Help me understand this simple C program.

[avi]

It all started with a question in Test Your C Skills with Yeshwant Kanetkar. Here is the actual question :

#include<stdio.h>
int main()
{
	float a=0.7;
	if(a<0.7)
	{
		printf("c\n");
	}
	else
	{
		printf("c++\n");
	}
	return 0;
}

I thought the output is 'C++', but the answer given as 'C'. So I ran the question in codepad, the answer is actually 'C'. C code - 14 lines - codepad

but when I changed the value of a to 1.7 or 2.7, I am getting output as 'C++'
code for 1.7 : C code - 14 lines - codepad
code for 2.7 : C code - 14 lines - codepad

When I checked the value stored, it was actually 0.69999 something, but not 0.7 !

Then I ran another program to print value of a float : http://codepad.org/er0zkzxS

#include<stdio.h>
int main()
{
float a=0.7, b=1.7, c=1.8, d=2.8;
printf("%.10f %.10f\n",0.7, a);
printf("%.10f %.10f\n",1.7, b);
printf("%.10f %.10f\n",1.8, c);
printf("%.10f %.10f\n",2.8, d);
return 0;
}

O/p:

0.7000000000 0.6999999881
1.7000000000 1.7000000477
1.8000000000 1.7999999523
2.8000000000 2.7999999523

So turns out float wasn't storing exact value. Instead it having value of 1.700000000000, it was having some extra digits at the end. However 1.8 was stored as 1.79999999 something.

It doesn't make sense to me at all. Why such behaviour ?

 

[tifosi]

If you look at the the way floating point numbers are represented in binary, you'll understand that they're all approximations. In fact, comparing floating point numbers is considered a bad programming practice. There are some tricks around it, but this sort of comparison is never recommended.

 

[avi]

can you suggest some links to read about this ? apart from wikipedia

 

[tifosi]

What Every Computer Scientist Should Know About Floating-Point Arithmetic

This paper is mathematical in nature. but it'll give you an idea.

You could also try stackoverflow for ways to get around such problems.

 

[booo]

http://www.planetpdf.com/codecuts/pdfs/aoa.pdf
chapter 14: page 789

 

[thangadurai.a]

In C and also in almost all C inherited families of the languages, when you want to use float values (raw values not float variables), always suffix it with the letter "f". In your case, the following if will work correctly if(a<0.7f)

 

[kvikram]

Try stack overflow. You will get super-informative replies in a few minutes!

 

[harryneopotter]

or u can use double. Floating point values have this precision problem since the Ice age !

 

[avi]

^LOL Yup, with double everything works perfectly.
C code - 10 lines - codepad
C code - 14 lines - codepad

@all - thank you very much for replying. Learnt something today, do not compare float values in C

 

[tifosi]

Double will work in the example, but there are several examples where double would fail too. The fact is, any kind of floating point representation is an approximation and can throw up bugs in the most unexpected places. Using an 'epsilon' style comparison is usually more reliable, and there are other practices too which you can find.

That said, double is good enough for general purpose computing. But if you're writing code for scientific computing or financial applications, you should be very careful with your floats and doubles.

 

[dinjo]

Never defined a datatype of float never helps.