From: sgk@REMOVEtroutmask.apl.washington.edu   
      
   On Tue, 19 Aug 2025 12:15:37 +0200, Kay Diederichs wrote:   
      
   > Am 8/17/25 um 05:18 schrieb Steven G. Kargl:   
   >> On Sat, 16 Aug 2025 20:40:13 -0500, Gary Scott wrote:   
   >>   
   >>> On 8/16/2025 7:02 PM, Steven G. Kargl wrote:   
   >>>> On Fri, 15 Aug 2025 22:24:52 +0000, Lawrence D’Oliveiro wrote:   
   >>>>   
   >>>>> On Thu, 14 Aug 2025 11:21:00 +0800, Woozy Song wrote:   
   >>>>>   
   >>>>>> A program was going wrong, and found   
   >>>>>> FRACTION(3.75)=0.9375 and FRACTION(2.5)=0.9375   
   >>>>>   
   >>>>> Broken as designed, according to the spec.   
   >>>>>   
   >>>>> Why on Earth does FRACTION have this meaning?   
   >>>>   
   >>>> To what spec are you referring?   
   >>>>   
   >>>> gfortran is giving the correct answer (although   
   >>>> I suspect OP has a transcription problem with   
   >>>> the second example).   
   >>>>   
   >>>   
   >>> I think most casual, non-math, non-floating point experts would expect   
   this:   
   >>   
   >> For someone programming in the Fortran language, I would expect   
   >> that they would consult some form of documentation to learn why   
   >> their expectation might be wrong.  For example, the gfortran   
   >> documentation (which comes with the compiler) contains   
   >>   
   >>   
   >> 8.119 ‘FRACTION’ -- Fractional part of the model representation   
   >> ===============================================================   
   >>   
   >> _Synopsis_:   
   >>       ‘Y = FRACTION(X)’   
   >>   
   >> _Description_:   
   >>       ‘FRACTION(X)’ returns the fractional part of the model   
   >>       representation of ‘X’.   
   >>   
   >> _Class_:   
   >>       Elemental function   
   >>   
   >> _Arguments_:   
   >>       X           The type of the argument shall be a ‘REAL’.   
   >>   
   >> _Return value_:   
   >>       The return value is of the same type and kind as the argument.  The   
   >>       fractional part of the model representation of ‘X’ is returned; it   
   >>       is ‘X * RADIX(X)**(-EXPONENT(X))’.   
   >>   
   >> _Example_:   
   >>            program test_fraction   
   >>              real :: x   
   >>              x = 178.1387e-4   
   >>              print *, fraction(x), x * radix(x)**(-exponent(x))   
   >>            end program test_fraction   
   >>   
   >   
   > Thank you, Steve.   
   >   
   > The following program, adapted from the Example in the documentation you   
   cite:   
   >   
   > program test_fraction   
   >    real :: x   
   >    x = 3.75   
   >    print *, fraction(x), x   
   >    print *, x * radix(x)**(-exponent(x))   
   > end program test_fraction   
   >   
   > prints:   
   >    0.937500000       3.75000000   
   >     0.00000000   
   >   
   > but I was expecting it to print:   
   >    0.937500000       3.75000000   
   >    0.937500000   
   >   
   > because the documentation says that fraction(x) is the same as  x *   
   radix(x)**(-exponent(x)) .   
   >   
   > I suggest that the documentation is wrong, and should say   
   > ' ... model representation of ‘X’ is returned; it is ‘X *   
   REAL(RADIX(X))**(-EXPONENT(X))’.   
   >   
   > Concerning the OP's finding that FRACTION(2.5) is printed as 0.9375:   
   > I cannot reproduce that with gfortran 11.5.0 or 13.3.1 or 14.2.1 which all   
   correctly calculate 0.625 .   
   >   
   > So to me this is a documentation bug, and possibly a bug in gfortran 12.2 .   
      
   Looks like you've found a documentation bug.  The Fortran standard   
   actually has   
      
     Result Value. The result has the value $X \times b^{-e}$,   
     where $b$ and $e$ are as defined in 16.4 for the representation   
     of X in the extended real model for the kind of X. ...   
      
   where I have marked-up the sentence in LaTeX.  IOW, when the gfortran   
   documentation was written, someone transcripted the equation   
   $X \times b^{-e}$ into Fortran without realizing RADIX returns an   
   integer.   
      
   --   
   steve   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)