is_integer incorrect for infinity inputs

[andrjohns]

Description

The checking function is_integer is defined using floor:

template <typename T>
inline bool is_integer(T x) {
  return floor(x) == x;
}

However, when floor is passed infinity, it returns it unmodified.

• If arg is ±∞, it is returned, unmodified

So floor(double inf) returns double inf, which int turn means that is_integer(double inf) returns true.

Example

#include <stan/math/prim/scal/fun/is_integer.hpp>
#include <iostream>
#include <limits>

int main() {
  using stan::math::is_integer;

  double dbl_inf = std::numeric_limits<double>::infinity();

  std::cout << std::boolalpha
            << "is_integer(double inf): " << is_integer(dbl_inf) 
            << std::endl;
}

andrew@Lappy:~/math$ g++ -std=c++1y  -I . is_integer_test.cpp 
andrew@Lappy:~/math$ ./a.out 
is_integer(double inf): true

Expected Output

is_integer(double inf): false

Current Version:

v2.19.1

[bob-carpenter]

If we literally just want to test the integer type, we should be using traits.

I think this is meant to test if a double value is equal to an integer. For instance, this program runs without printing:

#include <iostream>
#include <cmath>
int main() {
  for (int i = -1000; i < 1000; ++i) {
    double x = i;
    if (x != std::floor(x))
      std::cout << "i = " << i << "; x = " << x << std::endl;
  }
}

I'd thought it might be a problem given that it seems to represent some numbers as more like 2.9999999... than like 3, so I thought floor might be a problem.

[andrjohns]

I think this is meant to test if a double value is equal to an integer

Yep, I'd say that's what the aim is. I think we just need to have the function check if the input is infinite beforehand, like:

template <typename T>
inline bool is_integer(T x) {
  if (is_inf(x))
    return 0;
  return floor(x) == x;
}

[bob-carpenter]

Yes, for sure do the right check for infinity. That can be a one-liner: return floor(x) == x && !is_inf(x);

…

On May 15, 2019, at 1:51 AM, Andrew Johnson ***@***.***> wrote: I think this is meant to test if a double value is equal to an integer Yep, I'd say that's what the aim is. I think we just need to have the function check if the input is infinite beforehand, like: template <typename T> inline bool is_integer(T x) { if (is_inf(x)) return 0; return floor(x) == x; } — You are receiving this because you commented. Reply to this email directly, view it on GitHub, or mute the thread.