Float math issue

Hi all,

Version 2017.3.0p4 (25a5860ad58d) Personal

I have some trouble with float variables, after some tests I made this really simple script to check the issue:

Code (CSharp):

1.  
2.     public float testmath;
3.  
4.     void Start()
5.     {
6.         print("TEST 1");
7.         testmath = 0f;
8.         print("testmath start = " + testmath);
9.         testmath = testmath + 0.6f;
10.         print("testmath add = " + testmath);
11.         testmath = testmath - 0.6f;
12.         print("testmath sub = " + testmath);
13.  
14.         print("TEST 2");
15.         testmath = 0f;
16.         print("testmath start = " + testmath);
17.         testmath = testmath + 0.2f;
18.         print("testmath add = " + testmath);
19.         testmath = testmath - 0.2f;
20.         print("testmath sub = " + testmath);
21.  
22.         print("TEST 3");
23.         testmath = 0f;
24.         print("testmath start = " + testmath);
25.         testmath = testmath + 0.2f + 0.2f;
26.         print("testmath add = " + testmath);
27.         testmath = testmath - 0.2f - 0.2f;
28.         print("testmath sub = " + testmath);
29.  
30.         print("TEST 4");
31.         testmath = 0f;
32.         print("testmath start = " + testmath);
33.         testmath = testmath + 0.2f + 0.2f + 0.2f;
34.         print("testmath add = " + testmath);
35.         testmath = testmath - 0.2f - 0.2f - 0.2f;
36.         print("testmath sub = " + testmath);
37.  
38. }
39.  

Results:
TEST 1
testmath start = 0
testmath add = 0.6
testmath sub = 0

TEST 2
testmath start = 0
testmath add = 0.2
testmath sub = 0

TEST 3
testmath start = 0
testmath add = 0.4
testmath sub = 0

TEST 4
testmath start = 0
testmath add = 0.6
testmath sub = 1.490116E-08

0.2 is not the magic number with this issue happen, I also found an issue with the following simple code

Code (CSharp):

1.  
2.         testmath = 0f;
3.         print("testmath start = " + testmath);
4.         testmath = testmath + 0.4f + 0.4f + 0.4f;
5.         print("testmath add = " + testmath);
6.         testmath = testmath - 0.4f;
7.         print("testmath sub = " + testmath);
8.  

Result:
testmath start = 0
testmath add = 1.2
testmath sub = 0.8000001

This is not just a print issue, I also made:

Code (CSharp):

1.  
2.         if (testmath != 0.8f)
3.             print("Real issue");
4.  

And unity print the 'Real issue' string.
So there is a real bug with float variables or I'm doing sometrhing wrong?

Thanks!

 

For any code that you post on the forum, please tag it correctly: https://forum.unity.com/threads/using-code-tags-properly.143875/

That is unfortunately the way floats work thanks to floating point arithmetic. That result is expected, you will have to do some thing differently. For comparing floats you can use Mathf.Approximately.

 

Tags added.

About the 'result is expected', are you sure about it?
I can understand if I add and subtract different variables with different precisions, but If I add and subtract two constant why I must get two different values?

Other than this, I temporary solved adding a first cast to decimal during the add\sub and then cast againt to float but this doesn't seem a valid solution to me.

Code (CSharp):

1.  
2.         testmath = 0f;
3.         print("testmath start = " + testmath);
4.         testmath = (float) ((decimal) testmath + 0.2m + 0.2m + 0.2m);
5.         print("testmath add = " + testmath);
6.         testmath = (float)((decimal) testmath - 0.2m - 0.2m - 0.2m);
7.         print("testmath sub = " + testmath);
8.  

Result:
testmath start = 0
testmath add = 0.6
testmath sub = 0

 

Yes I am sure about it, that's just the way computer works. You could even get the same result if you copy and paste (duplicate) a object on your hierarchy. You will see that the scale or the rotation will not be the same. Instead of being 1 is 0.99999999 or anything else. I recommend you to do a research on computers and floating calculations.

As for your temporary solution. That might work now, but don't be surprised if stops working.

What are you trying to accomplish?

 

"I recommend you to do a research on computers and floating calculations."
Thanks for the recommendation, I reccomend you to do a research on how not to be rude when you are helping someone.

I'm working on zoom camera, I start added and subtract a constat when user use the mouse wheel.
Checking the variable on Inspector I saw that after the third zoom out my variable was not I expected.

 

They weren't being rude, from your question and what you've said here you are missing the basics on how computers and particularly how binary works with floating point numbers. The forums aren't somewhere where we are going to teach you basic computer science, that is something you can do in your own time so the help you are receiving here makes more sense.

 

Don't bite the hand that feeds you.

 

ok, so thanks for your time, I'll go studing floating point math.

That was not my intention, but usually when I speak with someone that knows less than me, and ask help, I try to be more soft.
Anyway probably was just me that read in a wrong way the Rotary-Heart reply.

 

I know that trying to solve a bug can be frustrating and I apologize if you found what I wrote rude, but I meant it on an educational way. There is literally nothing we can do, that's how the computer handle binary with floating point numbers.

As for what you are doing, if you need to compare it use my suggestion.

 

A useful thing to remember is that floating point is stored as a combination of binary fractions (1/2,1/4,1/8,1/16) so these are the only values that can be stored 'exactly'. For example, you can store 0.125 correctly because it's 1/8 but not 0.2.The code you posted would actually work ok with 0.25 but it's generally better practice not to compare floating point without some kind of approximation.

The accuracy is also dependent on the range so you can store values like 0.12345 and 1234500000000 but you can't store 1234500000000.12345 unless you're using more than 24-bit precision.

 

https://docs.microsoft.com/en-us/dotnet/csharp/language-reference/keywords/float

Observe where it says "Precision: 7 digits"

If, over time, you start forgetting stuff about floats and how they work, at least make sure that's one thing you remember: "7 digits".

That means if you have a floating point value of "1234567", you won't be able to add 0.001 to it with any meaningful results.

Likewise, if you have a floating point value of "0.000001", you won't be able to add 123 to it in a meaningful way.

Bottom line: If you need accuracy, don't use floats.