to_number.h

Go to the documentation of this file.

 
#ifndef GUL17_TO_NUMBER_H_
#define GUL17_TO_NUMBER_H_
 
#include <array>
#include <cmath>
#include <cstdint>
#include <cstdlib>
#include <exception>
#include <limits>
#include <optional>
#include <string_view>
#include <type_traits>
 
#include "gul17/internal.h"
#include "gul17/substring_checks.h"
 
namespace gul17 {
 
namespace detail {
 
constexpr inline bool is_digit(char c) noexcept
{
    return c >= '0' && c <= '9';
}
 
constexpr inline bool is_nan_specifier(char c) noexcept
{
    if (c >= '0' && c <= '9')
        return true;
    if (c >= 'a' && c <= 'z')
        return true;
    if (c >= 'A' && c <= 'Z')
        return true;
    if (c == '_')
        return true;
    return false;
}
 
template <typename NumberType, bool count_magnitude = false>
constexpr inline std::optional<NumberType> to_unsigned_integer(std::string_view str,
        NumberType* magnitude = nullptr) noexcept
{
#ifndef __GNUC__
    constexpr NumberType max_tenth = std::numeric_limits<NumberType>::max() / 10;
#endif
 
    if (str.empty())
        return std::nullopt;
 
    NumberType result = 0;
 
    for (char c : str)
    {
        if (!is_digit(c))
            return std::nullopt;
 
#ifdef __GNUC__
        if (__builtin_mul_overflow(result, NumberType{ 10 }, &result)) // NOLINT(cppcoreguidelines-pro-type-vararg)
            return std::nullopt;
 
        if (__builtin_add_overflow(result, static_cast<NumberType>(c - '0'), &result)) // NOLINT(cppcoreguidelines-pro-type-vararg)
            return std::nullopt;
#else
        if (result > max_tenth)
            return std::nullopt;
 
        result *= 10;
 
        auto last = result;
 
        result += c - '0';
        if (result < last)
            return std::nullopt;
#endif
        if /*constexpr*/ (count_magnitude)
            *magnitude *= NumberType{ 10 };
    }
 
    return result;
}
 
/* Parse a signed exponent specifier.
 * May start with a leading sign ('+' or '-'). The exponent value is limited to
 * the range of int. The used range with a long double conversion is usually in
 * the range -5000 to 5000, so this is not really a limitation.
 */
constexpr std::optional<int> parse_exponent(std::string_view str) noexcept
{
    bool negative = false;
 
    switch (str.front())
    {
    case '+':
        str.remove_prefix(1);
        break;
    case '-':
        str.remove_prefix(1);
        negative = true;
        break;
    default:
        break;
    }
 
    auto opt_exp = to_unsigned_integer<int>(str);
 
    if (!opt_exp)
        return std::nullopt;
 
    if (negative)
        return -*opt_exp;
    return *opt_exp;
}
 
// For some 'long double' types with a big mantissa uint64 is not large enough.
// We resort to __uint128, which is a non standard extension in GCC and clang.
// But only if we need to.
// Note that on some compilers there are no std::numeric_limits<> for the extension
// type, and then asserts later on will fail. But usually that compilers have small
// long double types.
template <typename NumberType>
using FloatConversionIntType =
    typename std::conditional<
        (std::numeric_limits<std::uint64_t>::digits10 >= std::numeric_limits<NumberType>::digits10),
        std::uint64_t,
        #ifdef __SIZEOF_INT128__ // GCC, clang, intel
            __uint128_t
        #else
            std::uint64_t
        #endif
    >::type;
 
template <typename NumberType>
constexpr inline std::optional<NumberType> to_normalized_float(std::string_view i1, std::string_view i2) noexcept
{
    static_assert(std::numeric_limits<FloatConversionIntType<NumberType>>::digits10
            >= std::numeric_limits<NumberType>::digits10,
            "FloatConversionIntType is too small for NumberType");
 
    i1 = i1.substr(0, std::min(i1.length(),
                size_t(std::numeric_limits<FloatConversionIntType<NumberType>>::digits10)));
    i2 = i2.substr(0, std::min(i2.length(),
                size_t(std::numeric_limits<FloatConversionIntType<NumberType>>::digits10) - i1.length()));
 
    FloatConversionIntType<NumberType> accu{ 0 };
 
    auto magnitude = FloatConversionIntType<NumberType>{ 1 };
 
    if (not i2.empty()) {
        auto f2 = to_unsigned_integer<FloatConversionIntType<NumberType>, true>(i2, &magnitude);
        if (not f2.has_value())
            return std::nullopt;
        accu = *f2;
    }
    if (not i1.empty()) {
        auto i2_magnitude = magnitude;
        auto f1 = to_unsigned_integer<FloatConversionIntType<NumberType>, true>(i1, &magnitude);
        if (not f1.has_value())
            return std::nullopt;
        accu += (*f1 * i2_magnitude);
    }
 
    return static_cast<NumberType>(accu) / static_cast<NumberType>(magnitude / 10); // NOLINT(bugprone-integer-division): Precision loss is not possible with normalized accu
}
 
template <typename NumberType>
struct ParseInfNanResult {
    bool result_valid;
    std::optional<NumberType> result;
};
 
template <typename NumberType>
constexpr inline ParseInfNanResult<NumberType> parse_inf_nan(std::string_view str) noexcept
{
    auto const strlength = str.length();
    if (strlength == 0)
        return { true, {} };
 
    if (gul17::starts_with_nocase(str, "inf")) {
        if (strlength == 3 /* strlen("inf") */ )
            return { true, std::make_optional(std::numeric_limits<NumberType>::infinity()) };
        if (strlength == 8 /* strlen("infinity") */
                and gul17::starts_with_nocase(str.substr(3), "inity"))
            return { true, std::make_optional(std::numeric_limits<NumberType>::infinity()) };
        return { true, {} };
    }
 
    if (gul17::starts_with_nocase(str, "nan")) {
        if (strlength == 3 /* strlen("nan") */ )
            return { true, std::make_optional(std::numeric_limits<NumberType>::quiet_NaN()) };
        if (strlength < 5 /* strlen("nan()") */ or str[3] != '(' or str.back() != ')')
            return { true, {} };
        str.remove_prefix(4);
        str.remove_suffix(1);
        while (str.length()) {
            if (not is_nan_specifier(str.front()))
                return { true, {} };
            str.remove_prefix(1);
        }
        // We do not use the NaN specifier
        return { true, std::make_optional(std::numeric_limits<NumberType>::quiet_NaN()) };
    }
    return { false, {} };
}
 
GUL_EXPORT
long double pow10(int exponent);
 
template <typename NumberType>
constexpr inline std::optional<NumberType> to_unsigned_float(std::string_view str) noexcept
{
    auto inf_nan = parse_inf_nan<NumberType>(str);
    if (inf_nan.result_valid)
        return inf_nan.result;
 
    int exponent = 0;
    auto e_pos = str.find_first_of("eE");
    if (e_pos != std::string_view::npos)
    {
        if (e_pos + 1 == str.size())
            return std::nullopt;
 
        auto str_exponent = str.substr(e_pos + 1);
 
        str = str.substr(0, e_pos);
 
        auto opt_exp = detail::parse_exponent(str_exponent);
 
        if (!opt_exp)
            return std::nullopt;
 
        exponent = *opt_exp;
    }
 
    std::string_view str_before_point{ str };
    std::string_view str_after_point;
 
    auto point_pos = str.find('.');
    if (point_pos != std::string_view::npos)
    {
        str_before_point = str.substr(0, point_pos);
        str_after_point = str.substr(point_pos + 1);
    }
 
    if (str_before_point.empty() && str_after_point.empty())
        return std::nullopt;
 
    // Get rid of leading zeros
    while (!str_before_point.empty() and str_before_point[0] == '0')
        str_before_point.remove_prefix(1);
 
    // Normalize the number
    if (str_before_point.empty()) {
        auto const old_digits = str_after_point.length();
        while (!str_after_point.empty() and str_after_point[0] == '0')
            str_after_point.remove_prefix(1);
 
        if (str_after_point.empty())
            return { 0 };
 
        str_before_point = str_after_point.substr(0, 1);
        str_after_point.remove_prefix(1);
        exponent -= static_cast<int>(old_digits - str_after_point.length());
    } else {
        exponent += static_cast<int>(str_before_point.length() - 1);
    }
 
    // Now the incoming number string is like this:
    // "s.tr_before_point" "str_after_point" E exponent
    //   ^                                           ^
    //   | here is the decimal dot, virtually        | corrected exponent
 
    using long_double = long double;
    using CalcType = std::conditional_t<
        std::greater<>()(sizeof(NumberType), sizeof(double)),
        long_double, double>;
 
    auto norm_val = to_normalized_float<CalcType>(str_before_point, str_after_point);
    if (not norm_val.has_value())
        return std::nullopt;
 
    return static_cast<NumberType>(detail::pow10(exponent) * *norm_val);
}
 
template <typename NumberType>
inline std::optional<NumberType> strtold_wrapper(std::string_view str) noexcept
{
    if (str.empty())
        return std::nullopt;
 
    try
    {
        auto input = std::string{ str };
        char* process_end;
        auto value = static_cast<NumberType>(std::strtold(input.c_str(), &process_end));
 
        if (input.data() + input.size() != process_end) // NOLINT(cppcoreguidelines-pro-bounds-pointer-arithmetic): Pointer arithmetic needed because strtold gives pointer back
            return std::nullopt;
        return value;
    }
    catch (const std::exception &)
    {
        return std::nullopt;
    }
}
 
} // namespace detail
 
 
// Overload for unsigned integer types.
template <typename NumberType>
constexpr inline std::enable_if_t<std::is_integral<NumberType>::value and
                                  std::is_unsigned<NumberType>::value,
                                  std::optional<NumberType>>
to_number(std::string_view str) noexcept
{
    return detail::to_unsigned_integer<NumberType>(str);
}
 
// Overload for signed integer types.
template <typename NumberType>
constexpr inline std::enable_if_t<std::is_integral<NumberType>::value and
                                  std::is_signed<NumberType>::value,
                                  std::optional<NumberType>>
to_number(std::string_view str) noexcept
{
    if (str.empty())
        return std::nullopt;
 
    if (str.front() == '-')
    {
        using UnsignedT = std::make_unsigned_t<NumberType>;
        constexpr auto max_abs_negative_value =
            static_cast<UnsignedT>(std::numeric_limits<NumberType>::max()) + 1;
 
        str.remove_prefix(1);
 
        auto result = detail::to_unsigned_integer<UnsignedT>(str);
        if (!result)
            return std::nullopt;
 
        if (*result == max_abs_negative_value)
            return std::numeric_limits<NumberType>::lowest();
        else if (*result > max_abs_negative_value)
            return std::nullopt;
 
        return static_cast<NumberType>(-static_cast<NumberType>(*result));
    }
 
    return detail::to_unsigned_integer<NumberType>(str);
}
 
// Overload for floating-point types.
template <typename NumberType>
constexpr inline std::enable_if_t<std::is_floating_point<NumberType>::value,
                                  std::optional<NumberType>>
to_number(std::string_view str) noexcept
{
    if (str.empty())
        return std::nullopt;
 
    if (
#ifdef _MSC_VER
#    pragma warning( push )
#    pragma warning( disable: 4127 ) // conditional expression is constant
#endif
            (std::numeric_limits<detail::FloatConversionIntType<NumberType>>::digits10
            <= std::numeric_limits<NumberType>::digits10)) {
#ifdef _MSC_VER
#    pragma warning( pop )
#endif
        // Too big for our approach. Resort to non-constexpr functionality.
        // This actually never happenes with the currently supported platforms / compilers.
        // (Except long double on Darwin)
        return detail::strtold_wrapper<NumberType>(str);
    }
 
    if (str.front() == '-')
    {
        str.remove_prefix(1);
        auto result = detail::to_unsigned_float<NumberType>(str);
        if (!result)
            return std::nullopt;
        return -*result;
    }
 
    return detail::to_unsigned_float<NumberType>(str);
}
 
// Overload for bool
template<>
constexpr inline std::optional<bool> to_number<bool>(std::string_view str) noexcept
{
    if (str.length() == 1) {
        if (str[0] == '1')
            return true;
        if (str[0] == '0')
            return false;
        return std::nullopt;
    }
    if (equals_nocase(str, "true"))
        return true;
 
    if (equals_nocase(str, "false"))
        return false;
 
    return std::nullopt;
}
 
 
} // namespace gul17
 
#endif
 
// vi:ts=4:sw=4:et:sts=4