Subject: Re: [boost] [gsoc] Interest in BGL v2?
From: Dave Abrahams (dave_at_[hidden])
Date: 2011-04-01 11:53:40
At Thu, 31 Mar 2011 19:20:47 -0700,
> On Thu, 31 Mar 2011 07:02:43 -0700, Dave Abrahams <dave_at_[hidden]>
> > At Thu, 31 Mar 2011 08:53:02 -0500,
> > Andrew Sutton wrote:
> >> I will say that our design does take a slightly more object-oriented
> >> approach (g.size(), for example), but that hardly precludes
> >> adaptation. Adaptors would have to be written as classes, but that
> >> shouldn't be a big deal.
> > I think you underestimate the cost of adaptation-via-wrapper and urge
> > you to stick with free functions unless you have a really good reason
> > to do otherwise.
> Can you elaborate?
Briefly: if you have to wrap things to get them to conform to a concept,
then they are no longer the original thing anymore. For example:
T x, y;
swap(x,w); // <=== compilation fails
A bigger issue is the question of whether the wrapper has value
semantics or not: do you pay to copy y into w, or do you merely refer
to y from w? The former can be costly, but the latter can be wrong.
I believe other cases of this general problem will crop up repeatedly
in real code.
-- Dave Abrahams BoostPro Computing http://www.boostpro.com
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk