So, I've worked some more at the focal-blur
missing-objects-bug. After doing some more
stuff, a new bug popped up. Now, objects
are traced incorrectly (media missing, highlights
missing, shading incorrect), and the effect varies
when changing only a translate-Value!

The actual code was something like
object{X rotate A rotate B translate C}

where C was something like
vnormalize(Position)*(.25+Length*1.25)

Changing vnormalize(Position) to

Position/Formerly_Calculated_vlength

resulted in the incorrectly rendered area moving
to an other position.
Also, when using focal blur, entire bunches of
objects disappear, AND some are just plain
cut off! The edges are then traced blurred... The
image entitled "Focal-Blur+Translate" that
shows this is found in the beta-test.binaries.

I have absolutely no clue, what this is all about.
When removing the translate-line, the scene
seems to render correctly, though I haven't
tested that with focal-blur yet.

I'll be posting image-variations of this bug with
some minor description on actual changes in
scene-file as follow-ups to that image.

Sooooo... I'm getting kinda frustrated, because
the scene is supposed to become a present, but with
all those bugs, I can't achieve the realism I wanted, yet
alone finish some sort of scene (with objects disappearing).

I'm not pushing, just expressing my motives, don't
take it personally. I have said many times, and am about
to once again, that I really appreciate the work the
POV-Team does, but right now I'm kinda sad... :-(

And I said sad, not angry!

--
Tim Nikias
Homepage: http://www.digitaltwilight.de/no_lights/index.html

Post a reply to this message

Hi Tim,

since I don't have much time left within a week, I couldn't really follow
the whole thread about your problems. But reading your lates posting just
brought up one idea:

I know that your'e working with meshes with smoothed triangles (as I did
with the fern). Have you ever tried to create the triangles as stand-alone
objects and not as part of a mesh? If there are problems with the triangles
(e.g. normals) you'll see a message. When using a mesh the triangles are
dropped silently.
It is a simple test with low effort (simply remove the mesh {} keyword) so I
thought you may to try it.
If you already verified that and it gave no further result, please excuse
me. As I mentioned, I didn't follow up the full thread.

regards
SY

Post a reply to this message

Hi!

I tested your suggestion, and yes, it does
work. But comparing 240MB Peak memory
to 1.6MB is something to think about, right?

Also, its not that just some triangles are missing
due to bad normals (though using your technique,
message window tells me there are some degenerate triangles
floating around), its that whole instances (object{}-placed)
of the mesh disappear, and some other are just cut off
above a certain line/height (I haven't found yet if
it is a certain line, or better, a certain percentage of
"viewing screen", or a certain height in POV).
Sometimes object simply vanish, at other times, they're
shaded incorrectly, or the interior is sort of messed up
(there, but weaker/stronger than when working
normally).

I am getting to the point of concluding that the bug is
somewhat the same as the one I posted on 17.03,
entitled "Media Bug (no media, only black objects appear)", because
I've just spotted a "version" of the image, where I can see
through other objects that lie behind the missing
meshes.

I'm still investigating all this, but I haven't found
any related bug yet. Also, I am really sad to find
that only one single person tried to confirm the
media-bug. I thought beta-testing was sort of a
group-effort, but I'm standing pretty much alone,
or at least I have that feeling.
Well, except for your answer and the Confirmant,
Peter Popov.



> Hi Tim,
>
> since I don't have much time left within a week, I couldn't really follow
> the whole thread about your problems. But reading your lates posting just
> brought up one idea:
>
> I know that your'e working with meshes with smoothed triangles (as I did
> with the fern). Have you ever tried to create the triangles as stand-alone
> objects and not as part of a mesh? If there are problems with the triangles
> (e.g. normals) you'll see a message. When using a mesh the triangles are
> dropped silently.
> It is a simple test with low effort (simply remove the mesh {} keyword) so I
> thought you may to try it.
> If you already verified that and it gave no further result, please excuse
> me. As I mentioned, I didn't follow up the full thread.
>
> regards
> SY

--
Tim Nikias
Homepage: http://www.digitaltwilight.de/no_lights/index.html

Post a reply to this message

Hoi!

> I tested your suggestion, and yes, it does
> work. But comparing 240MB Peak memory
> to 1.6MB is something to think about, right?

Of course. It wasn't meant to be a final solution or work-around. Just to
see what's going on with the triangles/normals.

> floating around), its that whole instances (object{}-placed)
> of the mesh disappear, and some other are just cut off
> above a certain line/height (I haven't found yet if

I see. Right, this is a different kind of problem. I didn't know that even
single triangles are 'cut' somewhere.

> any related bug yet. Also, I am really sad to find
> that only one single person tried to confirm the
> media-bug. I thought beta-testing was sort of a
> group-effort, but I'm standing pretty much alone,
> or at least I have that feeling.

Uhm, I can't really tell since I peek to the newsgroup about once a week or
so. Speaking for myself: I simply don't have the time. Neither to use 3.1g,
nor to test the beta's, nor to download new betas through a 32kB dial-in
line (even the 1MB whatever-compilation is painful for loading)....
But I agree with you in thinking that beta testing should be a "group-job".
At least to spread the test cases and multiply the "way of usage".

BTW: what is the special thing which prevents you to use 3.1g?

> Well, except for your answer and the Confirmant,
> Peter Popov.

I'm sorry that it didn't really help, but it's interesting that it did
return some further result.
regards
SY

Post a reply to this message

> Of course. It wasn't meant to be a final solution or work-around. Just to
> see what's going on with the triangles/normals.
>

Well, it works fine as a workaround for me, cause I've got
512 MB RAM... And lots of GB free space for Disc-Swapping...

> Uhm, I can't really tell since I peek to the newsgroup about once a week or
> so. Speaking for myself: I simply don't have the time. Neither to use 3.1g,
> nor to test the beta's, nor to download new betas through a 32kB dial-in
> line (even the 1MB whatever-compilation is painful for loading)....
> But I agree with you in thinking that beta testing should be a "group-job".
> At least to spread the test cases and multiply the "way of usage".
>

I don't meant to push anyone to do something they don't have
the time to. And of course you should only feel meant if
you actually download the betas, but just ignore bugs you've
encountered...

>
> BTW: what is the special thing which prevents you to use 3.1g?
>

Enhanced Media, trace() and photons/radiosity are the
things I really need if I want to create a very realistic image,
that also traces in less than a year to look good... ;-)

> I'm sorry that it didn't really help, but it's interesting that it did
> return some further result.
>

It helped! Perhaps someone got the final cue due to that,
but it always, even if its only to define the problem a little
more detailed.

--
Tim Nikias
Homepage: http://www.digitaltwilight.de/no_lights/index.html

Post a reply to this message

I've done some more research.

Using mesh and translate

Without Focal-Blur
***************
Ghost-effect
Objects are traced, but averaged
with background-color/sky_sphere

With Focal-Blur
************
Ghost-effect area simply vanishes

Using mesh and no translate

Everything works as expected

Using union and translate
Everything works as expected

The bug appears only with
Quality set to 6 or higher, so
probably has to do with textures.
(No warps in the textures, doesn't
matter if texturing is done for each
instance, or only once for the mesh)

When using no transparency on the
textures, the bug also doesn't appear,
but with any transmit-value, the
ghost-effect isn't altered.

When using -UV (not using vista
buffers), the ghost-effect doesn't take
place, but there's no object either.

Light Buffers, Hierarchy in Meshes,
Unnessecary-Bounding-Removal,
Automatic-Bounding, Split_Unions,
these all don't change a thing.


triangles is of no concern, I've added some lines
of code to check for degenerate triangles, and
my algorithms always return unit-length-normals.

Comparing union{}-workaround with mesh{}:
240MB vs 1.6MB Peak Memory...

I don't know if anyone can read any sense
out of this, but always feel free to ask.

--
Tim Nikias
Homepage: http://www.digitaltwilight.de/no_lights/index.html

Post a reply to this message