[Python-Dev] FP vs. tutorial (original) (raw)

Guido van Rossum guido@digicool.com
Mon, 21 May 2001 18:33:17 -0400

+ Yes, I'm serious about not including tutorial examples with platform-dependent output, unless they're explicitly meant to illustrate non-portable code.

Sure. Most examples can be rewritten to avoid platform-dependent output. But there should be one section on floating-point inaccuracies that shows a few of the kind of things you can expect on a typical platform, and 1.1 -> 1.1000000000000001 is pretty common.

+ Specific small examples notwithstanding, there is no uniformity across platforms in the last digit or so, because not even the IEEE- 754 standard requires that (while C is much sloppier than 754), and vendors generally don't implement anything better than the minimum necessary when it comes to f.p. (Sun is a notable exception).

So we'll have to add something like "the actual inexact output you see may differ from the inexact output in this example".

+ Happy to add text explaining the existence of surprises, and providing a URL. Do the floating-point morons on Python-Dev find this one comprehensible?:

http://www.lahey.com/float.htm

I was thinking more of immortalizing this one:

http://www.python.org/cgi-bin/moinmoin/RepresentationError

This can serve as a nice self-contained section on f.p. surprises.

--Guido van Rossum (home page: http://www.python.org/~guido/)