Subject: Re: [boost] [core] Breaking change to boost::ref in 1.56
From: Andrey Semashev (andrey.semashev_at_[hidden])
Date: 2014-07-14 05:06:44
On Monday 14 July 2014 11:08:01 Peter Dimov wrote:
> Andrey Semashev wrote:
> > I think Boost.Proto is one of the examples where undesired reference
> > recursion happened.
> Boost.Proto just uses cref in a generic context, where its argument
> sometimes happens to be a reference_wrapper. That's not undesired.
Generic code can still be written with the trait I proposed. What's undesired
is the reference recursiveness, especially if accidental.
> > If not because of performance, conceptually there are no references to
> > references in C++, and std::ref attempts to reflect that.
> There are references to reference_wrapper though.
Yes, you can also manually instantiate
reference_wrapper<reference_wrapper<T>>. Nothing's perfect. :)
> > Now that Boost.Proto does reference collapsing, reverting the change will
> > break it again (I think).
> That's possible, but I don't think so.
> I suspect that you don't actually have a motivating example in mind.
That's true, I don't have a practical example at hand. But neither I can see
why recursive references could be useful. With this kind of choice I prefer
reference collapsing, at least because it looks more natural and std-like.
> reference collapsing, how would you take advantage of it? What code have you
> written, or will write, that needs it?
In my probably uneducated view, generic libraries such as Proto or Phoenix
should do reference collapsing. If not by the means of boost::ref then
manually themselves. Whenever the user provides a reference_wrapper and the
library intends to save it by reference (e.g. into the expression tree),
reference collapsing should happen.
> Do you realize that the simple use
> #include <boost/ref.hpp>
> int main()
> int x = 0;
> boost::reference_wrapper<int> r = boost::ref( x );
> boost::reference_wrapper< boost::reference_wrapper<int> > r2 =
> boost::ref( r );
> boost::reference_wrapper<int> r3 = boost::ref( r );
> actually works without the overloads? They only serve to break r2.
I realize that the code will break but I disagree that the overloads only
serve to break it. Why would you want r2? I mean, in what case would you
intentionally write code to employ recursive references?