🔀 merged #379 and fixed conflicts
This commit is contained in:
commit
8d88a1da2a
3 changed files with 481 additions and 281 deletions
367
src/json.hpp
367
src/json.hpp
|
@ -10838,59 +10838,202 @@ basic_json_parser_66:
|
||||||
return result;
|
return result;
|
||||||
}
|
}
|
||||||
|
|
||||||
/*!
|
|
||||||
@brief parse floating point number
|
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
|
||||||
standard floating point number parsing function based on the type
|
|
||||||
supplied via the first parameter. Set this to @a
|
|
||||||
static_cast<number_float_t*>(nullptr).
|
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
|
||||||
the number
|
|
||||||
|
|
||||||
@return the floating point number
|
|
||||||
*/
|
|
||||||
long double str_to_float_t(long double* /* type */, char** endptr) const
|
|
||||||
{
|
|
||||||
return std::strtold(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
|
||||||
}
|
|
||||||
|
|
||||||
/*!
|
/*!
|
||||||
@brief parse floating point number
|
@brief parse string into a built-in arithmetic type as if
|
||||||
|
the current locale is POSIX.
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
note: in floating-point case strtod may parse
|
||||||
standard floating point number parsing function based on the type
|
past the token's end - this is not an error.
|
||||||
supplied via the first parameter. Set this to @a
|
|
||||||
static_cast<number_float_t*>(nullptr).
|
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
any leading blanks are not handled.
|
||||||
the number
|
|
||||||
|
|
||||||
@return the floating point number
|
|
||||||
*/
|
*/
|
||||||
double str_to_float_t(double* /* type */, char** endptr) const
|
struct strtonum
|
||||||
{
|
{
|
||||||
return std::strtod(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
public:
|
||||||
}
|
strtonum(const char* start, const char* end)
|
||||||
|
: m_start(start), m_end(end)
|
||||||
|
{}
|
||||||
|
|
||||||
/*!
|
/// return true iff parsed successfully as
|
||||||
@brief parse floating point number
|
/// number of type T.
|
||||||
|
///
|
||||||
|
/// @val shall contain parsed value, or
|
||||||
|
/// undefined value if could not parse.
|
||||||
|
template<typename T,
|
||||||
|
typename = typename std::enable_if<
|
||||||
|
std::is_arithmetic<T>::value>::type >
|
||||||
|
bool to(T& val) const
|
||||||
|
{
|
||||||
|
return parse(val, std::is_integral<T>());
|
||||||
|
}
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
/// return true iff token matches ^[+-]\d+$
|
||||||
standard floating point number parsing function based on the type
|
///
|
||||||
supplied via the first parameter. Set this to @a
|
/// this is a helper to determine whether to
|
||||||
static_cast<number_float_t*>(nullptr).
|
/// parse the token into floating-point or
|
||||||
|
/// integral type. We wouldn't need it if
|
||||||
|
/// we had separate token types for integral
|
||||||
|
/// and floating-point cases.
|
||||||
|
bool is_integral() const
|
||||||
|
{
|
||||||
|
const char* p = m_start;
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
if (!p)
|
||||||
the number
|
{
|
||||||
|
return false; // LCOV_EXCL_LINE
|
||||||
|
}
|
||||||
|
|
||||||
@return the floating point number
|
if (*p == '-' or * p == '+')
|
||||||
*/
|
{
|
||||||
float str_to_float_t(float* /* type */, char** endptr) const
|
++p;
|
||||||
{
|
}
|
||||||
return std::strtof(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
|
||||||
}
|
if (p == m_end)
|
||||||
|
{
|
||||||
|
return false; // LCOV_EXCL_LINE
|
||||||
|
}
|
||||||
|
|
||||||
|
while (p < m_end and* p >= '0'
|
||||||
|
and * p <= '9')
|
||||||
|
{
|
||||||
|
++p;
|
||||||
|
}
|
||||||
|
|
||||||
|
return p == m_end;
|
||||||
|
}
|
||||||
|
|
||||||
|
private:
|
||||||
|
const char* const m_start = nullptr;
|
||||||
|
const char* const m_end = nullptr;
|
||||||
|
|
||||||
|
// overloaded wrappers for strtod/strtof/strtold
|
||||||
|
// that will be called from parse<floating_point_t>
|
||||||
|
|
||||||
|
static void strtof(float& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtof(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
static void strtof(double& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtod(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
static void strtof(long double& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtold(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
template<typename T>
|
||||||
|
bool parse(T& value, /*is_integral=*/std::false_type) const
|
||||||
|
{
|
||||||
|
// replace decimal separator with locale-specific
|
||||||
|
// version, when necessary; data will point to
|
||||||
|
// either the original string, or buf, or tempstr
|
||||||
|
// containing the fixed string.
|
||||||
|
std::string tempstr;
|
||||||
|
std::array<char, 64> buf;
|
||||||
|
const size_t len = static_cast<size_t>(m_end - m_start);
|
||||||
|
|
||||||
|
// Since dealing with strtod family of functions,
|
||||||
|
// we're getting the decimal point char from the
|
||||||
|
// C locale facilities instead of C++'s numpunct
|
||||||
|
// facet of the current std::locale;
|
||||||
|
const auto loc = localeconv();
|
||||||
|
assert(loc != nullptr);
|
||||||
|
const char decimal_point_char = !loc->decimal_point ? '.'
|
||||||
|
: loc->decimal_point[0];
|
||||||
|
|
||||||
|
const char* data = m_start;
|
||||||
|
|
||||||
|
if (decimal_point_char != '.')
|
||||||
|
{
|
||||||
|
const size_t ds_pos = static_cast<size_t>(
|
||||||
|
std::find(m_start, m_end, '.') - m_start );
|
||||||
|
|
||||||
|
if (ds_pos != len)
|
||||||
|
{
|
||||||
|
// copy the data into the local buffer or
|
||||||
|
// tempstr, if buffer is too small;
|
||||||
|
// replace decimal separator, and update
|
||||||
|
// data to point to the modified bytes
|
||||||
|
if (len + 1 < buf.size())
|
||||||
|
{
|
||||||
|
std::copy(m_start, m_end, buf.data());
|
||||||
|
buf[len] = 0;
|
||||||
|
buf[ds_pos] = decimal_point_char;
|
||||||
|
data = buf.data();
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
tempstr.assign(m_start, m_end);
|
||||||
|
tempstr[ds_pos] = decimal_point_char;
|
||||||
|
data = tempstr.c_str();
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
char* endptr = nullptr;
|
||||||
|
value = 0;
|
||||||
|
// this calls appropriate overload depending on T
|
||||||
|
strtof(value, data, &endptr);
|
||||||
|
|
||||||
|
// note that reading past the end is OK, the data may be,
|
||||||
|
// for example, "123.", where the parsed token only
|
||||||
|
// contains "123", but strtod will read the dot as well.
|
||||||
|
const bool ok = endptr >= data + len
|
||||||
|
and len > 0;
|
||||||
|
|
||||||
|
if (ok and value == 0.0 and * data == '-')
|
||||||
|
{
|
||||||
|
// some implementations forget to negate the zero
|
||||||
|
value = -0.0;
|
||||||
|
}
|
||||||
|
|
||||||
|
return ok;
|
||||||
|
}
|
||||||
|
|
||||||
|
signed long long parse_integral(
|
||||||
|
char** endptr,
|
||||||
|
/*is_signed*/std::true_type) const
|
||||||
|
{
|
||||||
|
return std::strtoll(m_start, endptr, 10);
|
||||||
|
}
|
||||||
|
|
||||||
|
unsigned long long parse_integral(
|
||||||
|
char** endptr,
|
||||||
|
/*is_signed*/std::false_type) const
|
||||||
|
{
|
||||||
|
return std::strtoull(m_start, endptr, 10);
|
||||||
|
}
|
||||||
|
|
||||||
|
template<typename T>
|
||||||
|
bool parse(T& value, /*is_integral=*/std::true_type) const
|
||||||
|
{
|
||||||
|
char* endptr = nullptr;
|
||||||
|
errno = 0; // these are thread-local
|
||||||
|
const auto x = parse_integral(&endptr, std::is_signed<T>());
|
||||||
|
|
||||||
|
static_assert(std::is_signed<T>() // called right overload?
|
||||||
|
== std::is_signed<decltype(x)>(), "");
|
||||||
|
|
||||||
|
value = static_cast<T>(x);
|
||||||
|
|
||||||
|
return x == static_cast<decltype(x)>(value) // x fits into destination T
|
||||||
|
and (x < 0) == (value < 0) // preserved sign
|
||||||
|
and (x != 0 or is_integral()) // strto[u]ll did nto fail
|
||||||
|
and errno == 0 // strto[u]ll did not overflow
|
||||||
|
and m_start < m_end // token was not empty
|
||||||
|
and endptr == m_end; // parsed entire token exactly
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
/*!
|
/*!
|
||||||
@brief return number value for number tokens
|
@brief return number value for number tokens
|
||||||
|
@ -10899,125 +11042,69 @@ basic_json_parser_66:
|
||||||
number type (either integer, unsigned integer or floating point),
|
number type (either integer, unsigned integer or floating point),
|
||||||
which is passed back to the caller via the result parameter.
|
which is passed back to the caller via the result parameter.
|
||||||
|
|
||||||
This function parses the integer component up to the radix point or
|
integral numbers that don't fit into the the range of the respective
|
||||||
exponent while collecting information about the 'floating point
|
type are parsed as number_float_t
|
||||||
representation', which it stores in the result parameter. If there is
|
|
||||||
no radix point or exponent, and the number can fit into a @ref
|
|
||||||
number_integer_t or @ref number_unsigned_t then it sets the result
|
|
||||||
parameter accordingly.
|
|
||||||
|
|
||||||
If the number is a floating point number the number is then parsed
|
floating-point values do not satisfy std::isfinite predicate
|
||||||
using @a std:strtod (or @a std:strtof or @a std::strtold).
|
are converted to value_t::null
|
||||||
|
|
||||||
@param[out] result @ref basic_json object to receive the number, or
|
throws if the entire string [m_start .. m_cursor) cannot be
|
||||||
NAN if the conversion read past the current token. The latter case
|
interpreted as a number
|
||||||
needs to be treated by the caller function.
|
|
||||||
|
@param[out] result @ref basic_json object to receive the number.
|
||||||
*/
|
*/
|
||||||
void get_number(basic_json& result) const
|
void get_number(basic_json& result) const
|
||||||
{
|
{
|
||||||
assert(m_start != nullptr);
|
assert(m_start != nullptr);
|
||||||
|
assert(m_start < m_cursor);
|
||||||
|
|
||||||
const lexer::lexer_char_t* curptr = m_start;
|
strtonum num(reinterpret_cast<const char*>(m_start),
|
||||||
|
reinterpret_cast<const char*>(m_cursor));
|
||||||
|
|
||||||
// accumulate the integer conversion result (unsigned for now)
|
const bool is_negative = *m_start == '-';
|
||||||
number_unsigned_t value = 0;
|
|
||||||
|
|
||||||
// maximum absolute value of the relevant integer type
|
result.m_type = value_t::discarded;
|
||||||
number_unsigned_t max;
|
|
||||||
|
|
||||||
// temporarily store the type to avoid unecessary bitfield access
|
if (not num.is_integral())
|
||||||
value_t type;
|
|
||||||
|
|
||||||
// look for sign
|
|
||||||
if (*curptr == '-')
|
|
||||||
{
|
{
|
||||||
type = value_t::number_integer;
|
; // will parse as float below
|
||||||
max = static_cast<uint64_t>((std::numeric_limits<number_integer_t>::max)()) + 1;
|
|
||||||
curptr++;
|
|
||||||
}
|
}
|
||||||
else
|
else if (is_negative)
|
||||||
{
|
{
|
||||||
type = value_t::number_unsigned;
|
number_integer_t val{0};
|
||||||
max = static_cast<uint64_t>((std::numeric_limits<number_unsigned_t>::max)());
|
if (num.to(val))
|
||||||
}
|
|
||||||
|
|
||||||
// count the significant figures
|
|
||||||
for (; curptr < m_cursor; curptr++)
|
|
||||||
{
|
|
||||||
// quickly skip tests if a digit
|
|
||||||
if (*curptr < '0' or* curptr > '9')
|
|
||||||
{
|
{
|
||||||
if (*curptr == '.')
|
result.m_type = value_t::number_integer;
|
||||||
{
|
result.m_value = val;
|
||||||
// don't count '.' but change to float
|
|
||||||
type = value_t::number_float;
|
|
||||||
continue;
|
|
||||||
}
|
|
||||||
// assume exponent (if not then will fail parse): change to
|
|
||||||
// float, stop counting and record exponent details
|
|
||||||
type = value_t::number_float;
|
|
||||||
break;
|
|
||||||
}
|
|
||||||
|
|
||||||
// skip if definitely not an integer
|
|
||||||
if (type != value_t::number_float)
|
|
||||||
{
|
|
||||||
auto digit = static_cast<number_unsigned_t>(*curptr - '0');
|
|
||||||
|
|
||||||
// overflow if value * 10 + digit > max, move terms around
|
|
||||||
// to avoid overflow in intermediate values
|
|
||||||
if (value > (max - digit) / 10)
|
|
||||||
{
|
|
||||||
// overflow
|
|
||||||
type = value_t::number_float;
|
|
||||||
}
|
|
||||||
else
|
|
||||||
{
|
|
||||||
// no overflow
|
|
||||||
value = value * 10 + digit;
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// save the value (if not a float)
|
|
||||||
if (type == value_t::number_unsigned)
|
|
||||||
{
|
|
||||||
result.m_value.number_unsigned = value;
|
|
||||||
}
|
|
||||||
else if (type == value_t::number_integer)
|
|
||||||
{
|
|
||||||
// invariant: if we parsed a '-', the absolute value is between
|
|
||||||
// 0 (we allow -0) and max == -INT64_MIN
|
|
||||||
assert(value >= 0);
|
|
||||||
assert(value <= max);
|
|
||||||
|
|
||||||
if (value == max)
|
|
||||||
{
|
|
||||||
// we cannot simply negate value (== max == -INT64_MIN),
|
|
||||||
// see https://github.com/nlohmann/json/issues/389
|
|
||||||
result.m_value.number_integer = static_cast<number_integer_t>(INT64_MIN);
|
|
||||||
}
|
|
||||||
else
|
|
||||||
{
|
|
||||||
// all other values can be negated safely
|
|
||||||
result.m_value.number_integer = -static_cast<number_integer_t>(value);
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
// parse with strtod
|
number_unsigned_t val{0};
|
||||||
result.m_value.number_float = str_to_float_t(static_cast<number_float_t*>(nullptr), nullptr);
|
if (num.to(val))
|
||||||
|
|
||||||
// replace infinity and NAN by null
|
|
||||||
if (not std::isfinite(result.m_value.number_float))
|
|
||||||
{
|
{
|
||||||
type = value_t::null;
|
result.m_type = value_t::number_unsigned;
|
||||||
result.m_value = basic_json::json_value();
|
result.m_value = val;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// save the type
|
number_float_t val{0};
|
||||||
result.m_type = type;
|
if (result.m_type != value_t::discarded
|
||||||
|
or !num.to(val))
|
||||||
|
{
|
||||||
|
return; // already have a value from above
|
||||||
|
// or couldn't parse as float_t
|
||||||
|
}
|
||||||
|
|
||||||
|
result.m_type = value_t::number_float;
|
||||||
|
result.m_value = val;
|
||||||
|
|
||||||
|
// replace infinity and NAN by null
|
||||||
|
if (not std::isfinite(result.m_value.number_float))
|
||||||
|
{
|
||||||
|
result.m_type = value_t::null;
|
||||||
|
result.m_value = basic_json::json_value();
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
private:
|
private:
|
||||||
|
|
|
@ -9988,59 +9988,202 @@ class basic_json
|
||||||
return result;
|
return result;
|
||||||
}
|
}
|
||||||
|
|
||||||
/*!
|
|
||||||
@brief parse floating point number
|
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
|
||||||
standard floating point number parsing function based on the type
|
|
||||||
supplied via the first parameter. Set this to @a
|
|
||||||
static_cast<number_float_t*>(nullptr).
|
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
|
||||||
the number
|
|
||||||
|
|
||||||
@return the floating point number
|
|
||||||
*/
|
|
||||||
long double str_to_float_t(long double* /* type */, char** endptr) const
|
|
||||||
{
|
|
||||||
return std::strtold(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
|
||||||
}
|
|
||||||
|
|
||||||
/*!
|
/*!
|
||||||
@brief parse floating point number
|
@brief parse string into a built-in arithmetic type as if
|
||||||
|
the current locale is POSIX.
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
note: in floating-point case strtod may parse
|
||||||
standard floating point number parsing function based on the type
|
past the token's end - this is not an error.
|
||||||
supplied via the first parameter. Set this to @a
|
|
||||||
static_cast<number_float_t*>(nullptr).
|
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
any leading blanks are not handled.
|
||||||
the number
|
|
||||||
|
|
||||||
@return the floating point number
|
|
||||||
*/
|
*/
|
||||||
double str_to_float_t(double* /* type */, char** endptr) const
|
struct strtonum
|
||||||
{
|
{
|
||||||
return std::strtod(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
public:
|
||||||
}
|
strtonum(const char* start, const char* end)
|
||||||
|
: m_start(start), m_end(end)
|
||||||
|
{}
|
||||||
|
|
||||||
/*!
|
/// return true iff parsed successfully as
|
||||||
@brief parse floating point number
|
/// number of type T.
|
||||||
|
///
|
||||||
|
/// @val shall contain parsed value, or
|
||||||
|
/// undefined value if could not parse.
|
||||||
|
template<typename T,
|
||||||
|
typename = typename std::enable_if<
|
||||||
|
std::is_arithmetic<T>::value>::type >
|
||||||
|
bool to(T& val) const
|
||||||
|
{
|
||||||
|
return parse(val, std::is_integral<T>());
|
||||||
|
}
|
||||||
|
|
||||||
This function (and its overloads) serves to select the most approprate
|
/// return true iff token matches ^[+-]\d+$
|
||||||
standard floating point number parsing function based on the type
|
///
|
||||||
supplied via the first parameter. Set this to @a
|
/// this is a helper to determine whether to
|
||||||
static_cast<number_float_t*>(nullptr).
|
/// parse the token into floating-point or
|
||||||
|
/// integral type. We wouldn't need it if
|
||||||
|
/// we had separate token types for integral
|
||||||
|
/// and floating-point cases.
|
||||||
|
bool is_integral() const
|
||||||
|
{
|
||||||
|
const char* p = m_start;
|
||||||
|
|
||||||
@param[in,out] endptr recieves a pointer to the first character after
|
if (!p)
|
||||||
the number
|
{
|
||||||
|
return false; // LCOV_EXCL_LINE
|
||||||
|
}
|
||||||
|
|
||||||
@return the floating point number
|
if (*p == '-' or * p == '+')
|
||||||
*/
|
{
|
||||||
float str_to_float_t(float* /* type */, char** endptr) const
|
++p;
|
||||||
{
|
}
|
||||||
return std::strtof(reinterpret_cast<typename string_t::const_pointer>(m_start), endptr);
|
|
||||||
}
|
if (p == m_end)
|
||||||
|
{
|
||||||
|
return false; // LCOV_EXCL_LINE
|
||||||
|
}
|
||||||
|
|
||||||
|
while (p < m_end and* p >= '0'
|
||||||
|
and * p <= '9')
|
||||||
|
{
|
||||||
|
++p;
|
||||||
|
}
|
||||||
|
|
||||||
|
return p == m_end;
|
||||||
|
}
|
||||||
|
|
||||||
|
private:
|
||||||
|
const char* const m_start = nullptr;
|
||||||
|
const char* const m_end = nullptr;
|
||||||
|
|
||||||
|
// overloaded wrappers for strtod/strtof/strtold
|
||||||
|
// that will be called from parse<floating_point_t>
|
||||||
|
|
||||||
|
static void strtof(float& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtof(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
static void strtof(double& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtod(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
static void strtof(long double& f,
|
||||||
|
const char* str,
|
||||||
|
char** endptr)
|
||||||
|
{
|
||||||
|
f = std::strtold(str, endptr);
|
||||||
|
}
|
||||||
|
|
||||||
|
template<typename T>
|
||||||
|
bool parse(T& value, /*is_integral=*/std::false_type) const
|
||||||
|
{
|
||||||
|
// replace decimal separator with locale-specific
|
||||||
|
// version, when necessary; data will point to
|
||||||
|
// either the original string, or buf, or tempstr
|
||||||
|
// containing the fixed string.
|
||||||
|
std::string tempstr;
|
||||||
|
std::array<char, 64> buf;
|
||||||
|
const size_t len = static_cast<size_t>(m_end - m_start);
|
||||||
|
|
||||||
|
// Since dealing with strtod family of functions,
|
||||||
|
// we're getting the decimal point char from the
|
||||||
|
// C locale facilities instead of C++'s numpunct
|
||||||
|
// facet of the current std::locale;
|
||||||
|
const auto loc = localeconv();
|
||||||
|
assert(loc != nullptr);
|
||||||
|
const char decimal_point_char = !loc->decimal_point ? '.'
|
||||||
|
: loc->decimal_point[0];
|
||||||
|
|
||||||
|
const char* data = m_start;
|
||||||
|
|
||||||
|
if (decimal_point_char != '.')
|
||||||
|
{
|
||||||
|
const size_t ds_pos = static_cast<size_t>(
|
||||||
|
std::find(m_start, m_end, '.') - m_start );
|
||||||
|
|
||||||
|
if (ds_pos != len)
|
||||||
|
{
|
||||||
|
// copy the data into the local buffer or
|
||||||
|
// tempstr, if buffer is too small;
|
||||||
|
// replace decimal separator, and update
|
||||||
|
// data to point to the modified bytes
|
||||||
|
if (len + 1 < buf.size())
|
||||||
|
{
|
||||||
|
std::copy(m_start, m_end, buf.data());
|
||||||
|
buf[len] = 0;
|
||||||
|
buf[ds_pos] = decimal_point_char;
|
||||||
|
data = buf.data();
|
||||||
|
}
|
||||||
|
else
|
||||||
|
{
|
||||||
|
tempstr.assign(m_start, m_end);
|
||||||
|
tempstr[ds_pos] = decimal_point_char;
|
||||||
|
data = tempstr.c_str();
|
||||||
|
}
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
char* endptr = nullptr;
|
||||||
|
value = 0;
|
||||||
|
// this calls appropriate overload depending on T
|
||||||
|
strtof(value, data, &endptr);
|
||||||
|
|
||||||
|
// note that reading past the end is OK, the data may be,
|
||||||
|
// for example, "123.", where the parsed token only
|
||||||
|
// contains "123", but strtod will read the dot as well.
|
||||||
|
const bool ok = endptr >= data + len
|
||||||
|
and len > 0;
|
||||||
|
|
||||||
|
if (ok and value == 0.0 and * data == '-')
|
||||||
|
{
|
||||||
|
// some implementations forget to negate the zero
|
||||||
|
value = -0.0;
|
||||||
|
}
|
||||||
|
|
||||||
|
return ok;
|
||||||
|
}
|
||||||
|
|
||||||
|
signed long long parse_integral(
|
||||||
|
char** endptr,
|
||||||
|
/*is_signed*/std::true_type) const
|
||||||
|
{
|
||||||
|
return std::strtoll(m_start, endptr, 10);
|
||||||
|
}
|
||||||
|
|
||||||
|
unsigned long long parse_integral(
|
||||||
|
char** endptr,
|
||||||
|
/*is_signed*/std::false_type) const
|
||||||
|
{
|
||||||
|
return std::strtoull(m_start, endptr, 10);
|
||||||
|
}
|
||||||
|
|
||||||
|
template<typename T>
|
||||||
|
bool parse(T& value, /*is_integral=*/std::true_type) const
|
||||||
|
{
|
||||||
|
char* endptr = nullptr;
|
||||||
|
errno = 0; // these are thread-local
|
||||||
|
const auto x = parse_integral(&endptr, std::is_signed<T>());
|
||||||
|
|
||||||
|
static_assert(std::is_signed<T>() // called right overload?
|
||||||
|
== std::is_signed<decltype(x)>(), "");
|
||||||
|
|
||||||
|
value = static_cast<T>(x);
|
||||||
|
|
||||||
|
return x == static_cast<decltype(x)>(value) // x fits into destination T
|
||||||
|
and (x < 0) == (value < 0) // preserved sign
|
||||||
|
and (x != 0 or is_integral()) // strto[u]ll did nto fail
|
||||||
|
and errno == 0 // strto[u]ll did not overflow
|
||||||
|
and m_start < m_end // token was not empty
|
||||||
|
and endptr == m_end; // parsed entire token exactly
|
||||||
|
}
|
||||||
|
};
|
||||||
|
|
||||||
/*!
|
/*!
|
||||||
@brief return number value for number tokens
|
@brief return number value for number tokens
|
||||||
|
@ -10049,125 +10192,69 @@ class basic_json
|
||||||
number type (either integer, unsigned integer or floating point),
|
number type (either integer, unsigned integer or floating point),
|
||||||
which is passed back to the caller via the result parameter.
|
which is passed back to the caller via the result parameter.
|
||||||
|
|
||||||
This function parses the integer component up to the radix point or
|
integral numbers that don't fit into the the range of the respective
|
||||||
exponent while collecting information about the 'floating point
|
type are parsed as number_float_t
|
||||||
representation', which it stores in the result parameter. If there is
|
|
||||||
no radix point or exponent, and the number can fit into a @ref
|
|
||||||
number_integer_t or @ref number_unsigned_t then it sets the result
|
|
||||||
parameter accordingly.
|
|
||||||
|
|
||||||
If the number is a floating point number the number is then parsed
|
floating-point values do not satisfy std::isfinite predicate
|
||||||
using @a std:strtod (or @a std:strtof or @a std::strtold).
|
are converted to value_t::null
|
||||||
|
|
||||||
@param[out] result @ref basic_json object to receive the number, or
|
throws if the entire string [m_start .. m_cursor) cannot be
|
||||||
NAN if the conversion read past the current token. The latter case
|
interpreted as a number
|
||||||
needs to be treated by the caller function.
|
|
||||||
|
@param[out] result @ref basic_json object to receive the number.
|
||||||
*/
|
*/
|
||||||
void get_number(basic_json& result) const
|
void get_number(basic_json& result) const
|
||||||
{
|
{
|
||||||
assert(m_start != nullptr);
|
assert(m_start != nullptr);
|
||||||
|
assert(m_start < m_cursor);
|
||||||
|
|
||||||
const lexer::lexer_char_t* curptr = m_start;
|
strtonum num(reinterpret_cast<const char*>(m_start),
|
||||||
|
reinterpret_cast<const char*>(m_cursor));
|
||||||
|
|
||||||
// accumulate the integer conversion result (unsigned for now)
|
const bool is_negative = *m_start == '-';
|
||||||
number_unsigned_t value = 0;
|
|
||||||
|
|
||||||
// maximum absolute value of the relevant integer type
|
result.m_type = value_t::discarded;
|
||||||
number_unsigned_t max;
|
|
||||||
|
|
||||||
// temporarily store the type to avoid unecessary bitfield access
|
if (not num.is_integral())
|
||||||
value_t type;
|
|
||||||
|
|
||||||
// look for sign
|
|
||||||
if (*curptr == '-')
|
|
||||||
{
|
{
|
||||||
type = value_t::number_integer;
|
; // will parse as float below
|
||||||
max = static_cast<uint64_t>((std::numeric_limits<number_integer_t>::max)()) + 1;
|
|
||||||
curptr++;
|
|
||||||
}
|
}
|
||||||
else
|
else if (is_negative)
|
||||||
{
|
{
|
||||||
type = value_t::number_unsigned;
|
number_integer_t val{0};
|
||||||
max = static_cast<uint64_t>((std::numeric_limits<number_unsigned_t>::max)());
|
if (num.to(val))
|
||||||
}
|
|
||||||
|
|
||||||
// count the significant figures
|
|
||||||
for (; curptr < m_cursor; curptr++)
|
|
||||||
{
|
|
||||||
// quickly skip tests if a digit
|
|
||||||
if (*curptr < '0' or* curptr > '9')
|
|
||||||
{
|
{
|
||||||
if (*curptr == '.')
|
result.m_type = value_t::number_integer;
|
||||||
{
|
result.m_value = val;
|
||||||
// don't count '.' but change to float
|
|
||||||
type = value_t::number_float;
|
|
||||||
continue;
|
|
||||||
}
|
|
||||||
// assume exponent (if not then will fail parse): change to
|
|
||||||
// float, stop counting and record exponent details
|
|
||||||
type = value_t::number_float;
|
|
||||||
break;
|
|
||||||
}
|
|
||||||
|
|
||||||
// skip if definitely not an integer
|
|
||||||
if (type != value_t::number_float)
|
|
||||||
{
|
|
||||||
auto digit = static_cast<number_unsigned_t>(*curptr - '0');
|
|
||||||
|
|
||||||
// overflow if value * 10 + digit > max, move terms around
|
|
||||||
// to avoid overflow in intermediate values
|
|
||||||
if (value > (max - digit) / 10)
|
|
||||||
{
|
|
||||||
// overflow
|
|
||||||
type = value_t::number_float;
|
|
||||||
}
|
|
||||||
else
|
|
||||||
{
|
|
||||||
// no overflow
|
|
||||||
value = value * 10 + digit;
|
|
||||||
}
|
|
||||||
}
|
|
||||||
}
|
|
||||||
|
|
||||||
// save the value (if not a float)
|
|
||||||
if (type == value_t::number_unsigned)
|
|
||||||
{
|
|
||||||
result.m_value.number_unsigned = value;
|
|
||||||
}
|
|
||||||
else if (type == value_t::number_integer)
|
|
||||||
{
|
|
||||||
// invariant: if we parsed a '-', the absolute value is between
|
|
||||||
// 0 (we allow -0) and max == -INT64_MIN
|
|
||||||
assert(value >= 0);
|
|
||||||
assert(value <= max);
|
|
||||||
|
|
||||||
if (value == max)
|
|
||||||
{
|
|
||||||
// we cannot simply negate value (== max == -INT64_MIN),
|
|
||||||
// see https://github.com/nlohmann/json/issues/389
|
|
||||||
result.m_value.number_integer = static_cast<number_integer_t>(INT64_MIN);
|
|
||||||
}
|
|
||||||
else
|
|
||||||
{
|
|
||||||
// all other values can be negated safely
|
|
||||||
result.m_value.number_integer = -static_cast<number_integer_t>(value);
|
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
// parse with strtod
|
number_unsigned_t val{0};
|
||||||
result.m_value.number_float = str_to_float_t(static_cast<number_float_t*>(nullptr), nullptr);
|
if (num.to(val))
|
||||||
|
|
||||||
// replace infinity and NAN by null
|
|
||||||
if (not std::isfinite(result.m_value.number_float))
|
|
||||||
{
|
{
|
||||||
type = value_t::null;
|
result.m_type = value_t::number_unsigned;
|
||||||
result.m_value = basic_json::json_value();
|
result.m_value = val;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
// save the type
|
number_float_t val{0};
|
||||||
result.m_type = type;
|
if (result.m_type != value_t::discarded
|
||||||
|
or !num.to(val))
|
||||||
|
{
|
||||||
|
return; // already have a value from above
|
||||||
|
// or couldn't parse as float_t
|
||||||
|
}
|
||||||
|
|
||||||
|
result.m_type = value_t::number_float;
|
||||||
|
result.m_value = val;
|
||||||
|
|
||||||
|
// replace infinity and NAN by null
|
||||||
|
if (not std::isfinite(result.m_value.number_float))
|
||||||
|
{
|
||||||
|
result.m_type = value_t::null;
|
||||||
|
result.m_value = basic_json::json_value();
|
||||||
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
private:
|
private:
|
||||||
|
|
|
@ -383,7 +383,7 @@ TEST_CASE("regression tests")
|
||||||
};
|
};
|
||||||
|
|
||||||
// change locale to mess with decimal points
|
// change locale to mess with decimal points
|
||||||
std::locale::global(std::locale(std::locale(), new CommaDecimalSeparator));
|
auto orig_locale = std::locale::global(std::locale(std::locale(), new CommaDecimalSeparator));
|
||||||
|
|
||||||
CHECK(j1a.dump() == "23.42");
|
CHECK(j1a.dump() == "23.42");
|
||||||
CHECK(j1b.dump() == "23.42");
|
CHECK(j1b.dump() == "23.42");
|
||||||
|
@ -407,8 +407,34 @@ TEST_CASE("regression tests")
|
||||||
CHECK(j3c.dump() == "10000");
|
CHECK(j3c.dump() == "10000");
|
||||||
//CHECK(j3b.dump() == "1E04"); // roundtrip error
|
//CHECK(j3b.dump() == "1E04"); // roundtrip error
|
||||||
//CHECK(j3c.dump() == "1e04"); // roundtrip error
|
//CHECK(j3c.dump() == "1e04"); // roundtrip error
|
||||||
|
|
||||||
|
std::locale::global(orig_locale);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
SECTION("issue #379 - locale-independent str-to-num")
|
||||||
|
{
|
||||||
|
setlocale(LC_NUMERIC, "de_DE.UTF-8");
|
||||||
|
|
||||||
|
// disabled, because locale-specific beharivor is not
|
||||||
|
// triggered in AppVeyor for some reason
|
||||||
|
#if 0
|
||||||
|
{
|
||||||
|
// verify that strtod now uses commas as decimal-separator
|
||||||
|
CHECK(std::strtod("3,14", nullptr) == 3.14);
|
||||||
|
|
||||||
|
// verify that strtod does not understand dots as decimal separator
|
||||||
|
CHECK(std::strtod("3.14", nullptr) == 3);
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
|
||||||
|
// verify that parsed correctly despite using strtod internally
|
||||||
|
CHECK(json::parse("3.14").get<double>() == 3.14);
|
||||||
|
|
||||||
|
// check a different code path
|
||||||
|
CHECK(json::parse("1.000000000000000000000000000000000000000000000000000000000000000000000000").get<double>() == 1.0);
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
SECTION("issue #233 - Can't use basic_json::iterator as a base iterator for std::move_iterator")
|
SECTION("issue #233 - Can't use basic_json::iterator as a base iterator for std::move_iterator")
|
||||||
{
|
{
|
||||||
json source = {"a", "b", "c"};
|
json source = {"a", "b", "c"};
|
||||||
|
|
Loading…
Reference in a new issue