Re: [Boost-users] Forward declarations and iterator_facade<>

Subject: Re: [Boost-users] Forward declarations and iterator_facade<>
From: David Abrahams (dave_at_[hidden])
Date: 2010-10-17 05:10:34

At Fri, 15 Oct 2010 16:49:40 -0600,
OvermindDL1 wrote:
> 
> On Fri, Oct 15, 2010 at 11:02 AM, David Abrahams <dave_at_[hidden]> wrote:
> > At Fri, 15 Oct 2010 17:24:42 +0100,
> > Phil Endecott wrote:
> >>
> >> Dear Experts,
> >>
> >> I have code something like this:
> >>
> >> struct S;
> >>
> >> class iter:
> >> Â  Â public boost::iterator_facade<iter,S,......>
> >> { ... };
> >>
> >> struct S {
> >> Â  Â ...
> >> };
> >>
> >>
> >> This used to work, but stopped recently; I believe it may have been
> >> after upgrading gcc, not after upgrading Boost itself, but I'm
> >> uncertain about that. Â The problem is that something inside
> >> iterator_facade calls boost::is_POD<S>, which (not unreasonably)
> >> doesn't seem to work for the forward-declared S.
> >
> > I guess you were just lucky, because we never tested for or intended
> > to provide that capability.
> >
> >> Re-ordering the code to remove the circular dependencies will be
> >> painful, so if anyone has any other hints I would love to hear them.
> >
> > I'm afraid there's not going to be any other choice for you unless we
> > are somehow able to add that capability in Boost and you are able to
> > wait for it.
> 
> Shouldn't he be able to just specialize boost::is_pod on his forward
> declared type as true?

Yeah (or false If it's not actually POD).  That would be a perfectly
legit thing to do; good idea.

namespace boost
{
  template <>
  struct is_pod<MyPod> : boost::mpl::true_ {};

  template <>
  struct is_pod<MyNonPod> : boost::mpl::false_ {};
}

It might even work.  No guarantees of course, but it's a pretty good
bet.

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com