William F Pokorny <ano### [at] anonymousorg> wrote:
> On 6/23/22 15:48, Chris R wrote:
> > I took my main scene and eliminated everything but the transparent box, and even
> > added media to it and everything looked fine.  I added one scene element at a
> > time back and it was all fine until I added in the drip of water I had modeled
> > falling from the downspout.  As soon as I added that object back into the scene,
> > the transparent box with media became a reflective mirror instead!
> >
> > The drip of water is modeled as a quartic.  It had a material to look like
> > water, (so transparent, ior 1.33, reflective, etc.).  When I changed the
> > material to just a simple gray pigment, the media box was no longer visible or
> > reflective, but the media also disappeared.
>
> Unsure...
>
> What version of POV-Ray is being used?
>

I am using the latest 3.8 beta.

> Are the shapes involved part of the same union or merge?

My scene, as is my usual practice, is a single union of all of the individual
objects in the scene.

>
> The only obvious thing I can think of would be if the water material was
> defined by name and happened to have the same name as the one used for
> the media box - and the water material definition was seen last.
>

The material for the media box was defined inline with the box declaration, so
there could be no name collision.

> If you can get a simpler scene showing the issues I'll add it to my
> collection for debug, though no promises as to when I'd dig. I've been
> busy with other than POV-Ray of late.
>

I will play around with it and see if I can find something simple and
repeatable.  I had the scene down to just a few elements, but there are a lot of
my own include files involved, so it will take a bit to try and reproduce it.

> --- Deeper and less likely possibility...
>
> With some older shape types - due ill advised automatic polynomial order
> reduction code combined with the lack of tight automatic / manual
> bounding for some shapes - we sometimes end up with an additional root
> (and so another surface) being introduced(a) by the auto-order reduction
> itself.
>
> Maybe something like this is going on but, that surface by order
> reduction would have to have a very particular position to see the
> results you posted.
>
> A test for this fail would be to change your quartic to a simple sphere
> or similar and see if the apparent reflection propagation to the box
> problem persists?
>
> (a) - One of the shipped scenes long has had a vertical line in the
> result due the auto-order reduction changing two roots (4th order eqn)
> to three roots (third order eqn) for one or more of the scene's shapes.
> I cannot remember which it is for sure - primitiv.pov maybe? The povr
> branch fixed this particular exposure(b) by ripping out the automatic
> order reduction code.
>
> (b) - Order reduction introduces slight root/surface shifts in the
> intended roots even when it mostly works in a consistent way. The
> reduction code also sometimes flips 'sturm on' to fixed solvers, without
> notice due the order reduction, on a per ray->surface equation basis.
> The sturm vs fixed solver results for identical equations are often not
> themselves identical and this occasionally introduce at same surface,
> subtle, consistency issues - which are sometimes visible. And there's
> more 'solver noise' possible due the order reduction I won't detail -
> partly because I no longer remember all the possible implications!
>
> Bill P.

Thanks, if I can find some time to reproduce the problem with a simple example
I'll post it.  Would you prefer it in bug reports, sources, or here in the
images newsgroup?


-- Chris R.

Post a reply to this message