parsing strings with value modifiers ('-', '%') at the end

Question

I try to get to grips with parsing.

I have some data that comes in a de-de format with additional information at the end of the string.

I managed to get the de-de part correct but I struggle in getting the - and % parsed correctly. I read up on codecvt but I do not understand the topic.

Here is a reflection of what I understand so far and an example of what I need to do.

#include <string>
#include <locale>
#include <iostream>
#include <sstream>

using namespace std;

#define EXPECT_EQ(actual, expected) { \
    if (actual != expected) \
    { \
        cout << "expected " << #actual << " to be " << expected << " but was " << actual << endl; \
    } \
}

double parse(wstring numstr)
{
    double value;
    wstringstream is(numstr);
    is.imbue(locale("de-de"));
    is >> value;
    return value;
}

int main()
{
    EXPECT_EQ(parse(L"123"), 123); //ok
    EXPECT_EQ(parse(L"123,45"), 123.45); //ok
    EXPECT_EQ(parse(L"1.000,45"), 1000.45); //ok
    EXPECT_EQ(parse(L"2,390%"), 0.0239); //% sign at the end
    EXPECT_EQ(parse(L"1.234,56-"), -1234.56); //- sign at the end
}

The output is:

expected parse(L"2,390%") to be 0.0239 but was 2.39
expected parse(L"1.234,56-") to be -1234.56 but was 1234.56

How can I imbue my stream so that it reads the - and % sign like I need it to?

Answer 1

The codecvt facet is the wrong place to look here. The codecvt facet is only intended to deal with converting an external representation of a character into an internal representation of the same character (e.g., UTF-8 in the file, UTF-32/UCS-4 internally).

For parsing numbers like this, you're looking for the num_get facet. The basic idea is that you'll create a class derived from std::num_get that overrides do_get for (at least) the types of numbers you care about.

In a typical case, you only do a "real" implementation for a few types (e.g., long long and long double) and have the functions for all the smaller types delegate to those, then convert the result to the target type.

Here's a fairly simple num_get facet. For the moment, it only attempts to provide the special processing for type double. To keep the example from getting too outrageously long, I've simplified the processing a bit:

• It doesn't try to parse exponents on the numbers (e.g., the '99' in 1e99).
• It doesn't try to deal with a suffix of %- (but will do -%).
• It's hard-coded to treat ',' as the decimal point and '.' as the thousands separator.
• It makes no attempt at sanity checking thousands separators. e.g., 1,,,3 will parse as 13.

Within those limitations here's some code:

#include <ios>
#include <string>
#include <locale>
#include <iostream>
#include <sstream>
#include <iterator>
#include <cctype>

using namespace std;

template <class charT, class InputIterator = istreambuf_iterator<charT> >
class read_num : public std::num_get < charT > {
public:
    typedef charT char_type;
    typedef InputIterator iter_type;
protected:
    iter_type do_get(iter_type in, iter_type end, ios_base& str, ios_base::iostate& err, double& val) const {
        double ret = 0.0;

        bool negative = false;
        using uc = std::make_unsigned<charT>::type;

        while (std::isspace((uc)*in))
            ++in;
        if (*in == '-') {
            negative = true;
            ++in;
            while (std::isspace((uc)*in))
                ++in;
        }
        while (std::isdigit((uc)*in)) {
            ret *= 10;
            ret += *in - '0';
            ++in;
            if (*in == '.')
                ++in;
        }
        if (*in == ',') {
            ++in;
            double place = 10.0;
            while (std::isdigit((uc)*in)) {
                ret += (*in - '0') / place;
                place *= 10;
                ++in;
            }
        }
        if (*in == '-') {
            negative = true;
            ++in;
        }
        if (*in == '%') {
            ret /= 100.0;
            ++in;
        }
        if (negative)
            ret = -ret;
        val = ret;
        return in;
    }
};

Realistically, under the circumstances you probably don't want to do things this way-- you probably want to delegate to the existing facet to read the number proper, then at the end of what it parses, look for a - and/or % and react appropriately (and probably diagnose an error if, for example, you find both leading and trailing '-').

Answer 2

I'd tackle this head-on: let's get to grips with parsing here.

You'd end up writing that somewhere anyways, so I'd forget about the need to create an (expensive) string stream first.

Weapon Of Choice: Boost Spirit

Note,

• I parse the string using it's iterators directly. My code is pretty generic as to the type of floating point number used.

• You can pretty much search replace double by e.g. boost::multiprecision::cpp_dec_float (or make it a template argument) and be parsing. Because I predict that you needed to parser decimal floating point numbers, not binary floating point numbers. You're losing accuracy in the conversion.

UPDATE: extended sample Live On Coliru

The Simple Grammar

At it's core, the grammar is really simple:

if (parse(numstr.begin(), numstr.end(), mynum >> matches['-'] >> matches['%'],
            value, sign, pct)) 
{
    if (sign) value = -value;
    if (pct)  value /= 100;

    return value;
}

There you have it. Of couse, we need to define mynum so it parses the unsigned real numbers as expected:

using namespace qi;
real_parser<double, de_numpolicy<double> > mynum;

The Magic: real_policies<>

The documentation goes a long way to explaining how to tweak real number parsing using real_policies. Here's the policy I came up with:

template <typename T>
    struct de_numpolicy : qi::ureal_policies<T>
{
    //  No exponent
    template <typename It>                static bool parse_exp(It&, It const&)          { return false; } 
    template <typename It, typename Attr> static bool parse_exp_n(It&, It const&, Attr&) { return false; } 

    //  Thousands separated numbers
    template <typename It, typename Attr>
    static bool parse_n(It& first, It const& last, Attr& attr)
    {
        qi::uint_parser<unsigned, 10, 1, 3> uint3;
        qi::uint_parser<unsigned, 10, 3, 3> uint3_3;

        if (parse(first, last, uint3, attr)) {
            for (T n; qi::parse(first, last, '.' >> uint3_3, n);)
                attr = attr * 1000 + n;

            return true;
        }

        return false;
    }

    template <typename It>
        static bool parse_dot(It& first, It const& last) {
            if (first == last || *first != ',')
                return false;
            ++first;
            return true;
        }
};

Full Demo

Live On Coliru

#include <boost/spirit/include/qi.hpp>
#include <iostream>


#define EXPECT_EQ(actual, expected) { \
    double v = (actual); \
    if (v != expected) \
    { \
        std::cout << "expected " << #actual << " to be " << expected << " but was " << v << std::endl; \
    } \
}

namespace mylib {
    namespace qi = boost::spirit::qi;

    template <typename T>
        struct de_numpolicy : qi::ureal_policies<T>
    {
        //  No exponent
        template <typename It>                static bool parse_exp(It&, It const&)          { return false; } 
        template <typename It, typename Attr> static bool parse_exp_n(It&, It const&, Attr&) { return false; } 

        //  Thousands separated numbers
        template <typename It, typename Attr>
        static bool parse_n(It& first, It const& last, Attr& attr)
        {
            qi::uint_parser<unsigned, 10, 1, 3> uint3;
            qi::uint_parser<unsigned, 10, 3, 3> uint3_3;

            if (parse(first, last, uint3, attr)) {
                for (T n; qi::parse(first, last, '.' >> uint3_3, n);)
                    attr = attr * 1000 + n;

                return true;
            }

            return false;
        }

        template <typename It>
            static bool parse_dot(It& first, It const& last) {
                if (first == last || *first != ',')
                    return false;
                ++first;
                return true;
            }
    };

    template<typename Char, typename CharT, typename Alloc>
    double parse(std::basic_string<Char, CharT, Alloc> const& numstr)
    {
        using namespace qi;
        real_parser<double, de_numpolicy<double> > mynum;

        double value;
        bool sign, pct;

        if (parse(numstr.begin(), numstr.end(), mynum >> matches['-'] >> matches['%'],
                    value, sign, pct)) 
        {
            // std::cout << "DEBUG: " << std::boolalpha << " '" << numstr << "' -> (" << value << ", " << sign << ", " << pct << ")\n";
            if (sign) value = -value;
            if (pct)  value /= 100;

            return value;
        }

        assert(false); // TODO handle errors
    }

} // namespace mylib

int main()
{
    EXPECT_EQ(mylib::parse(std::string("123")),       123);      // ok
    EXPECT_EQ(mylib::parse(std::string("123,45")),    123.45);   // ok
    EXPECT_EQ(mylib::parse(std::string("1.000,45")),  1000.45);  // ok
    EXPECT_EQ(mylib::parse(std::string("2,390%")),    0.0239);   // %  sign at the end
    EXPECT_EQ(mylib::parse(std::string("1.234,56-")), -1234.56); // -  sign at the end
}

If you uncomment the "DEBUG" line, it prints:

DEBUG:  '123' -> (123, false, false)
DEBUG:  '123,45' -> (123.45, false, false)
DEBUG:  '1.000,45' -> (1000.45, false, false)
DEBUG:  '2,390%' -> (2.39, false, true)
DEBUG:  '1.234,56-' -> (1234.56, true, false)