1MATH_ERROR(7) Linux Programmer's Manual MATH_ERROR(7)
2
6 math_error - detecting errors from mathematical functions
7
9 #include <math.h>
10 #include <errno.h>
11 #include <fenv.h>
12
14 When an error occurs, most library functions indicate this fact by re‐
15 turning a special value (e.g., -1 or NULL). Because they typically re‐
16 turn a floating-point number, the mathematical functions declared in
17 <math.h> indicate an error using other mechanisms. There are two er‐
18 ror-reporting mechanisms: the older one sets errno; the newer one uses
19 the floating-point exception mechanism (the use of feclearexcept(3) and
20 fetestexcept(3), as outlined below) described in fenv(3).
21 A portable program that needs to check for an error from a mathematical
23 function should set errno to zero, and make the following call
24 feclearexcept(FE_ALL_EXCEPT);
26 before calling a mathematical function.
28 Upon return from the mathematical function, if errno is nonzero, or the
30 following call (see fenv(3)) returns nonzero
31 fetestexcept(FE_INVALID | FE_DIVBYZERO | FE_OVERFLOW |
33 FE_UNDERFLOW);
34 then an error occurred in the mathematical function.
36 The error conditions that can occur for mathematical functions are de‐
38 scribed below.
39 Domain error
41 A domain error occurs when a mathematical function is supplied with an
42 argument whose value falls outside the domain for which the function is
43 defined (e.g., giving a negative argument to log(3)). When a domain
44 error occurs, math functions commonly return a NaN (though some func‐
45 tions return a different value in this case); errno is set to EDOM, and
46 an "invalid" (FE_INVALID) floating-point exception is raised.
47 Pole error
49 A pole error occurs when the mathematical result of a function is an
50 exact infinity (e.g., the logarithm of 0 is negative infinity). When a
51 pole error occurs, the function returns the (signed) value HUGE_VAL,
52 HUGE_VALF, or HUGE_VALL, depending on whether the function result type
53 is double, float, or long double. The sign of the result is that which
54 is mathematically correct for the function. errno is set to ERANGE,
55 and a "divide-by-zero" (FE_DIVBYZERO) floating-point exception is
56 raised.
57 Range error
59 A range error occurs when the magnitude of the function result means
60 that it cannot be represented in the result type of the function. The
61 return value of the function depends on whether the range error was an
62 overflow or an underflow.
63 A floating result overflows if the result is finite, but is too large
65 to represented in the result type. When an overflow occurs, the func‐
66 tion returns the value HUGE_VAL, HUGE_VALF, or HUGE_VALL, depending on
67 whether the function result type is double, float, or long double. er‐
68 rno is set to ERANGE, and an "overflow" (FE_OVERFLOW) floating-point
69 exception is raised.
70 A floating result underflows if the result is too small to be repre‐
72 sented in the result type. If an underflow occurs, a mathematical
73 function typically returns 0.0 (C99 says a function shall return "an
74 implementation-defined value whose magnitude is no greater than the
75 smallest normalized positive number in the specified type"). errno may
76 be set to ERANGE, and an "underflow" (FE_UNDERFLOW) floating-point ex‐
77 ception may be raised.
78 Some functions deliver a range error if the supplied argument value, or
80 the correct function result, would be subnormal. A subnormal value is
81 one that is nonzero, but with a magnitude that is so small that it
82 can't be presented in normalized form (i.e., with a 1 in the most sig‐
83 nificant bit of the significand). The representation of a subnormal
84 number will contain one or more leading zeros in the significand.
85
87 The math_errhandling identifier specified by C99 and POSIX.1 is not
88 supported by glibc. This identifier is supposed to indicate which of
89 the two error-notification mechanisms (errno, exceptions retrievable
90 via fetestexcept(3)) is in use. The standards require that at least
91 one be in use, but permit both to be available. The current (version
92 2.8) situation under glibc is messy. Most (but not all) functions
93 raise exceptions on errors. Some also set errno. A few functions set
94 errno, but don't raise an exception. A very few functions do neither.
95 See the individual manual pages for details.
96 To avoid the complexities of using errno and fetestexcept(3) for error
98 checking, it is often advised that one should instead check for bad ar‐
99 gument values before each call. For example, the following code en‐
100 sures that log(3)'s argument is not a NaN and is not zero (a pole er‐
101 ror) or less than zero (a domain error):
102 double x, r;
104 if (isnan(x) || islessequal(x, 0)) {
106 /* Deal with NaN / pole error / domain error */
107 }
108 r = log(x);
110 The discussion on this page does not apply to the complex mathematical
112 functions (i.e., those declared by <complex.h>), which in general are
113 not required to return errors by C99 and POSIX.1.
114 The gcc(1) -fno-math-errno option causes the executable to employ im‐
116 plementations of some mathematical functions that are faster than the
117 standard implementations, but do not set errno on error. (The gcc(1)
118 -ffast-math option also enables -fno-math-errno.) An error can still
119 be tested for using fetestexcept(3).
120
122 gcc(1), errno(3), fenv(3), fpclassify(3), INFINITY(3), isgreater(3),
123 matherr(3), nan(3)
124 info libc
126
128 This page is part of release 5.13 of the Linux man-pages project. A
129 description of the project, information about reporting bugs, and the
130 latest version of this page, can be found at
131 https://www.kernel.org/doc/man-pages/.
132Linux 2021-03-22 MATH_ERROR(7)