[CF-metadata] bounds

From: Russ Rew <russ>
Date: Thu, 22 May 2003 13:12:45 -0600

John Caron wrote:

> I would prefer "equal within tolerence". if you need a definition of
> tolerence, i would use something like
> abs( (a-b)/b) < 1.0E-5
>
> The reason is that post processors shouldnt have to preserve floating
> point to the last bitter bit. ;^}

I disagree for a couple of reasons:

 - Post processors don't have to preserve floating point to the last
   bitter bit, each one just has to be consistent in changing the same
   floating point value to a different floating point value. I can't
   imagine any post processor (or other program) that would sometimes
   change 0.66667 to 0.6666 and other times to 0.6667 (I'm just using
   base 10 for illustration). As long as the same processor munges
   the same floating point value in the same way, equality tests will
   work OK. Some programs depend on that fact for processing missing
   values for floating point.

 - Approximate comparison is more complex than

     abs( (a-b)/b) < 1.0E-5
   
   in case b is identically zero, and because there is no "epsilon"
   that is always appropriate for such tests. You can use an epsilon
   appropriate for IEEE single precision floating point, but then you
   can contrive cases where that would be wrong and exact equality
   would be the only appropriate test.
   
I think Jonathan is right to recommend exact equality tests in this
case.

--Russ
Received on Thu May 22 2003 - 13:12:45 BST