c++ double precision errors

[grahamwheat]

I'm making a program in which the collision detection determines the distance that an object will move and orders any other object which would be overlapped by that move to move enough to get out of the way. For example, the player sitting on a block moving up at a speed of 1.0 would be ordered to move 1.0 pixels up in order to avoid an overlap.
The program is far too long to submit in its entirety, but this is the focus of the matter:

#include <iostream>

using namespace std;

int main (int argc, char *argv[])
{
double var = 0;
double var2 = 10;
for (int i = 0; i < 100; i++)
{
var += 0.003;
var2 += 0.003;
}
if (var2-var == 10.0000000) cout << "equal" << endl;
cout << var2-var << endl;
cout << "Press ENTER to continue..." << endl;
cin.get();
return 0;
}

In line 14, I found that the difference is less than .00000000000001. var2-var equals 10 at non-decimal values.

Why is this happening? I am using Dev-C++ 4.9.9.2 and the default c++ compiler.

[vb_master]

I'm making a program in which the collision detection determines the distance that an object will move and orders any other object which would be overlapped by that move to move enough to get out of the way. For example, the player sitting on a block moving up at a speed of 1.0 would be ordered to move 1.0 pixels up in order to avoid an overlap.
The program is far too long to submit in its entirety, but this is the focus of the matter:

#include <iostream>

using namespace std;

int main (int argc, char *argv[])
{
double var = 0;
double var2 = 10;
for (int i = 0; i < 100; i++)
{
var += 0.003;
var2 += 0.003;
}
if (var2-var == 10.0000000) cout << "equal" << endl;
cout << var2-var << endl;
cout << "Press ENTER to continue..." << endl;
cin.get();
return 0;
}

In line 14, I found that the difference is less than .00000000000001. var2-var equals 10 at non-decimal values.

Why is this happening? I am using Dev-C++ 4.9.9.2 and the default c++ compiler.

Use long double instead of double.

[grahamwheat]

Thanks

I tried to implement it, but it doesn't seem to work after OpenGL has been enabled.

[Sharp Sapphire]

for (int i = 0; i < 100; i++)
{
var += 0.003;
var2 += 0.003;
}

Did you chop code out of that.. cause you could just do

var += 0.003*100;
var2+=0.003*100;

...

[APHawkes]

When it comes to floating-point values it's darn near impossible to get exact values. Finding if x=10.0000 is just as hard as trying to find when x=10.39740. Round numbers don't help, and are typically worse than weird decimal values.

You have a few options. First, you could use two integer/long variables to represent values similar to scientific notation. That will give precise numbers, but you'd have to override all the typical operators to make it smooth.
Second choice is to find a math library which already has types like this. It's already been done, I'm sure.

The third option is to allow for tolerances on your collisions. This is what most people do. If you want to know when x = 10.000000 you say:

Code: Select all

y= x - 10.0d;
if (y < 0.05 || y > -0.05) {
  /*your code here*/
}

thus giving a tolerance of +-0.05

[jdmlight]

This thread is two years old. Why was this bumped???