🚀 go-pugleaf

In article <1ghtanINNcf9@cs.tut.fi> pjh@cs.tut.fi (Haavisto Petri) writes:

>I don't have the standard, but it seems clear to me that given an
>expression of type double with a positive value, the following
>should always be true:
>
>		(long) (expression) == (long) floor(expression)

Thank you for all those who answered (both privately and in the net).
As I thought, there's no excuse for the compiler. The problem was not
merely academic since the bug makes the Julian Day Number functions
given in Numerical Recipes in C (and in many other places) subtly fail.

Petri Haavisto

PS. Since there was considerable interest I will post a short program
reproducing the bug:
---
/* Demonstrate a feature of Microsoft C v6.00 (MS-DOS) involving
   conversion from double to long.
   (Compiler switches: /W3 /AL /G2 /FPi87, no other switches tried). */

#include <stdio.h>
#include <math.h>

#define N1 2415080L
#define N2 1867216.25
#define N3 36524.25

long v1(long l) { return (long) ((l - N2) / N3);}
long v2(long l) { return (long) floor((l - N2) / N3);}

int main(void)
{
  printf(" Version 1: %ld\n Version 2: %ld\n", v1(N1), v2(N1));
  return 0;
}
---
This prints:

 Version 1: 14