Celestia Forums

I have some models in free-space (L5). When they undergo an eclipse of the Sun by the Earth (a relatively regular occurrence at L5/L4), they are no longer lit by any light source. At this time they "blink out" completely and are invisible for what I assume is the duration of totality.

I think this is something that can happen in other 3D programs when there's no light source, so was wondering if this is a fundamental "feature" of the graphics engine or is it a bug?

Obviously, there's workarounds, but they're not nescessarily very elegant, or realistic, but firstly I'd just like to find out what other peoples experiences are of this behaviour, and if this is something which is a bug and therefore fixable.

Cheers

What symptoms do you see in which of the different render paths? (Type Ctrl-V to step through them.)

My understanding is that the current code in Celestia is not intended to cast shadows on or by 3D models. There used to be code that tried to do it, but it was extremely buggy.

It's not that there's no shadows Selden, it's that the models don't get rendered at all!! They simply disappear and where they were is empty space.
They still exist as objects in the solar system, they're just invisible (as if they'd reached an "Ending" date), and then when the sun comes back, they re-appear just as suddenly. (EDIT:about 50 minutes in real-time later)
Maybe they've gone into hyperspace

I'll check out the render paths as you suggest. BTW: v1.4.0

EDIT: Happens in both basic and multi-texture (I assume multi-texture means OpenGL 2.0 path??)

Sorry: Multitexture is not OpenGL 2.0.
Celestia has at least 5 Render paths, some of them available only on Nvidia cards:

Basic
Multitexture
OpenGL Vertex Program
OpenGL Vertex Program/Nvidia combiners
OpenGL v2.0

I managed to create a relatively simple test case. It fails (the model disappears) in all Render paths.



And here's a Cel:// URL for just before the model blinks out. On slower systems, Celestia may not open its window until after the "blink" has happened, so you may need to click on it again after Celestia has started.

cel://Follow/Sol:Earth2:l5/2027-08-27T1 ... 91&lm=1024

I'll move this thread to the Bugs forum.

Thanks Selden,

Selden wrote:Sorry: Multitexture is not OpenGL 2.0.

Not as sorry as I am!!
Further notes:
1. I tried a few different models to rule out it being a problem with my model and it happens regardless of the model.
2. If I put an extra sun nearby, the model no longer disappears, as its now illuminated by that light source..
(Haven't tried making textures emissive to see if that makes a difference)

**UPDATE**
Since logging this bug, I have experienced models disappearing for extended periods of time - days or months (even when lit by the sun), so this problem does not seem to be limited to solar eclipses.

I haven't worked out yet what the common circumstances are, but one thing I have noticed is that it does appear to be more likely to happen the further the model is from the Sun. eg. Models at Jupiter seem to suffer more from this problem than those at earth. I'm beginning to wonder if this problem may have something to do with model albedo and the amount of light available...

I'd be interested to know if other people have experienced this problem, and what were the circumstances,

ie.
Date/Time period of disappearance,
location,
model involved,
model albedo.
any transparent parts on model?
View->Render settings,
Render Path,
or anything else you think may be relevant....

Hopefully, this will highlight what the common circumstance and cause of the problem is, so please log your experiences here.

selden wrote:Sorry: Multitexture is not OpenGL 2.0.
Celestia has at least 5 Render paths, some of them available only on Nvidia cards:

Basic
Multitexture
OpenGL Vertex Program
OpenGL Vertex Program/Nvidia combiners
OpenGL v2.0

I managed to create a relatively simple test case. It fails (the model disappears) in all Render paths.

Code: Select all

# demo of model disappearing when going into shadow

# fake earth so L5 and Earth2 orbits are coplanar
# making an eclipse easier to find

"Earth2" "Sol"
{
   Texture "earth.*"
   Radius 6378.140
   
   EllipticalOrbit {   
      Period            1.0000
      SemiMajorAxis     1.0000
      Eccentricity      0.0167
      Inclination       0.0001
      AscendingNode   348.739
      LongOfPericenter 102.947
           MeanLongitude      0
   }
}

"l5" "Sol/Earth2"
{
   Mesh "iss.3ds"
   Radius   100
   Emissive true # to make it easy to see

   EllipticalOrbit {
      Period           27.321661
      SemiMajorAxis    384400
      Eccentricity     0.054900
   }
}

And here's a Cel:// URL for just before the model blinks out. On slower systems, Celestia may not open its window until after the "blink" has happened, so you may need to click on it again after Celestia has started.

cel://Follow/Sol:Earth2:l5/2027-08-27T1 ... 91&lm=1024

I'll move this thread to the Bugs forum.

yes

symaski62 wrote:yes

symaski62,

Does "yes" mean:
"yes, you can see the model and it doesn't disappear"
or:
"yes, you have the problem and the model disappears"

Your image suggests that you can see the model OK during the entire period of eclipse. Is this what you mean?

# demo of model disappearing when going into shadow

# fake earth so L5 and Earth2 orbits are coplanar
# making an eclipse easier to find

"Earth2" "Sol"
{
Texture "earth.*"
Radius 6378.140

EllipticalOrbit {
Period 1.0000
SemiMajorAxis 1.0000
Eccentricity 0.0167
Inclination 0.0001
AscendingNode 348.739
LongOfPericenter 102.947
MeanLongitude 0
}
}

"L5" "Sol/Earth2"
{
#Mesh "iss.3ds"
Radius 100
Emissive true # to make it easy to see

EllipticalOrbit {
Period 27.321661
SemiMajorAxis 384400
Eccentricity 0.054900
}
}


not eclipse !

" Emissive true # to make it easy to see " "it broken"

tried your test. It worked fine. Model did not disappear. EDIT: correction, it did disappear after all. I had the time rate up too fast so I "missed" the blink the first time.

I'm not sure if you are talking about the same problem that I am seeing. The picture you posted shows the model still visible, although dark. This would be correct behaviour in an eclipse. The problem I am talking about is where models blink out completely...
Here is a demo of what I'm talking about:

1.7MB mpg:


Linux Users: http://www.sitesled.com/members/tentsforpaddies/files/video/eclipse.mpg

Is this the same problem as you are having?

Hi there,
I just made the test with Selden's URL and the ISS does not disappear.
I cycled through all the render paths. All are working here.

EDIT: Captain, I cannot see your pic. All I get is a black page.

Boux wrote:EDIT: Captain, I cannot see your pic. All I get is a black page.

It's not a pic, it's a video (1.7 MB). I suspect that thinking it was a pic, you didn't wait for it to load. If you're on dial-up you might have to be patient (that blank page could take 5 mins to load) or alternatively, right-click and save it to your hard-drive, then open it with any mpeg-player. If you haven't already, you may want to associate MPG type files with your player. (Assuming you're using windows) If you're MAC user, try quicktime.

Let me know if still can't see it.

Yep, it's an mpeg video.
The problem is that I am running Linux.
I will be able to watch it only if I can download it or have a direct link to it (not through a thumbnail).
Maybe you can edit your post and give an URL to the file.

Boux wrote:Yep, it's an mpeg video.
The problem is that I am running Linux.
I will be able to watch it only if I can download it or have a direct link to it (not through a thumbnail).
Maybe you can edit your post and give an URL to the file.

DONE!

Has anyone else experienced models disappearing outside the circumstances described above?
It seems to me that this problem is not nescessarily directly related to eclipses or the absence of light (because I've seen it happen to fully lit models), but perhaps the angle/orientation of the model to the light source, or perhaps they're 3 separate bugs....

Hi Captain,
The video clearly shows the model disappearing as soon as it is in the eclipse shadow.
As said, I cannot repoduce this here (and I never noticed such a behaviour in the past with any kind of model).
Really wierd.
Does this happen with any kind of model?
The visual effect reminds me of the nebula bug in FTxx where nebulae were disappearing when within a narrow angle in the center of the field of view.

Chuft-Captain asked,
"Has anyone else experienced models disappearing outside
the circumstances described above?
It seems to me that this problem is not nescessarily directly related
to eclipses or the absence of light (because I've seen it happen to fully
lit models), but perhaps the angle/orientation of the model to the
light source, or perhaps they're 3 separate bugs..."


You can make a thousand stars disappear.
Here is the URL

cel://Follow/Sol/2006-01-05T12:15:27.97 ... 3&lm=49152

Increase the star magnitude with ] so 15th magnitude stars are seen.
The alignment is sensitive so the stars are missing only when
you put the celestial grid's north pole at Sol. (or south pole at Sol).

Curiously, I seem to have just the opposite problem. When viewed in conjunction with a source of illumination, be it star or planet (okay, reflected illumination in the latter case), many models suddenly become emmissive.

As for dissapearing models, that does indeed happen with models that are made to be large. I've been playing with Banks orbitals lately (basically small ring worlds that orbit a star like a planet rather than surround it), and if they exceed 10,000 kilometers in diameter, they will dissapear when eclipsing a sun. Curiously, if I make them larger than 50,000 kilometers, they will not appear at all.

...John...

Curiouser, and curiouser...

There's definitely something strange going on here.

Boux, I think the reason you are having trouble replicating Selden's demo is that the behaviour of this bug is not consistent. What I have discovered is that the disappearance happens at different times and for different periods of time. This is seemingly random, but suggests to me a dependence on the orientation of light-source, object, and observer.
This makes it very hard to consistently replicate. (It also only blinks out for about 20-30 secs, so if you're running the time-rate at 100x or more, you'll miss it).

I ran Selden's test in a single session and captured the blink-out to video (see the clips below).
In the first clip, I was "following" the object, I then pressed "G" to goto the object and then "Y" to sync orbit. This changes the orientation, and the time of disappearance is therefore different in the second clip.
In past sessions, the time of blink-out has been different yet again.

So, there seems to be several variables involved, but seems to be somehow related to orientation of light-source, object, and observer.

Interestingly, the depth-sorting problem associated with transparent models also seems to exhibit behaviour dependent on orientation of light, object, and observer, so I wonder if there is some fundamental defect responsible for both problems.

Anyway, my hope is that this thread will provide some clues to the developers that may help identify exactly what's going on.

http://www.sitesled.com/members/tentsfo ... link01.avi

http://www.sitesled.com/members/tentsfo ... link02.avi

Boux wrote:Does this happen with any kind of model?

Read the beginning of the thread.

GM,

The plot thickens....however I couldn't replicate your disappearing stars. Perhaps you might post before and after pics, or a short video clip showing the effect?

Cheers