your solution would be something very simple to explain to my collegues
> and adds no burden to creation of elements.  If it is someday in Boost,
> I would use it right away.

well, you should start to try to use it soon. Even some is of another
oppinion,
I'm certain that all major design decisions are in place.

> I think the const-copy semantic of elements in STL containers is
> basically forcing to use shared_ptrs even when elements are not shared,
> so I think your solution of having equivalent containers for pointers is
> the good solution.

indeed. It's the only way to get the zero-overhead C++ hallmark.

> Like others pointed, I think you example could be more convincing.  You
> should show how it simplifies everything in day-to-day work with
> basically all operations.

yes, I should. I'm just wondering if those who said that actually browsed to
the eaxmple section
and viewed the examples.

> About ptr_map, I wonder if a map with the key a pointer should be
> provided (and another with both key and value pointers).  Same thing for
> multimap.

So you want three types of maps. Can you give a moivating example? I have
seen use for the
mapped type to be a pointer.

There is a slight problem with two pointers, namely that an exception-safe
insert cannot
take two naked pointer values, it would have to be two auto_ptr's.

> About releasing functions returning auto_ptrs and adopting ones taking
> one, even if I have used this convention before, I feel it is far from
> standard

yeah std::auto_ptr.release() does  not return another auto_ptr <g>.

>and I don't remember having seeing it in Boost.

nothing like this is in boost.

> It is however
> more safe.

.exactly.

Thanks for your comments, sorry I not elaborating very much, but I've just
been celebrating christmas for two days without too  much sleep.(and I'm
still a biit drunk)

br and merry Christnas

Thorsten