std::numeric_limits<T>::signaling_NaN

From cppreference.com
 
 
Utilities library
General utilities
Date and time
Function objects
Formatting library (C++20)
(C++11)
Relational operators (deprecated in C++20)
Integer comparison functions
(C++20)(C++20)(C++20)   
(C++20)
Swap and type operations
(C++14)
(C++11)
(C++11)
(C++11)
(C++17)
Common vocabulary types
(C++11)
(C++17)
(C++17)
(C++17)
(C++11)
(C++17)
(C++23)
Elementary string conversions
(C++17)
(C++17)
 
 
 
static T signaling_NaN() throw();
(until C++11)
static constexpr T signaling_NaN() noexcept;
(since C++11)

Returns the special value "signaling not-a-number", as represented by the floating-point type T. Only meaningful if std::numeric_limits<T>::has_signaling_NaN == true. In IEEE 754, the most common binary representation of floating-point numbers, any value with all bits of the exponent set and at least one bit of the fraction set represents a NaN. It is implementation-defined which values of the fraction represent quiet or signaling NaNs, and whether the sign bit is meaningful.

Return value

T std::numeric_limits<T>::signaling_NaN()
/* non-specialized */ T()
bool false
char 0
signed char 0
unsigned char 0
wchar_t 0
char8_t (since C++20) 0
char16_t (since C++11) 0
char32_t (since C++11) 0
short 0
unsigned short 0
int 0
unsigned int 0
long 0
unsigned long 0
long long (since C++11) 0
unsigned long long (since C++11) 0
float implementation-defined (may be FLT_SNAN)
double implementation-defined (may be DBL_SNAN)
long double implementation-defined (may be LDBL_SNAN)

Notes

A NaN never compares equal to itself. Copying a NaN is not required, by IEEE-754, to preserve its bit representation (sign and payload), though most implementation do.

When a signaling NaN is used as an argument to an arithmetic expression, the appropriate floating-point exception may be raised and the NaN is "quieted", that is, the expression returns a quiet NaN.

Example

Demonstrates the use of a signaling NaN to raise a floating-point exception:

#include <cfenv>
#include <iostream>
#include <limits>
 
#pragma STDC_FENV_ACCESS on
 
void show_fe_exceptions()
{
    int n = std::fetestexcept(FE_ALL_EXCEPT);
 
    if (n & FE_INVALID)
        std::cout << "FE_INVALID is raised\n";
    else if (n == 0)
        std::cout << "no exceptions are raised\n";
 
    std::feclearexcept(FE_ALL_EXCEPT);
}
 
int main()
{
    double snan = std::numeric_limits<double>::signaling_NaN();
    std::cout << "After sNaN was obtained, ";
    show_fe_exceptions();
 
    double qnan = snan * 2.0;
    std::cout << "After sNaN was multiplied by 2, ";
    show_fe_exceptions();
 
    double qnan2 = qnan * 2.0;
    std::cout << "After the quieted NaN was multiplied by 2, ";
    show_fe_exceptions();
 
    std::cout << "The result is " << qnan2 << '\n';
}

Output:

After sNaN was obtained, no exceptions are raised
After sNaN was multiplied by 2, FE_INVALID is raised
After the quieted NaN was multiplied by 2, no exceptions are raised
The result is nan

See also

identifies floating-point types that can represent the special value "signaling not-a-number" (NaN)
(public static member constant)
[static]
returns a quiet NaN value of the given floating-point type
(public static member function)
(C++11)
checks if the given number is NaN
(function)