🚀 go-pugleaf

> >      However, given that the conversion
> >      algorithm from external to internal format is the same, I would  
> argue that
> >      ALL instances of the SAME decimal value will result in the SAME  
> binary
> >      floating point value internally and can therefore be compared for  
> equality
> >      with impunity.
> 
> Actually, it's worse than this.  Sometimes floating point values can be  
> different depending on whether or not they've been in and out of  
> registers, etc.  Moral:  Use something like this:
> 
> #define EQU(_x,_y)  (((_x) > (_y)) ? (((_x)-(_y))<EPSILON) :  
> (((_y)-(_x))<EPSILON))
> #define EPSILON .00001  // or whatever

One can use the -ffloat-store option to "cure" this problem but at some
cost in terms of performance (depending, of course, on the application).

     -ffloat-store
          Don't store floating-point variables in registers.
          This prevents undesirable excess precision due to the
          floating registers keeping more precision than a double
          is supposed to have.

          For most programs, the excess precision does no harm,
          but a few programs rely on the precise definition of
          IEEE floating point.  Use -ffloat-store for such pro-
          grams.

mwagner@next.com