Re: axisScaleMin/axisScaleMax

[Fritz Z. <za...@ee...>]

On Tue, 17 Jul 2001, Steven Knight wrote:

> On Tue, Jul 17, 2001 at 10:01:18PM +0200, Fritz Zaucker wrote:
> > In a case, where there is no clear invalid value (and why should
> > HUGE_VAL be invalid?), wouldn't it be better to return -1/0 for
> > invalid/valid axes and return the min/max value in a variable passed
> > as argument?
> >
> > Sure, it makes for a bit more lengthy code, but at least it is clear
> > then and does not depend on HUGE_VAL being available on all platforms.
>
> I checked and HUGE_VAL is defined on win32, linux, irix and sun.  Also on all
> these systems HUGE_VAL is defined in math.h.
>
> >
> > Cheers,
> > Fritz
> >
> > P.S.: One could even make a function axisScaleMinMax() returning both
> >       values, as they might be needed together most of the time
> >       anyway.
> >
>
> While a function like axisScaleMinMax() might be useful, I would suggest
> keeping axisScaleMin() and axisScaleMax() for the simple reason that
> there may be sometimes when you only need the min or max value on the
> axis scale.

I have no problem with having both, but I still think that using
HUGE_VAL is sort of a hack (although not a bad one). Or is somewhere
said that HUGE_VAL should be considered invalid?

Another clean way would be to return a NaN (which can be tested with
isnan(3C). Unfortunately it is not straightforward to create a NaN (or
at least I don't know it). My coworker Tobias Oetiker (author of MRTG
and RRDtool) uses

    #define NaN 0.0/0.0

but says that one has to deal with issues of floating point exceptions
on various OSes.

So perhaps using ou of band signaling (as I had originally suggested)
is the cleanest way ...

Cheers,
Fritz

-- 
Dr. Fritz Zaucker, Head IT Support Group
Department of Electrical Engineering,  Federal Institute of Technology
ETZ J97, Gloriastrasse 35, CH-8092 Zurich, Switzerland
Tel.: +41-1-632-5241 Fax: +41-1-632-1194 http://people.ee.ethz.ch/~zaucker/
E-mail: za...@ee... (see home page for PGP key)