MATH_ERRNO, MATH_ERREXCEPT, math_errhandling

Description

The macro constant math_errhandling expands to an expression of type int that is either equal to MATH_ERRNO, or equal to MATH_ERREXCEPT, or equal to their bitwise OR (MATH_ERRNO | MATH_ERREXCEPT).

The value of math_errhandling indicates the type of error handling that is performed by the floating-point operators and functions:

Constant Explanation MATH_ERREXCEPT indicates that floating-point exceptions are used: at least FE_DIVBYZERO, FE_INVALID, and FE_OVERFLOW are defined in <cfenv>. MATH_ERRNO indicates that floating-point operations use the variable errno to report errors.

If the implementation supports IEEE floating-point arithmetic (IEC 60559), math_errhandling & MATH_ERREXCEPT is required to be non-zero.

The following floating-point error conditions are recognized:

Condition	Explanation	errno	floating-point exception	Example
Domain error	the argument is outside the range in which the operation is mathematically defined (the description of each function lists the required domain errors)	EDOM	FE_INVALID	std::acos(2)
Pole error	the mathematical result of the function is exactly infinite or undefined	ERANGE	FE_DIVBYZERO	std::log(0.0), 1.0/0.0
Range error due to overflow	the mathematical result is finite, but becomes infinite after rounding, or becomes the largest representable finite value after rounding down	ERANGE	FE_OVERFLOW	std::pow(DBL_MAX,2)
Range error due to underflow	the result is non-zero, but becomes zero after rounding, or becomes subnormal with a loss of precision	ERANGE or unchanged (implementation-defined)	FE_UNDERFLOW or nothing (implementation-defined)	DBL_TRUE_MIN/2
Inexact result	the result has to be rounded to fit in the destination type	unchanged	FE_INEXACT or nothing (unspecified)	std::sqrt(2), 1.0/10.0

Declarations

Since C++11

#define MATH_ERRNO        

#define MATH_ERREXCEPT    

#define math_errhandling  /*implementation defined*/

Notes

Whether FE_INEXACT is raised by the mathematical library functions is unspecified in general, but may be explicitly specified in the description of the function (e.g. std::rint vs std::nearbyint)

Before C++11, floating-point exceptions were not specified, EDOM was required for any domain error, ERANGE was required for overflows and implementation-defined for underflows.

Examples

#include <iostream>
#include <cfenv>
#include <cmath>
#include <cerrno>
#include <cstring>
#pragma STDC FENV_ACCESS ON
int main()
{
    std::cout 
        << "MATH_ERRNO is "
        << (math_errhandling & MATH_ERRNO ? "set" : "not set") << '\n'
        << "MATH_ERREXCEPT is "
        << (math_errhandling & MATH_ERREXCEPT ? "set" : "not set") << '\n';
    std::feclearexcept(FE_ALL_EXCEPT);
    errno = 0;
    std::cout 
        <<  "log(0) = " 
        << std::log(0) << '\n';
    if(errno == ERANGE)
            std::cout 
                << "errno = ERANGE (" 
                << std::strerror(errno) << ")\n";
    if(std::fetestexcept(FE_DIVBYZERO))
        std::cout 
            << "FE_DIVBYZERO (pole error) reported\n";
}

Possible Result

MATH_ERRNO is set
MATH_ERREXCEPT is set
log(0) = -inf
errno = ERANGE (Numerical result out of range)
FE_DIVBYZERO (pole error) reported