Precision of float and double

[John Stark]

One question in Enthuware is about precision of floats and doubles. It turnes out that

float f = 1.0f / 3.0f;
System.out.println((f * 3.0f) == 1.0f);

prints 'true'. When I change it to

float f = 1.0f / 3.0f;
System.out.println((f * 3.0f) == 1.0);// 1.0 is double now

it still prints 'true'.
But

float f = 1.0f / 3.0f;
System.out.println((f * 3.0) == 1.0);// 3.0 is also double now

prints 'false'. I don't really understand this behavior.

Another question is: Obviously in Java it is OK to compare floats or doubles directly:

double d = 1.0;
if (d == 1.0) System.out.println("equal");

In C you don't do that because 1.0 might not be exactly 1.0. You do something like:

if (fabs(d-1.0) < 0.00000001)

So you check whether the difference between the number is sufficiently small.

Why can I compare doubles directly in Java? Can anybody explain?

Thanks,

John

[fred rosenberger]

I'm not sure about the details about why it works in some of your cases and not others, but i wouldn't ever count on it working.

Most high-level languages follow the IEEE standard on floats. I've never read it, but it basically defines how they should work.

I would not bet the farm on your examples still evaluating true/false the same with other computations, so I think your idea of comparing within a certain delta is probably the right way to go in Java, too.

[John Stark]

That's what I was basically thinking.

Thanks,

John