Next: , Up: Library Functions

5.5.1 Portability of C Functions

Most usual functions can either be missing, or be buggy, or be limited on some architectures. This section tries to make an inventory of these portability issues. By definition, this list always requires additions. Please help us keeping it as complete as possible.

On ancient hosts, exit returned int. This is because exit predates void, and there was a long tradition of it returning int.

On current hosts, the problem more likely is that exit is not declared, due to C++ problems of some sort or another. For this reason we suggest that test programs not invoke exit, but return from main instead.

The C standard says a call free (NULL) does nothing, but some old systems don't support this (e.g., NextStep).
The C99 standard says that isinf and isnan are macros. On some systems just macros are available (e.g., HP-UX and Solaris 10), on some systems both macros and functions (e.g., glibc 2.3.2), and on some systems only functions (e.g., IRIX 6 and Solaris 9). In some cases these functions are declared in nonstandard headers like <sunmath.h> and defined in non-default libraries like -lm or -lsunmath.

The C99 isinf and isnan macros work correctly with long double arguments, but pre-C99 systems that use functions typically assume double arguments. On such a system, isinf incorrectly returns true for a finite long double argument that is outside the range of double.

The best workaround for these issues is to use gnulib modules isinf and isnan (see Gnulib). But a lighter weight solution involves code like the following.

          #include <math.h>
          #ifndef isnan
          # define isnan(x) \
              (sizeof (x) == sizeof (long double) ? isnan_ld (x) \
               : sizeof (x) == sizeof (double) ? isnan_d (x) \
               : isnan_f (x))
          static inline int isnan_f  (float       x) { return x != x; }
          static inline int isnan_d  (double      x) { return x != x; }
          static inline int isnan_ld (long double x) { return x != x; }
          #ifndef isinf
          # define isinf(x) \
              (sizeof (x) == sizeof (long double) ? isinf_ld (x) \
               : sizeof (x) == sizeof (double) ? isinf_d (x) \
               : isinf_f (x))
          static inline int isinf_f  (float       x)
          { return !isnan (x) && isnan (x - x); }
          static inline int isinf_d  (double      x)
          { return !isnan (x) && isnan (x - x); }
          static inline int isinf_ld (long double x)
          { return !isnan (x) && isnan (x - x); }

Use AC_C_INLINE (see C Compiler) so that this code works on compilers that lack the inline keyword. Some optimizing compilers mishandle these definitions, but systems with that bug typically have many other floating point corner-case compliance problems anyway, so it's probably not worth worrying about.

The C standard says a call malloc (0) is implementation dependent. It can return either NULL or a new non-null pointer. The latter is more common (e.g., the GNU C Library) but is by no means universal. AC_FUNC_MALLOC can be used to insist on non-NULL (see Particular Functions).
Posix prefers setenv to putenv; among other things, putenv is not required of all Posix implementations, but setenv is.

Posix specifies that putenv puts the given string directly in environ, but some systems make a copy of it instead (e.g., glibc 2.0, or BSD). And when a copy is made, unsetenv might not free it, causing a memory leak (e.g., FreeBSD 4).

On some systems putenv ("FOO") removes ‘FOO’ from the environment, but this is not standard usage and it dumps core on some systems (e.g., AIX).

On MinGW, a call putenv ("FOO=") removes ‘FOO’ from the environment, rather than inserting it with an empty value.

The C standard says a call realloc (NULL, size) is equivalent to malloc (size), but some old systems don't support this (e.g., NextStep).
signal handler
Normally signal takes a handler function with a return type of void, but some old systems required int instead. Any actual int value returned is not used; this is only a difference in the function prototype demanded.

All systems we know of in current use return void. The int was to support K&R C, where of course void is not available. The obsolete macro AC_TYPE_SIGNAL (see AC_TYPE_SIGNAL) can be used to establish the correct type in all cases.

In most cases, it is more robust to use sigaction when it is available, rather than signal.

The C99 standard says that if the output array isn't big enough and if no other errors occur, snprintf and vsnprintf truncate the output and return the number of bytes that ought to have been produced. Some older systems return the truncated length (e.g., GNU C Library 2.0.x or IRIX 6.5), some a negative value (e.g., earlier GNU C Library versions), and some the buffer length without truncation (e.g., 32-bit Solaris 7). Also, some buggy older systems ignore the length and overrun the buffer (e.g., 64-bit Solaris 7).
The C standard says sprintf and vsprintf return the number of bytes written. On some ancient systems (SunOS 4 for instance) they return the buffer pointer instead, but these no longer need to be worried about.
On various old systems, e.g., HP-UX 9, sscanf requires that its input string be writable (though it doesn't actually change it). This can be a problem when using gcc since it normally puts constant strings in read-only memory (see Incompatibilities of GCC). Apparently in some cases even having format strings read-only can be a problem.
Posix specifies that strerror_r returns an int, but many systems (e.g., GNU C Library version 2.2.4) provide a different version returning a char *. AC_FUNC_STRERROR_R can detect which is in use (see Particular Functions).
AIX 4.3 provides a broken version which produces the following results:
          strnlen ("foobar", 0) = 0
          strnlen ("foobar", 1) = 3
          strnlen ("foobar", 2) = 2
          strnlen ("foobar", 3) = 1
          strnlen ("foobar", 4) = 0
          strnlen ("foobar", 5) = 6
          strnlen ("foobar", 6) = 6
          strnlen ("foobar", 7) = 6
          strnlen ("foobar", 8) = 6
          strnlen ("foobar", 9) = 6

_SC_PAGESIZE is standard, but some older systems (e.g., HP-UX 9) have _SC_PAGE_SIZE instead. This can be tested with #ifdef.
The Posix spec says that unlink causes the given file to be removed only after there are no more open file handles for it. Some non-Posix hosts have trouble with this requirement, though, and some DOS variants even corrupt the file system.
On MinGW, unsetenv is not available, but a variable ‘FOO’ can be removed with a call putenv ("FOO="), as described under putenv above.
The C99 standard provides va_copy for copying va_list variables. It may be available in older environments too, though possibly as __va_copy (e.g., gcc in strict pre-C99 mode). These can be tested with #ifdef. A fallback to memcpy (&dst, &src, sizeof (va_list)) gives maximum portability.
va_list is not necessarily just a pointer. It can be a struct (e.g., gcc on Alpha), which means NULL is not portable. Or it can be an array (e.g., gcc in some PowerPC configurations), which means as a function parameter it can be effectively call-by-reference and library routines might modify the value back in the caller (e.g., vsnprintf in the GNU C Library 2.1).
Signed >>
Normally the C >> right shift of a signed type replicates the high bit, giving a so-called “arithmetic” shift. But care should be taken since Standard C doesn't require that behavior. On those few processors without a native arithmetic shift (for instance Cray vector systems) zero bits may be shifted in, the same as a shift of an unsigned type.
Integer /
C divides signed integers by truncating their quotient toward zero, yielding the same result as Fortran. However, before C99 the standard allowed C implementations to take the floor or ceiling of the quotient in some cases. Hardly any implementations took advantage of this freedom, though, and it's probably not worth worrying about this issue nowadays.