(fromRational (1%5)) :: Double

Scott Turner p.turner@computer.org
Thu, 19 Apr 2001 06:44:37 -0400

At 10:02 2001-04-19 +0400, S.D.Mechveliani wrote:
>Hugs finds
>                        Prelude> (fromRational (1%5)) :: Double
>                        0.2
>Double  has mantissa in a binary representation. So, the interpreter 
>has to convert a decimal 5 to a binary 101B, evaluate  1B / 101B  
>obtaining an infinite sequence of binary digits and take the first  m
>of them required for Double. 
>Printing the result should yield something like  0.1999...

You've got it all correct.  The trick is in the "something like" which
happens during printing.  The exact result would be 
which is unwieldy, and can be misleading about its precision.  Those last
38 digits must not be mistaken for something useful.  Fortran would show a
decimal precision related to the binary precision of the DOUBLE, i.e.
In Fortran, the number of decimal digits displayed is independent of the
numeric value.  But this is not the only possible choice.  Haskell uses a
technique which is becoming more popular, in which the number of digits
displayed depends on the value.  It finds the minimum number of digits
which will convert back to the original binary number.  0.2 when converted
to binary has the same value
therefore 0.2 is printed.

Scott Turner
p.turner@computer.org       http://www.billygoat.org/pkturner