NAN macro expands to a quiet NaN (Not
A Number). Similarly both the
functions generate a quiet NaN value without raising an invalid exception.
The argument s should point to either an empty string
or a hexadecimal representation of a non-negative integer (e.g. 0x1234). In
the latter case, the integer is encoded in some free bits in the
representation of the NaN, which sometimes stores machine-specific
information about why a particular NaN was generated. There are 22 such bits
available for float variables, 51 bits for
double variables, and at least 51 bits for a
long double. If s is improperly
formatted or represents an integer that is too large, then the particular
encoding of the quiet NaN that is returned is indeterminate.
Calling these functions with a non-empty string isn't portable.
Another operating system may translate the string into a different NaN
encoding, and furthermore, the meaning of a given NaN encoding varies across
machine architectures. If you understood the innards of a particular
platform well enough to know what string to use, then you would have no need
for these functions anyway, so don't use them. Use the
NAN macro instead.
nanl() functions and the
NAN macro conform to ISO/IEC
9899:1999 (“ISO C99”).