Subject: Re: [boost] attempting sandbox svg_plot templates
From: Paul A. Bristow (pbristow_at_[hidden])
Date: 2012-08-25 05:50:00
> -----Original Message-----
> From: boost-bounces_at_[hidden] [mailto:boost-bounces_at_[hidden]] On Behalf Of Jan Hudec
> Sent: Friday, August 24, 2012 7:04 PM
> To: boost_at_[hidden]
> Subject: Re: [boost] attempting sandbox svg_plot templates
>
> On Tue, Aug 21, 2012 at 15:09:02 -0400, Roger Martin wrote:
> > Run into many template and friend issues. Anybody getting this lib to work?
> >
> > Much coming from
> https://svn.boost.org/svn/boost/sandbox/SOC/2007/visualization/boost/svg_plot/uncertain.hpp.
> > First two:
> > ..............
> > In file included from
> > ../../testing-grounds/divcon-support_libs/local/r.linux-x86_64.gcc-sha
> > red/include/boost/svg_plot/svg_2d_plot.hpp:32:0,
> > from tests/plotting_tests.cpp:21:
> > ../../testing-grounds/divcon-support_libs/local/r.linux-x86_64.gcc-
> shared/include/boost/svg_plot/uncertain.hpp:241:11:
> > required from here
> > ../../testing-grounds/divcon-support_libs/local/r.linux-x86_64.gcc-
> shared/include/boost/svg_plot/uncertain.hpp:97:24:
> > error: template-id ‘operator<< <>’ for ‘std::ostream&
> > boost::svg::operator<<(std::ostream&, const boost::svg::unc<false>&)’
> > does not match any template declaration
> > ../../testing-grounds/divcon-support_libs/local/r.linux-x86_64.gcc-
> shared/include/boost/svg_plot/uncertain.hpp:102:24:
> > error: template-id ‘operator<< <>’ for ‘std::ostream&
> > boost::svg::operator<<(std::ostream&, const
> > std::pair<boost::svg::unc<false>, boost::svg::unc<false> >&)’ does not
> > match any template declaration ..............
>
> It's making friends with templates that don't exist. I don't think the involved templates can work in any
> compler, ever.
>
> There is however big difference between Gcc and Visual C++ in treating of templates. Gcc parses
> everything as it comes across it and insists that it's well-formed, while VC++ delays lot of it until
> instantiation, apparently in an attempt to relieve programmers from having to write `typename` and
> `template` in many cases. That means Gcc will complain about things like this while VC++ won't.
>
> However the templates seem pretty bogus. There is a
>
> template <bool correlated> class unc;
>
> followed up by
>
> template<typename correlated> std::ostream& operator<< (std::ostream& os, const unc<false>& u);
>
> followed up by
>
> friend std::ostream& operator<< <> (std::ostream& os, const unc<correlated>& u);
>
> and that simply does not add up under any compiler. The middle template is not using it's template
> argument in it's signature, so it's neither possible to give it's specialization as the friend declaration is
> trying to do, nor is it actually useful, because you'd need to give the template parameter explicitly, so the
> only possible call syntax is
>
> operator<< <whatever> (out, u)
>
> and that certainly does not look like what was intended.
>
> So I suggest you just try to comment the offending declarations out and see if it gets you somewhere.
I'm sure you are right - but MSVC seems to swallow it and it works.
Friends sometime seem more trouble than they are worth ;-)
But I'm getting round to being able to work with both compilers, aided by the original 'complainant' Roger Martin.
Watch this space...
Paul