Open Radar

 
Area:
Foundation

Summary:
When you print a NSDecimalNumber to the console using NSLog, the number is converted to a double first, potentially resulting in floating point rounding errors. This does not happen if you convert the NSDecimalNumber to a string by any other method first.

Steps to Reproduce:
NSDecimalNumber *d = [NSDecimalNumber decimalNumberWithString:@"99.9"];
NSLog(@"%@", d);     //prints "99.90000000000001"
NSLog(@"%@", d.description);     //prints "99.9"
NSLog(@"%@", d.stringValue);     //prints "99.9"
NSLog(@"%@", [d descriptionWithLocale:nil]);     //prints "99.9"

Expected Results:
All of the NSLog calls should print "99.9" for this number.

Actual Results:
If the NSDecimalNumber is passed directly to NSLog, it is converted to a double somewhere.

It seems this is happening during the mapping to os_log. Inside _os_log_fmt_flatten_data, it tests if the object is an instance of NSNumber and then directly calls -objCType, which for this decimal number returns "d" (double). Then -doubleValue is called on the decimal number, which creates the rounding error.

Version/Build:
Xcode 9.2 (also the same on Xcode 10.0)