POV-Ray : Newsgroups : povray.bugreports : PovWin Crash: Seems related to meshes and reflection{1} : Re: PovWin Crash: Seems related to meshes and reflection{1} Server Time
14 May 2024 20:24:57 EDT (-0400)
  Re: PovWin Crash: Seems related to meshes and reflection{1}  
From: Reactor
Date: 8 Dec 2009 18:45:00
Message: <web.4b1ee421b8f29f1ad6143abb0@news.povray.org>
Martin <don### [at] onocom> wrote:
> Hello!
>
> I was testing some texturing in meshes (because i'm very bad texturing
> :P) exported from Wings 3D, when pov crashed while rendering



Duplicated it on my own!  Interesting, never seen this before.  Ok, tested an
identical scene under 64-bit Windows XP SP2, 32-bit version of POV 3.6.2, and
the scene renders just fine.  The crashes occurred in the 64-bit version 3.7 b34
(also under Windows XP 64 SP2)

To trigger the crash seems to require a mesh with more than about 5000 vertices,
I was able to get the crash with a 6000 vertex mesh.  The key seems to be
something around the normals and the textures applied on a per vertex basis, as
I exported and rendered the same mesh with:
* uv_vectors and normal_vectors (crashed)
* normal_vectors only (crashed)
* uv_vectors only (completed successfully)
* neither uv or normal vectors (completed successfully)
(I ran meshes with only smoothed normals and a combination of smooth and hard
edges to get the above using a max_trace_level of 1 )
Then, with some fancy find-and-replacery, I removed the texture index after each
face, removed the texture_list{}, then re-ran a crashing mesh with the texture
applied to the entire mesh.  This allowed the render to complete successfully.

Here is where it gets weird:  While minimizing a scene for this, I found that
changing the aspect ratio very slightly can sometimes prevent a crash.  That is,
changing  right <4/3,0,0> to right <4.0001/3,0,0> allow renders that previously
crashed to complete successfully.  I tried a few others, but could not find any
sort of pattern - sometimes a given aspect ratio would crash, other times it
would complete successfully.  Same with changing the output resolution - crashes
became more random instead of certain.

A problem I ran into while trying to duplicate this is that... well, I couldn't
always.  I wanted to (partially) rule out the exporter outputting improperly
formatted meshes, so I exported some using the subdivision option and others
smoothed manually.  The manually smoothed ones did not crash, but the mesh
topography was slightly different (different number of vertexes, slightly
different shape)  from running the smooth command twice, as opposed to telling
it to subdivide for 2 steps (this isn't so strange, I suspect it is because the
subdivide command tessellates into triangles first, but the mesh had
quadrilaterals when I smoothed).

I hope this helps someone who has a better understanding of these things than I
do. ;)

-Reactor


Post a reply to this message

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.