CODEWITHSUNDEEP
c++c++11
user8147810
user8147810

Reputation:

How to handle -nan output in c++

Let's say I have a function that calculates Euclidean distance between a pair of points. point and point_pair are two structs defined as:

struct point {
    int x, y;
}

and

struct point_pair {
    point a, b;
}

The following functions calculates the distance taking a pair of points as an input:

double calc_distance(point_pair pair)
{
  return (sqrt((pair.a.x - pair.b.x) * (pair.a.x - pair.b.x) + (pair.a.y - pair.b.y) * (pair.a.y - pair.b.y)));
}

The function works fine for small point pair values; but for a point pair such as: 

651760491 595516649
716636914 955747792

The output is -nan

I'm not sure how to workaround this, is there something else I should be using in place of double?

Here is the entire code: https://pastebin.com/5XEr9bTD

Upvotes: 1

Views: 161

Answers (2)

bipll
bipll

Reputation: 11940

The best would be to not use int for point coordinates at all:

#include <iostream>
#include <cmath>

struct point { double x, y; };

struct point_pair { point a, b; };

double calc_distance(point_pair pair) {
    return (std::sqrt((pair.a.x - pair.b.x) * (pair.a.x - pair.b.x) + (pair.a.y - pair.b.y) * (pair.a.y - pair.b.y)));
}

int main() {
    point_pair pp{{651760491, 595516649}
            , {716636914, 955747792}};
    std::cout << calc_distance(pp) << '\n';
}

But if you have a reason to do that, well, then at least convert coordinate differences to double:

struct point { int x, y; };

struct point_pair { point a, b; };

double calc_distance(point_pair pair) {
    return (std::sqrt(double(pair.a.x - pair.b.x) * (pair.a.x - pair.b.x) + double(pair.a.y - pair.b.y) * (pair.a.y - pair.b.y)));
}

Because otherwise you face integer overflow at large orders, and signed overflow is an UB in C++.

Upvotes: 2

Severin Pappadeux
Severin Pappadeux

Reputation: 20080

Right way to do it would be use of standard hypot() function.

https://en.cppreference.com/w/cpp/numeric/math/hypot

"Computes the square root of the sum of the squares of x and y, without undue overflow or underflow at intermediate stages of the computation."

Upvotes: 5

Related Questions