Question about floats in Java/in general

[Petar Stoyanov]

Let's say I have the following 3 situations:

a)

float a = 1.0e20f ;
float b = 1.0e-20f ;
System.out.println(a/b);

b)

float c = 1.0e8f ;
float d = 1.0f;
System.out.println(c+d);

c)

float e = 1.5e38f ;
float f = 0.4e38f ;
System.out.println(e-f);

Why is the output wrong by all 3 of them?
in a) the result should be 1, because 1e20/1e-20 is the same as 1/1 and its 1 but java outputs Infinity
in b) the expected result is 1.00000001e8 but java outputs 1.0ef. The same as the float c
in c) the expected result is 1.0000001e38, but it should be 1.1e38

[Fred Kleinschmidt]

 

Why is the output wrong by all 3 of them?
in a) the result should be 1, because 1e20/1e-20 is the same as 1/1 and its 1 but java outputs Infinity
in b) the expected result is 1.00000001e8 but java outputs 1.0ef. The same as the float c
in c) the expected result is 1.0000001e38, but it should be 1.1e38

a) 1e20/1e-20 = 1e40, not 1
   1e20 is not the same as 1, nor is 1e-20 the same as 1.
But the real problem is that floating point numbers have only about a maximum of 6 significant figures.
See  http://www.cygnus-software.com/papers/comparingfloats/comparingfloats.htm
Do not confuse range, accuracy, and significant figures.

Try the same code using double instead of float and see what, if anything, changes. Do you understand why?

[Paul Clapham]

Fred, the new version of the floating-point document you posted is here. Your link now points to a page which links to the original and also to what I posted, which the author thinks is better now.

[Campbell Ritchie]

Don't use floats unless some other code requires them.