On Sunday 01 December 2002 12:52, mth wrote:
> As part of the conversion to the cdk Atom class I noticed that the
> physical coordinate representation of atoms (in Angstroms) went from float
> -> double ... Point3f -> Point3d.
>
> It really doesn't matter to me. And remember that I am not a chemist ...
>
> Q: What was the reasoning behind using doubles instead of floats? Is it
> the case that this level of precision (or range) is needed?
>
> Previously, there was a mixture of floats and doubles used in the
> implementation. I had been working my through eliminating all doubles from
> the implementation and converting everything to floats.
>
> Since the new Atom class uses double, I am getting ready to go through the
> transform/rendering code and convert it all to doubles.
>
> Q: I now that the answer is yes, but ... Please confirm that the Atom
> location will continue to be Point3d (double).

Yes, I noticed this too... Haven't had time to report this yet... I did not 
give it much attention until I started subclassing the CDK classes... In 
general one would like to use double, and hence CDK uses that... As we 
are now basing Jmol on CDK, we should use doubles instead  of floats too...
Even though in our case doubles maybe a bit of an overkill...

I did not report it yet, as I was wondering if the increased precision had
effects on the performance... (which it should...) but haven't had time to do 
some real test... though the effect seems rather marginal... but this may 
also be caused by the removal of a number of method calls... which was one of 
the advantageous effects of the move to CDK ...

Anyway, as you already said, we should start using doubles instead of 
floats... 

And yes, it is unlikely that CDK will start using floats...

Egon


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf
_______________________________________________
Jmol-developers mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/jmol-developers

Reply via email to