use struct instead

This commit is contained in:
Janek Kozicki
2021-07-19 00:56:41 +02:00
parent d01e7e9b77
commit 98bef0c0ca
4 changed files with 32 additions and 20 deletions

View File

@@ -35,15 +35,18 @@ constexpr T max(T a, T b)
}
// TODO: this looks like a duplicate from test_cpp_double_float_comparision.cpp file.
template<typename FloatingPointType> constexpr bool is_floating_point = std::is_floating_point<FloatingPointType>::value
// FIXME: this looks like a duplicate from test_cpp_double_float_comparision.cpp file.
template<typename FloatingPointType> struct is_floating_point {
static const bool value;
};
template<typename FloatingPointType> const bool is_floating_point<FloatingPointType>::value = std::is_floating_point<FloatingPointType>::value
#ifdef BOOST_MATH_USE_FLOAT128
or std::is_same<FloatingPointType,boost::multiprecision::float128>::value
#endif
;
template <typename FloatingPointType,
typename std::enable_if<is_floating_point<FloatingPointType>, bool>::type = true>
typename std::enable_if<is_floating_point<FloatingPointType>::value, bool>::type = true>
FloatingPointType uniform_real()
{
static std::random_device rd;
@@ -67,14 +70,14 @@ NumericType uniform_integral_number()
template <typename NumericType,
typename std::enable_if<std::is_integral<NumericType>::value && !is_floating_point<NumericType>, bool>::type = true>
typename std::enable_if<std::is_integral<NumericType>::value && !is_floating_point<NumericType>::value, bool>::type = true>
NumericType get_rand()
{
return uniform_integral_number<NumericType>();
}
template <typename FloatingPointType,
typename std::enable_if<is_floating_point<FloatingPointType>, bool>::type = true>
typename std::enable_if<is_floating_point<FloatingPointType>::value, bool>::type = true>
FloatingPointType get_rand()
{
return uniform_real<FloatingPointType>();