Re: [Gzz] OrthoCoordsys???, reverting

[Tuomas Lukka]

On Sat, Aug 24, 2002 at 12:34:46PM +0200, Benja Fallenstein wrote:
> Tuomas Lukka wrote:
> 
> ><me>:
> > 
> >
> >>>*backward*. I think it should be possible to find *some* way to 
> >>>implement this in GL quickly for now... we can always change it later.
> >>>     
> >>>
> >>Ok, I'll try to find a way to do that, but with the understanding
> >>   
> >>
> >
> >           // a REAL kludge
> >         key = new Integer(key.hashCode() % 27482963 * 427189 + 
> >         (keys[into]!=null?keys[into].hashCode() : 89761));
> >
> 
> :)
> 
> >This might be cheaper for the AWT side as well than the current code.
> >
> 
> I like it as a means of getting GL to run, but not as a permanent 
> solution: a), it's bound to make for interpolation errors at times 
> (e.g.: you show the same view in two windows side-by-side. Then there 
> are LOTS of coordsys/parent pairs which have the same pair of keys-- but 
> then the parents' parents (the two windows) are different, which the 
> kludge ignores. 

Actually, it *DOESN'T* ignore that, since the KEY is changed.
Which means that the parents of the coordsys have different keys.

Every child has as its key a combination of its own and its parent's
key.

They are unique. 

So real interpolation errors happen EXTREMELY rarely. A little less
rarely than purely random ints would meet but still quite rarely.

> And b), I don't buy into it being more efficient: it 
> creates one object per coordsys, which is precisely what I'm trying to 
> avoid in AWT. This is a place where I would like to see some serious 
> benchmarks first before changing anything ;-)

Yes, actually I was thinking of storing ints instead of keys for efficiency.

        Tuomas