forked from boostorg/conversion
Use metafunction (at compile-time), if possible; use run-time function only for broken libraries.
[SVN r35596]
This commit is contained in:
@@ -17,11 +17,8 @@
|
||||
|
||||
#ifdef BOOST_NO_LIMITS_COMPILE_TIME_CONSTANTS
|
||||
#include <boost/assert.hpp>
|
||||
#define BOOST_DETAIL_LCAST_ASSERT(cond) BOOST_ASSERT(cond)
|
||||
#else
|
||||
#include <boost/static_assert.hpp>
|
||||
#define BOOST_DETAIL_LCAST_ASSERT(cond) BOOST_STATIC_ASSERT(cond)
|
||||
// Remember, static_cast is evaluated even inside dead branches.
|
||||
#endif
|
||||
#if BOOST_WORKAROUND(BOOST_MSVC, < 1300)
|
||||
# include <boost/type_traits/is_same.hpp>
|
||||
@@ -30,29 +27,80 @@
|
||||
|
||||
namespace boost { namespace detail {
|
||||
|
||||
#ifdef _MSC_VER
|
||||
#pragma warning (push)
|
||||
// conditional expression is constant
|
||||
#pragma warning (disable : 4127)
|
||||
#ifndef BOOST_NO_LIMITS_COMPILE_TIME_CONSTANTS
|
||||
// Calculate an argument to pass to std::ios_base::precision from
|
||||
// lexical_cast. See alternative implementation for broken standard
|
||||
// libraries in lcast_get_precision below. Keep them in sync, please.
|
||||
template<class T>
|
||||
struct lcast_precision
|
||||
{
|
||||
typedef std::numeric_limits<T> limits;
|
||||
|
||||
BOOST_STATIC_CONSTANT(bool, use_default_precision =
|
||||
!limits::is_specialized || limits::is_exact
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(bool, is_specialized_bin =
|
||||
!use_default_precision &&
|
||||
limits::radix == 2 && limits::digits > 0
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(bool, is_specialized_dec =
|
||||
!use_default_precision &&
|
||||
limits::radix == 10 && limits::digits10 > 0
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(std::streamsize, streamsize_max =
|
||||
boost::integer_traits<std::streamsize>::const_max
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(unsigned int, precision_dec = limits::digits10 + 1U);
|
||||
|
||||
BOOST_STATIC_ASSERT(!is_specialized_dec ||
|
||||
precision_dec <= streamsize_max + 0UL
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(unsigned long, precision_bin =
|
||||
2UL + limits::digits * 30103UL / 100000UL
|
||||
);
|
||||
|
||||
BOOST_STATIC_ASSERT(!is_specialized_bin ||
|
||||
limits::digits + 0UL < ULONG_MAX / 30103UL &&
|
||||
precision_bin > limits::digits10 + 0UL &&
|
||||
precision_bin <= streamsize_max + 0UL
|
||||
);
|
||||
|
||||
BOOST_STATIC_CONSTANT(std::streamsize, value =
|
||||
is_specialized_bin ? precision_bin
|
||||
: is_specialized_dec ? precision_dec : 6
|
||||
);
|
||||
};
|
||||
#endif
|
||||
|
||||
template<class T>
|
||||
inline std::streamsize lcast_get_precision()
|
||||
{
|
||||
#if !defined(BOOST_NO_LIMITS_COMPILE_TIME_CONSTANTS)
|
||||
return lcast_precision<T>::value;
|
||||
#else // Follow lcast_precision algorithm at run-time:
|
||||
|
||||
typedef std::numeric_limits<T> limits;
|
||||
|
||||
bool const is_floating = limits::is_specialized && !limits::is_exact;
|
||||
if(is_floating)
|
||||
bool const use_default_precision =
|
||||
!limits::is_specialized || limits::is_exact;
|
||||
|
||||
if(!use_default_precision)
|
||||
{ // Includes all built-in floating-point types, float, double ...
|
||||
// and UDT types for which digits (significand bits) is defined (not zero)
|
||||
|
||||
bool const is_specialized_binary = is_floating &&
|
||||
bool const is_specialized_bin =
|
||||
limits::radix == 2 && limits::digits > 0;
|
||||
bool const is_specialized_decimal = is_floating &&
|
||||
bool const is_specialized_dec =
|
||||
limits::radix == 10 && limits::digits10 > 0;
|
||||
std::streamsize const precision_maxarg =
|
||||
boost::integer_traits<std::streamsize>::const_max;
|
||||
if(is_specialized_binary)
|
||||
std::streamsize const streamsize_max =
|
||||
(boost::integer_traits<std::streamsize>::max)();
|
||||
|
||||
if(is_specialized_bin)
|
||||
{ // Floating-point types with
|
||||
// limist::digits defined by the specialization.
|
||||
|
||||
@@ -60,20 +108,18 @@ inline std::streamsize lcast_get_precision()
|
||||
unsigned long const precision = 2UL + digits * 30103UL / 100000UL;
|
||||
// unsigned long is selected because it is at least 32-bits
|
||||
// and thus ULONG_MAX / 30103UL is big enough for all types.
|
||||
BOOST_DETAIL_LCAST_ASSERT(!is_specialized_binary ||
|
||||
BOOST_ASSERT(
|
||||
digits < ULONG_MAX / 30103UL &&
|
||||
precision > limits::digits10 + 0UL &&
|
||||
precision <= precision_maxarg + 0UL
|
||||
precision <= streamsize_max + 0UL
|
||||
);
|
||||
return precision;
|
||||
}
|
||||
else if(is_specialized_decimal)
|
||||
else if(is_specialized_dec)
|
||||
{ // Decimal Floating-point type, most likely a User Defined Type
|
||||
// rather than a real floating-point hardware type.
|
||||
unsigned int const precision = limits::digits10 + 1U;
|
||||
BOOST_DETAIL_LCAST_ASSERT(!is_specialized_decimal ||
|
||||
precision <= precision_maxarg + 0UL
|
||||
);
|
||||
BOOST_ASSERT(precision <= streamsize_max + 0UL);
|
||||
return precision;
|
||||
}
|
||||
}
|
||||
@@ -89,6 +135,7 @@ inline std::streamsize lcast_get_precision()
|
||||
// radix = 10 and digits10 == the number of decimal digits.
|
||||
|
||||
return 6;
|
||||
#endif
|
||||
}
|
||||
|
||||
template<class T>
|
||||
@@ -113,13 +160,7 @@ inline void lcast_set_precision(std::ios_base& stream
|
||||
stream.precision(s > t ? s : t);
|
||||
}
|
||||
|
||||
#ifdef _MSC_VER
|
||||
#pragma warning (pop)
|
||||
#endif
|
||||
|
||||
}}
|
||||
|
||||
#undef BOOST_DETAIL_LCAST_ASSERT
|
||||
|
||||
#endif // BOOST_DETAIL_LCAST_PRECISION_HPP_INCLUDED
|
||||
|
||||
|
Reference in New Issue
Block a user