"Warp" <war### [at] tagpovrayorg> wrote in message 
news:443e1278@news.povray.org...
> Chris B <c_b### [at] btconnectcomnospam> wrote:
>> If this is part of your aircraft, then you seem to be using 0.035 povray
>> units to represent about 1 metre.
>> If you use 1 metre = 1 povray unit, then you'll get far less of the sorts 
>> of
>> problem you are getting.
>
>  I don't think that's a problem, unless he is using measures of the order
> of magnitude of one nanometer.
>
>  Double-precision floating point numbers are quite flexible, and using
> 0.035 as unit is about exactly as accurate as using 1 as unit. It's only
> when you start going to real extremes (like using 0.000000035 as unit)
> when you might start getting problems.
>
> -- 
>                                                          - Warp

Hi Warp,

I was just going on the two problems that Gyscos has reported.

Alain indicated that scaling up by a factor of about 20 gets rid of the 
artifacts reported in this thread and I find that scaling up by about the 
same factor makes the two sets of artifacts from the thread 'Light passing 
through my wall' get smaller and then dissapear, because the camera ends up 
20 times further away and the artefacts seem to be of a fixed size.

I'm not sure the problem I looked at is necessarily to do with the accuracy 
of the floating point calculations (though it seems worth considering). In 
the simplified example I posted in the other thread, all of the objects were 
of a fair size. The camera is close-in to emphasize the artifacts, but not 
close enough to get to the sort of extremes you mention. In my example in 
that thread, commenting out the second box  makes the artifacts dissapear, 
which is odd as it's nowhere near the offending objects and is well out of 
camera shot. I looked at this on POV-Ray 3.6 and interestingly Alain has 
indicated that with a 3.7 beta the artifacts are always present.

So I'm not sure what the cause of either of these problems is, and I can see 
that it would take a fair time to change the units throughout his SDL, but 
that does seem to me to be a workaround for both problems and one that 
Gyscos may wish to consider. In addition, from a personal perspective, I 
found it quite difficult to look at the problem having to think about 
everything to 4 decimal places, but of course that's just down to what 
you're used to.

Regards,
Chris B.

Post a reply to this message