Will this introduce errors if a/b=0.50000000000001?
How can I guarantee the conversion between float and int won't
introduce any errors?
Thanks,
Peng
Your worried about an insignificant, tiny, non-measureable error whilst
you are attempting to introduce a significant error by suggesting that
10/3 should be 4? Thats a 20% error! What happens when a 20% error gets
propagated?
What are the requirements for that 3 to become a 4? and why?
have you looked at modulus? 10%3, which will result 1 as an adjustment?
Although i still can't understand why.
Do you realize that you'ld have to multiply that doubles's error as
displayed by 1.0 x 10^14 in order for it to impact a cast to an integer
value? In fact, since an integer has both a max and min limitation with
a relatively puny and microscopic valid range, that floating number has
an amazing, mind-boggling count of significant figures.
Here, observe what happens to an integer when it reaches its max
allowed value:
#include <iostream>
#include <limits>
int main()
{
int min = std::numeric_limits<int>::min();
int max = std::numeric_limits<int>::max();
std::cout << "range of an integer on this platform:";
std::cout << "\nmin: " << min;
std::cout << "\nmax: " << max;
std::cout << "\nrollover = " << ++max; // rollover
std::cout << std::endl;
return 0;
}
/* your mileage will vary... i'm running a 64 bit platform
range of an integer on this platform:
min: -2147483648
max: 2147483647
rollover = -2147483648 !!!
*/
Now observe the unadulterated double:
/*
range of a double on this platform:
min: 2.22507e-308
max: 1.79769e+308
*/
thats around 4e616 !!! care to compare a roundoff error of 1.0e-14 to
that range?
floating numbers are not whole numbers and they do typically roundof
the last digit by 0.5 or so. But lets take that into context here.