PEP 260: simplify xrange()

Michael Hudson mwh at python.net
Wed Jun 27 20:16:52 CEST 2001


Jeff Shannon <jeff at ccvcorp.com> writes:

> To make a long story short, the discussion regarding FP boils down
> to the fact that "1.0" cannot be exactly represented in FP.  

Not true.

> Every time you increment your "frange", you are introducing nasty
> rounding errors, and by the time you've done several thousand
> iterations, they're going to be very significant rounding errors.
> And the worst part is that there will be *no indication* of where
> the wrong answers are coming from unless you happen to monitor your
> frange index throughout a debugging run...  and even then it'll only
> be clear if you know that the FP problem exists.

This is true, though.

Glib truism:

  Floating point arithmetic *is* exact (given a rounding mode), but
  it's not exactly what you think.

My pedant-gland throbs a bit when I see people write "floating point
numbers are approximate"... 

Cheers,
M.

-- 
  That one is easily explained away as massively intricate
  conspiracy, though.            -- Chris Klein, alt.sysadmin.recovery



More information about the Python-list mailing list