Understanding float variable comparison in if()

Question

Unable to find the reason for the following piece of code:

#include <stdio.h>
int main()
{
    float f = 0.1;
    if (f == 0.1)
      printf("True");
    else
      printf("False");
    return 0;
}

The output is false.

#include <stdio.h>
int main()
{
    float f = 0.1;
    if (f == (float)0.1)
      printf("True");
    else
      printf("False");
    return 0;
}

Now shows the correct output. Whats the reason behind this?

Also what is the reason of this behavior.

#include <stdio.h>
main()
{
    int n = 0, m = 0;
        if (n > 0)
            if (m > 0)
                    printf("True");
        else 
            printf("False");
}

Please never compare floating point numbers for equality. Floating point numbers almost never compare equal. — fuz, Aug 22 '13 at 10:01
..and C tag opens today's 'annoying question' competition with an old classic. (credit: [Martin James](http://stackoverflow.com/users/758133/martin-james)) — Lightness Races in Orbit, Aug 22 '13 at 10:13
@H2CO3 thank you for pointing that. I searched previously failed to get the same. I will take a look at the answer. — Megharaj, Aug 22 '13 at 10:13
@Megharaj You're welcome - you may also profit from reading [this](http://floating-point-gui.de) often-cited article. — , Aug 22 '13 at 10:14
@FUZxxl There are plenty of opportunities for using floating-point equality properly. Your advice is misplaced. Floating-point equality works. In this particular example it is the conversion from `double` to `float` that may produce results surprising to the naive programmer, not equality. — Pascal Cuoq, Aug 22 '13 at 12:12
@PascalCuoq Can you elaborate? I can not figure out a sensible case where it actually makes sense to compare floating point numbers for equality; except perhaps, in case you want to sort them, but even then you can compare with greater-then. — fuz, Aug 22 '13 at 12:49
@FUZxxl With the knowledge that `HUGE_VAL` is `+inf` for IEEE 754 systems, see this question: http://stackoverflow.com/questions/11421756/weverything-yielding-comparing-floating-point-with-or-is-unsafe . For another example, this post contains a function to convert from `double` to `unsigned long long` without using a conversion; the function uses floating-point equality: http://blog.frama-c.com/index.php?post/2013/05/01/A-conversionless-conversion-function2 — Pascal Cuoq, Aug 22 '13 at 12:56
@PascalCuoq This is a case for which you have the function isinf() on any C99 compliant system. It is not guaranteed that if something compares equal to HUGE_VAL on non IEEE 754 compliant systems it is really equal to positive infinity. — fuz, Aug 22 '13 at 13:10

score 6 · Accepted Answer · edited Jun 20 '20 at 09:12

6

0.1 literal is double. You loose precision here float f = 0.1;

You could say we loose precision during comparison again, so why isn't f == 0.1 true anyway? Because float extends to double, not the opposite. In C smaller type always extends to the bigger one.

Simplified your example we can say that double(float(1.0)) != 1.0

Possible solutions:

use double instead of float as a type of f.
use cast as in your 2nd example
use float literals - replace all 0.1 with 0.1f

Better solution

Floating point variables have a lot of problems with comparisons. They, including this one, can be solved by defining your own comparison function:

bool fp_equal(double a, double b, double eps = FLT_EPSILON) {
  return fabs(a - b) < fabs(eps * a);
}

The second part of the question:

Why the answer is false is because else part always corresponds to the innermost if block. So you were confused by formatting, the code is equivalent to:

#include <stdio.h>

int main()
{
    int n = 0, m = 0;
    if (n > 0) {
        if (m > 0) {
            printf("True");
        }
        else {
            printf("False");
        }
    }
}

edited Jun 20 '20 at 09:12

Community

1
1

answered Aug 22 '13 at 09:59

sasha.sochka

14,395
10
44
68

Another solution is to use a float literal `f`, `float f = 0.1f; if (f == 0.1f)` ... – David Ranieri Aug 22 '13 at 10:06
@AlterMann, it's written in my asnwer. Note the 3rd bullet point. – sasha.sochka Aug 22 '13 at 10:09
@sasha Please can you answer for my third example in the question(just now edited) – Megharaj Aug 22 '13 at 10:10
@Megharaj, for your third example, put brackets before each `if` and see what happens :) – David Ranieri Aug 22 '13 at 10:14
@Megharaj: that last part doesn't really belong under your original question. Don't do unrelated ".. while I have your attention .." questions. – Jongware Aug 22 '13 at 10:17
@sasha.sochka thanks a lot for the answer. Its bad on me that I dint notice that. – Megharaj Aug 22 '13 at 10:18
1

You probably should account for the magnitude of the floating-point values when doing such a comparison. http://www.c-faq.com/fp/fpequal.html – jamesdlin Aug 22 '13 at 10:25
Rather than use the magic number `1e-20`, consider `FLT_EPSILON` or some factor of it. – chux - Reinstate Monica Aug 22 '13 at 12:09
You fixed the magnitude problem, but it still won't work if `a` is negative. – jamesdlin Aug 22 '13 at 23:56

score 2 · Answer 2 · answered Aug 22 '13 at 10:19

Answer to your second question (third example):

According to your indention, the code doesn't do what you exepect it to:

#include <stdio.h>
main()
{ 
    int n = 0, m = 0;
        if (n > 0)
            if (m > 0)
                 printf("True");
        else 
            printf("False");
}

Here the else belongs to the inner if, so it's equal to

#include <stdio.h>
main()
{ 
    int n = 0, m = 0;
        if (n > 0) {
            if (m > 0) {
                 printf("True");
            } else {  
                 printf("False");
            }
        }
}

but I think you meant:

#include <stdio.h>
main()
{ 
    int n = 0, m = 0;
        if (n > 0) {
            if (m > 0) {
                 printf("True");
            }
        } else {  
            printf("False");
        }
}

It's a good example for why one should alwys user brackets in if statements.

And proper indentation. And a return type for `main`. – Lightness Races in Orbit Aug 22 '13 at 10:20 — Lightness Races in Orbit, Aug 22 '13 at 10:20
@Ligthness absolutely correct – Ingo Leonhardt Aug 22 '13 at 10:22 — Ingo Leonhardt, Aug 22 '13 at 10:22

Understanding float variable comparison in if()

2 Answers2

Linked