FLT_MAX and FLT_MIN confusion...

registering · 03-16-2004, 08:35 AM

Hi all,
I'm trying to figure out the minimum and maximum values for a float my system can handle. However when I execute this code:

Code:

using namespace std;

#include <iostream>
#include <cfloat>   //calls /usr/lib/gcc-lib/i386-redhat-linux/3.2.2/include/float.h
#include <iomanip>

int main(int argc, char *argv[], char *envp[])
{
   float  max = FLT_MAX;  //should be 3.40282347e+38F
   float  min = FLT_MIN;  //should be 1.17549435e-38F (not negative??)
   
   cout << "\nFLT_MAX:          " << fixed << max; 
   cout << "\nFLT_MIN:          " << fixed << min;    
   
   max -= 125.0;
   cout << "\nMAX MINUS 125:    " << fixed << max;
   
   min += 125.0;
   cout << "\nMIN PLUS 125:     " << fixed << min; 
   
   max *= -1; 
   cout << "\nMAX NEGATIVE:    " << fixed << max;
   min *= -1;
   cout << "\nMIN NEGATIVE:    " << fixed << min;
   
   cout << endl;
   
   return (0);
   
} //end main(...)

I get this output:

FLT_MAX: 340282346638528859811704183484516925440.000000
FLT_MIN: 0.000000
MAX MINUS 125: 340282346638528859811704183484516925440.000000
MIN PLUS 125: 125.000000
MAX NEGATIVE: -340282346638528859811704183484516925440.000000
MIN NEGATIVE: -125.000000

Why is FLT_MIN zero? Why can't max be decreased? Thanks for any pointers.

roelofs · 10-18-2006, 03:05 PM

You've probably figured this out in the intervening two and a half years, but the problem with the minima is your use of the "fixed" formatting keyword. That limits the number of (fraction) decimal places to six, and obviously it's not going to show much when you need 37 zeros before you start printing "interesting" numbers.

Adding the following line at the top...

Code:

   cout << "\nFLT_MIN:          " << min;

...prints the following non-trivial value:

Code:

   FLT_MIN:          1.17549e-38

In addition, it appears that "fixed" is sticky, much like "hex" (IIRC); that is, it stays in effect until you cancel it. I can't find the URL to my favorite reference for this stuff (somewhere on the GNU/GCC site), but that's what I recall.