Converting String to Float or double value changes the decimal values

Question

I am trying to convert string to float but it changes the decimal values. e.g I have a string:

NSString *str = @"33.641832";
float value = [str floatValue]; //33.6418304

the value give me 33.6418304 instead of 33.641832. I want to convert exactly as in the string.

Florian Burel · Answer 1 · 2016-08-15T13:32:05.620

1

This is due to the lack of float precision.

By my experience, Float are precise to the 7 significant digit approximatly.

33.641832 is stored as 33641832 *10^-6 so you have 8 significant digit (33641832).

You should use double if you want more precision.

edited Aug 15 '16 at 13:32

answered Aug 15 '16 at 13:17

Florian Burel

The number is actually stored in binary, not decimal. – Simon Byrne Aug 15 '16 at 13:34
@SimonByrne Huh? Are you serious? – Droppy Aug 15 '16 at 13:36
@SimonByrne yes obviously, but it is store on 4 bytes: 1 bit for the sign, 8 for the exp value (base), and 22 bits for the significant digit. More info : https://en.wikipedia.org/wiki/Single-precision_floating-point_format. – Florian Burel Aug 15 '16 at 13:37
Yes, see https://en.wikipedia.org/wiki/Single-precision_floating-point_format – Simon Byrne Aug 15 '16 at 13:38
It is actually stored as `1.051307201385498046875 * 2^5` – Simon Byrne Aug 15 '16 at 13:40
well it is stored as 0 and 1, we can all agree to that. we are all grown up dev here. so obviously, int float double, string, object, pointers are stored as binary. what I mean is that in order to "represent" the idea of decimal (you cannot cut a bit) the pattern used is to mutliply the decimal by a power of ten until you obtain the smallest integer (called the fraction). the fraction is stored as a binary in a part of the 4 bytes and the power of ten (called the base) is stored in another – Florian Burel Aug 15 '16 at 13:43
My point was that the exponent is stored as a power of 2, not 10. – Simon Byrne Aug 15 '16 at 19:49

score 0 · Answer 2 · answered Aug 15 '16 at 13:34

0

See the table in this post by Rick Regan:

A float type is only guaranteed to accurately "round-trip" numbers with 6 significant digits. For doubles, the guarantee is 15.

answered Aug 15 '16 at 13:34

Simon Byrne

2 Answers2