Re: [Gnuplot-bugs] gnuplot-4.2.3 (and earlier): serious floating-point bug, and a simple fix

[Hans-Bernhard B. <HBB...@t-...>]

Nelson H. F. Beebe wrote:

> The correct way to compute a Euclidean norm is NOT sqrt(x*x + y*y),
> but hypot(x,y).  

There's a flaw in that argument that's arguably just as big as the bug 
you found: hypot() is, for practical intents and purposes, a 
non-standard function.  Neither ISO Standard C90 nor POSIX have it.  So 
using hypot() would make gnuplot a lot less portable.  The fix is, in 
other words, too simple to be usable.

Does solving a highly unrealistic underflow/overflow problem (let's face 
it: *nothing* in the real world is as small as 1e-160 times something 
else) really warrant breaking compilation on quite a number of platforms?