Rounding behavior when NSNumberFormatter reads string?

Question

I have this sample code:

NSNumberFormatter *formatter = [[NSNumberFormatter alloc] init];
NSString *numberString = @"9.2";
NSNumber *number = [formatter numberFromString:numberString];
NSLog(@"String: %@, Number: %@", numberString, number);

The printed result is:

String: 9.2, Number: 9.199999999999999

I don't understand why this would happen. Am I missing a setting?

Welcome to the wonderful world of floating point numbers. The closest `double` to 9.2 is actually 9.199999999999999289457264239899814128875732421875. — Daniel Fischer, Apr 20 '12 at 22:14
It looks as if the formatter is converting it to a float in order to create the number.... Odd — lnafziger, Apr 20 '12 at 22:14
Hmm, strange that I haven't run into this before. Is there someplace I can read more about this? — Cory Imdieke, Apr 20 '12 at 22:19
@DanielFischer an NSObject? It doesn't specify a storage type but can store representations of many different types... — lnafziger, Apr 20 '12 at 22:22
Internally I believe it's still storing it as a double, so if there are accuracy issues they would show through there as well. — Cory Imdieke, Apr 20 '12 at 22:24
While I didn't realize that NSNumber stored as a double, here is an answer that I gave on another question that explains why you get what you have: http://stackoverflow.com/a/10049569/937822 — lnafziger, Apr 20 '12 at 22:26

score 2 · Accepted Answer · answered Apr 20 '12 at 22:27

2

According to this reference, an NSNumber can store any ordinary C numeric type. Since this is obtained by parsing a non-integer, the two types to choose from are float and double, and by the displayed value, the chosen representation is double.

Since decimal fractions like 9.2 aren't exactly representable as doubles, you get the closest representable number. That is displayed to default precision, which is about 16 decimal places.

answered Apr 20 '12 at 22:27

Daniel Fischer

181,706
17
308
431

Thanks for the additional info. I knew there were issues with very large or very small numbers, but didn't expect an issue like that for a seemingly-average sized number like that. Anyway thanks for pointing me in the right direction. – Cory Imdieke Apr 20 '12 at 22:30
The size doesn't matter, most numbers cannot be exactly represented in either decimal representation (`NSDecimalNumber`) or base 2 (what `float` and `double` use). However, the inexactness is usually very small and doesn't matter (if it does, though, things get hairy), so for 99.99% of applications at least one of `NSNumber` and `NSDecimalNumber` does fine. – Daniel Fischer Apr 20 '12 at 22:39

Rounding behavior when NSNumberFormatter reads string?

1 Answers1