Re: [avrdude-dev] Adding elapsed time

[Theodore A. Roth]

On Wed, 30 Jul 2003, Joerg Wunsch wrote:

> As Theodore A. Roth wrote:
>
> > It's usually a bad idea to test for equality with a float since the
> > equality may never be true due to the nature of floats (does 0 ==
> > 10^-400?).
>
> Comparision against constant 0.0 is guaranteed to work though.  0.0 is
> one of the FP numbers that has a fixed bit pattern, like +Inf, +NaN,
> -Inf, -NaN.  Note that there's a +0.0 as well as a -0.0, but they are
> supposed to compare identical (which they don't do on the AVR, btw. --
> there are more lurking compatibility bugs in that respect, but that's
> merely off-topic here ;-).
>
> Other powers of two can also be compared exactly (1.0 for example),
> but that's already an implementation detail, while the mentioned
> absolute constants are IMHO guaranteed to exist and to be comparable
> by the IEEE standard.

I don't think that that invalidates my statement. As a casual user of
floating point math, I would rather stick to a general rule that works
in all cases than remember all the gory details that are only true is
certain circumstances. I find that this makes my code more
maintainable and bug free.

I can see one use of an equality check though, to check for div by
zero (kinda puedo-code, so don't consider it anything but a hack
example):


float reciprocal (float foo)
{
  if (0.0 == foo)
     return NaN;
  else
     return 1.0/foo;
}


This is why I don't use C++ or perl if I can help it. Too many
features to remember how to use them all correctly (to be honest,
python isn't much better in that respect). But that is _way_ off-topic
now. ;-)

Now, back on topic. I think Brian's second elapsed time patch is more
in tune with the design of the report_progress() function.

Ted Roth