Andreas Schwab wrote: > Jim Meyering <[email protected]> writes: > >> The statement "long double x = NAN;" (inside glibc's strtold) leaves many >> bits of "x" uninitialized. > > You are looking at padding bits, which have unspecified contents.
I realize that they are unspecified. That is why I did not claim that this is a POSIX violation. POSIX says little about what strtold must do for a "NaN" string. However, I am dismayed that with glibc's strtold the values of those bits is not deterministic. That makes sorting NaN values obtained from strtold unnecessarily hard.
