[vtkusers] Mesa/VTK problem on RedHat 7.2 (also with Mesa 4.0)

[Dave Reed]

> From: Prabhu Ramachandran <prabhu at aero.iitm.ernet.in>
> 
> >>>>> "DR" == Dave Reed <dreed at capital.edu> writes:
> 
>     DR> When I moved from RedHat 7.1 to RedHat 7.2, I started getting
>     DR> missing polygons in simple VTK examples (using vtk3.2). It was
>     DR> suggested that I try Mesa 4.0 so I did, but the same thing
>     DR> still happens. I also tried VTK 4.0, but the same thing still
>     DR> happens. I completely removed the libGL and libGLU files that
>     DR> came with RedHat 7.2, installed Mesa-4.0, compiled VTK 4.0 and
>     DR> it still happens.
> 
>     DR> http://capital2.capital.edu/faculty/dreed/medical1.jpg
> 
>     DR> shows an example - there are polygons missing just under the
>     DR> nose.
> 
>     DR> The same program works fine (no missing polygons) with
>     DR> NVidia's drivers with a TNT2 card. So either it's a bug in
>     DR> Mesa or some strange interaction between VTK and Mesa. I
>     DR> suspect it's a bug in Mesa, but I'm not certain.
> 
> Yes, I have the same problem with VTK and Mesa on debian machines.  It
> all works right with nVidia's OpenGL.  I reported it here:
> 
> http://public.kitware.com/pipermail/vtkusers/2001-September/007664.html
> 
> Ann Gentile replied saying that the problem was with display lists.
> Read the full thread for details.
> 
> I've tested it with the following:
> 
>      Mesa 3.1: Horrible behaviour -- basically entire actors would
>      simply disappear when I did something (rotated the view or
>      changed property of something), this made using VTK very painful.
> 
>      Mesa 3.4.2: Much better behaviour but still there would be sudden
>      disappearance of actors.
> 
> I have checked this with both VTK 3.2 and VTK 4.0 with the same
> problems.
> 
> As regards testing if it really is display lists - if you are using
> VTK 4.0, Ann's patch wont work.  You can try this instead: In
> Rendering/vtkOpenGLPolyDataMapper.cxx (CVS tree as of 15th Nov.)
> change
> 
>      249   //
>      250   // if something has changed regenerate colors and display lists
>      251   // if required
>      252   //
>      253   if ( this->GetMTime() > this->BuildTime || 
>      254        input->GetMTime() > this->BuildTime ||
>      255        act->GetProperty()->GetMTime() > this->BuildTime ||
>      256        ren->GetRenderWindow() != this->LastWindow)
>      257     {
>      258     // sets this->Colors as side effect
>      259     this->MapScalars(act->GetProperty()->GetOpacity());
>      260 
> 
> To read:
> 
>      249   //
>      250   // if something has changed regenerate colors and display lists
>      251   // if required
>      252   //
>      253   if (1)
>      254     {
>      255     // sets this->Colors as side effect
>      256     this->MapScalars(act->GetProperty()->GetOpacity());
>      257 
> 
> 
> Note that I have not tried this myself but if Ann's patch is correct I
> guess this should be correct too.  According to Hank Childs
> <childs at hyper.llnl.gov> this problem occurs because of a problem in
> display lists.  Maybe you could confirm this behaviour by trying this
> change and checking if you still have missing polygons.  This might
> help the mesa folks.


Yes, making that change to the VTK code fixes the problem. So the
problem is with Mesa display lists and triangle strips?

Dave