Hi,
thanks to all for your answers.
> I don't think your suggested comparison does
> give a total order.
Yes, that's the point.
> It seems you just need to sort the points in
> order of angle
Exactly.
> so within one quadrant you can just sort on tangent
> (p.y-z.y)/(p.x-z.x) together with some sorts to order the quadrants,
> so you only need data from one point at a time, which you should be able
> to express as one or more xsl:sort instructions easily enough.
Yep, that's true as well. Well, of course I could compute the angles,
and sort the points in order of them. If there is a tie and two points have the
same angle I would have to discard the one that is closest to my reference point.
But computing angles is based on slow inaccurate trigonometric functions.
The computation I presented in my first mail is fast and accurate. So I would
prefer it.
But anyway, I understand that it is not possible the way I thought. But I
will have a look at the fxsl library and try to figure out if it can help.
Again thanks for the answers and the hints :-)
Claus