Boost Karma - non consuming predicate

Question

I need to print a std::complex but omitting imaginary part if it's equal zero. So I have a rule with two productions:

karma::rule<OutputIterator, std::complex<double>()> complexRule = 
    '(' << double_ << ", " double_ << ')'
    | double_ << omit[double_];

This way Karma will always choose the first production, so I need some kind of predicate which will make a decission. Boost Karma tutorial comes with that solution which requires adapting std::complex as a three element tuple.

BOOST_FUSION_ADAPT_ADT(
    std::complex<double>,
    (bool, bool, obj.imag() != 0, /**/)
    (double, double, obj.real(), /**/)
    (double, double, obj.imag(), /**/)
)

but unfortunately I cannot do that since other code is using std::complex adapted as two element tuple. Is there a way to solve that problem without adding predicate straight into Fusion adapter?

I was trying to use karma::eps generator as a predicate

auto rule = eps( ... ) << '(' << double_ << ", " << double_ << ')'
          | double_ << omit[double_];

but I don't know what Phoenix expression should I put inside eps( ... ), and as Epsilon Generator doesn't consume any attribute I'm not sure if it is possible to access std::complex from it?

Answer 1

I'd personally stay away from adapting this as a sequence (I'm not sure how you adapted it as a two-element fusion sequence in the first place).

However it's done, it won't be generic (so you'll have use separate adaptations for different type arguments (float, double, long double, boost::multiprecision::number<boost::multiprecision::cpp_dec_float<50>> etc.).

This seems like a job for Spirit's customization points:

namespace boost { namespace spirit { namespace traits {

    template <typename T>
        struct extract_from_attribute<typename std::complex<T>, boost::fusion::vector2<T, T>, void>
        {
            typedef boost::fusion::vector2<T,T> type;

            template <typename Context>
                static type call(std::complex<T> const& attr, Context& context)
                {
                    return { attr.real(), attr.imag() };
                }
        };

} } }

Now you can just use any std::complex<T> with a rule/expression expecting fusion sequence:

 rule = 
        '(' << karma::double_ << ", " << karma::duplicate [ !karma::double_(0.0) << karma::double_ ] << ')' 
      | karma::double_ << karma::omit [ karma::double_ ];

Note how

• I used duplicate[] to test for 0.0 before emitting the output
• On the other branch I used omit to consume the imaginary part without displaying anything

Here's a full demo, Live On Coliru

#include <boost/spirit/include/karma.hpp>
#include <complex>

namespace boost { namespace spirit { namespace traits {

    template <typename T>
        struct extract_from_attribute<typename std::complex<T>, boost::fusion::vector2<T, T>, void>
        {
            typedef boost::fusion::vector2<T,T> type;

            template <typename Context>
                static type call(std::complex<T> const& attr, Context& context)
                {
                    return { attr.real(), attr.imag() };
                }
        };

} } }

namespace karma = boost::spirit::karma;

int main()
{
    karma::rule<boost::spirit::ostream_iterator, boost::fusion::vector2<double, double>()> 
        static const rule = 
                            '(' << karma::double_ << ", " << karma::duplicate [ !karma::double_(0.0) << karma::double_ ] << ')' 
                          | karma::double_ << karma::omit [ karma::double_ ];

    std::vector<std::complex<double>> const values {
                { 123, 4 },
                { 123, 0 },
                { 123, std::numeric_limits<double>::infinity() },
                { std::numeric_limits<double>::quiet_NaN(), 0 },
                { 123, -1 },
            };

    std::cout << karma::format_delimited(*rule, '\n', values);
}

Output:

(123.0, 4.0)
123.0
(123.0, inf)
nan
(123.0, -1.0)