Boost C++ Libraries: Ticket #11012: chrono_io v2 fail to compile with boost::chrono::duration< boost::rational >

attachment set

Tue, 10 Feb 2015 15:52:57 GMT

attachment → chrono_rational.tgz

unbuildable example

status, component changed; owner set

viboes — Fri, 13 Feb 2015 18:27:39 GMT

owner set to viboes
status new → assigned
component None → chrono

Why do you want the following to compile?

 boost::chrono::duration< boost::rational<int> >

duration expects an integral type.

Mon, 16 Feb 2015 08:26:50 GMT

Hello Vicente,

thank you for your answer.

Replying to viboes:

Why do you want the following to compile?
 boost::chrono::duration< boost::rational<int> >

Well, this specific line does compile, and even the io part did compile under version 1. The compilation only fails with io version 2.

I expected the compilation not to break between versions.

Moreover, the page en.cppreference.com/w/cpp/chrono/duration documents rep as:

Rep, an arithmetic type representing the number of ticks

I'm not sure what "arithmetic type" precisely means in this context, but using a rational type, which supports the arithmetic operations, seemed legit (and does actually work, except for io version 2).

Note that I just discovered this trait class: en.cppreference.com/w/cpp/types/is_arithmetic, maybe there is a precise definition of what an arithmetic type is somewhere in the standard.

Replying to viboes:

duration expects an integral type.

That seems untrue: there is explicit support for floating point numbers.

I guess the crux here is to first understand whether or not std::chrono supports boost::rational-like types as Rep, and then support the same types for the io part.

viboes — Mon, 16 Feb 2015 14:55:36 GMT

My bad. You are right and std::chrono should support boost::rational.

As you mention Boost.Chrono IO V2 doesn't supports it.

V1 needed just that the Rep supports streaming

    return os << d.count() << ' ' << f.template name<Period>(d.count());

V2 request num_put facet.

        return std::use_facet<std::num_put<CharT, iter_type> >(ios.getloc()).put(s, ios, fill,
            static_cast<long int> (d.count()));
            //d.count());

When I comment/uncomment the previous lines I got an ambiguous call to put. This was a bug, and when I replace your example by

    typedef boost::chrono::duration< float > RationalSeconds;
    RationalSeconds d(0.5);
    std::cout << d << std::endl;

The decimal part is truncated. I get

0 seconds

instead of

0.5 seconds

I have added a trait that promotes int to long int

    namespace detail
    {
      template <class T>
      struct propagate {
        typedef T type;
      };
      template <>
      struct propagate<boost::int_least32_t> {
        typedef boost::int_least64_t type;
      };
    }

and used it instead of the cast

            static_cast<long int> (d.count()));

as follows

            static_cast<typename detail::propagate<Rep>::type> (d.count()));

This will however not take in account boost::rational as a valid representation.

There are other places where I have added more constraints to the representation.

        return do_get_unit(style, rt_ratio(Period()), static_cast<intmax_t>(d.count()));

I 'll need to rethink how V2 can take care of IO of Rep not supported by the facets num_get/num_put and many other issues. :(

viboes — Mon, 16 Feb 2015 14:58:32 GMT

https://github.com/boostorg/chrono/commit/300c0c69631de7dcc9e555bdc8ff217d9a25bb88

Mon, 16 Feb 2015 16:54:01 GMT

Vincente,

many thanks for your work on chrono (I should have said that first)

Replying to viboes:

I 'll need to rethink how V2 can take care of IO of Rep not supported by the facets num_get/num_put and many other issues. :(

yes, good luck for that.

Cheers

milestone changed

viboes — Wed, 18 Feb 2015 18:12:54 GMT

milestone To Be Determined → Boost 1.58.0

You are welcome.

I have found a workaround. While not completely satisfactory from the design point of view, it seems to work.

Please, could you take a look at https://github.com/boostorg/chrono/commit/83b881fad4ecb539c63357c3482c95daf2ca7388

viboes — Wed, 18 Feb 2015 18:30:44 GMT

While the last commit works for stream inserter (<<), I don't have yet a clear idea on how to take care of stream extractor (>>).

Wed, 18 Feb 2015 21:35:16 GMT

Thank you for the patch

Replying to viboes:

Please, could you take a look at [...]

I commented there.

Wed, 18 Feb 2015 21:36:47 GMT

Replying to viboes:

Please, could you take a look at ..

I commented there. Cheers

status changed; resolution set

viboes — Mon, 09 Mar 2015 07:25:26 GMT

status assigned → closed
resolution → fixed

I will close this issue, even if the design is not clean.