[boost] Re: Traits-based design and ODR

From: Andrei Alexandrescu (andrewalex_at_[hidden])
Date: 2002-04-30 15:11:13

"Beman Dawes" <bdawes_at_[hidden]> wrote in message
> I skimmed MC++D quickly, but couldn't find your rationale for choice of
> defaults.  I'm assuming it was something like "the safest, most commonly
> needed for a wide range of general uses."
>
> Instead, defaults would have to be chosen as "the most widely useful for
> interfacing between libraries."
>
> Now it may be that these two criteria result in the same set of
> defaults.  But you need to review your choice of defaults to make sure
that
> is the case.

The current choices are the ones that are most generic and safe:
non-intrusive reference counting, assertion-based checking, and no implicit
conversion. If I'm not wrong, they correspond to the behavior of shared_ptr.

It's unclear to me whether these are the best defaults for interfacing
libraries, although that seems to be the case.

In my opinion, imposing one smart pointer type as a communication tool for
all libraries is not a very worthwhile goal, and might even become a red
herring, as I will detail in a later post. There are many concerns about
inter-module communication, including type safety, efficiency, and ease of
use. Imposing one smart pointer might jeopardize applications on all these
three dimensions.

Andrei