www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.learn - float comparison

reply nocide <benutzer example.com> writes:
unittest {
     bool fuzzyCmp(float a, float b) {
         return abs(a-b) < 0.000001f;
     }
     float len1 = sqrt(8.0f);
     float len2 = sqrt(8.0f);
     assert(len1 == len2);              // passes
     assert(fuzzyCmp(len1,sqrt(8.0f))); // passes
     assert(len1 == sqrt(8.0f));        // fails!!
}


The comparison of the float variable passes, but the comparison to the 
return value of the sqrt-function fails.?!
Aug 25 2012
next sibling parent reply Jonathan M Davis <jmdavisProg gmx.com> writes:
On Saturday, August 25, 2012 10:28:44 nocide wrote:

 unittest {
      bool fuzzyCmp(float a, float b) {
          return abs(a-b) < 0.000001f;
      }
      float len1 = sqrt(8.0f);
      float len2 = sqrt(8.0f);
      assert(len1 == len2);              // passes
      assert(fuzzyCmp(len1,sqrt(8.0f))); // passes
      assert(len1 == sqrt(8.0f));        // fails!!
 }
 
 
 The comparison of the float variable passes, but the comparison to the
 return value of the sqrt-function fails.?!



Comparing floating point numbers with == is almost always the wrong thing to 
do.

http://floating-point-gui.de/

- Jonathan M Davis
Aug 25 2012
parent nocide <benutzer example.com> writes:
Am 25.08.2012 10:36, schrieb Jonathan M Davis:

 On Saturday, August 25, 2012 10:28:44 nocide wrote:

 unittest {
       bool fuzzyCmp(float a, float b) {
           return abs(a-b) < 0.000001f;
       }
       float len1 = sqrt(8.0f);
       float len2 = sqrt(8.0f);
       assert(len1 == len2);              // passes
       assert(fuzzyCmp(len1,sqrt(8.0f))); // passes
       assert(len1 == sqrt(8.0f));        // fails!!
 }


 The comparison of the float variable passes, but the comparison to the
 return value of the sqrt-function fails.?!



 Comparing floating point numbers with == is almost always the wrong thing to
 do.

 http://floating-point-gui.de/

 - Jonathan M Davis



Many thanks!

nocide
Aug 25 2012
prev sibling parent "bearophile" <bearophileHUGS lycos.com> writes:
Jonathan M Davis:


 Comparing floating point numbers with == is almost always the 
 wrong thing to do.



The second part of the answer is a suggestion to use 
std.math.feqrel.

Bye,
bearophile
Aug 25 2012