Test case:
NSLog(@\"%f\", M_PI);
NSLog(@\"%@\", [NSString stringWithFormat:@\"%f\", M_PI]);
NSLog(@\"%@\", [NSNumber numberWithDouble:M_PI]);
You should use long double with a maximum formatting of 20 digits @.20Lg. The long doubles are 80-bit floating points, so you will not get a better precision than that. Be aware also that as of XCode 4.3.2 the constants are not in long double notation, even if the many digits suggest an uberlong double ;-)
NSLog(@"%.21g", M_PI);
// with cast because M_PI is not defined as long double
NSLog(@"%.21Lg", (long double)M_PI);
// with corrected long double representation (#.####L):
// v from here on overhead
NSLog(@"%.21Lg", 3.14159265358979323846264338327950288L);
// alternative for creating PI
NSLog(@"%.21Lg", asinl(1.0)*2.0);
// and a funny test case:
NSLog(@"%.21Lg", asinl(1.0)*2.0 - M_PI); // on second thought, not that funny: should be 0.0
the results are:
p[5528:f803] 3.141592653589793116 (actually 16 digits standard double precision) p[5528:f803] 3.141592653589793116 p[5528:f803] 3.14159265358979323851 p[5528:f803] 3.14159265358979323851 p[5575:f803] 1.22514845490862001043e-16 (should have been 0.0)