MantisBT - ParaView
View Issue Details
0003767ParaView(No Category)public2006-09-13 16:352009-12-09 16:17
Kevin Hobbs 
Berk Geveci 
normalmajoralways
closedfixed 
 
 
0003767: Clip hang
Some time recently clip has stopped working in Paraview from CVS HEAD. All I have to do is open paraview, bring up a source like wavelet, and clip it, and the render window becomes mostly unresponsive. I cannot rotate the object, change the clipping plane, and the object drawn does not change when the window is resized. I can click on buttons and delete filters but the only way to render something else is to close paraview.

Paraview was compiled with gcc.

Paraview was linked against Mesa-6.4.2 without dri compiled from source.

This is Linux gargon.hooperlab 2.6.17-1.2174_FC5 #1 SMP Tue Aug 8 15:30:44 EDT 2006 x86_64 x86_64 x86_64 GNU/Linux
No tags attached.
txt bt.txt (7,147) 1969-12-31 19:00
https://www.vtk.org/Bug/file/5667/bt.txt
Issue History
2009-12-09 14:51Berk GeveciProject@3@ => ParaView
2009-12-09 16:17Berk GeveciStatustabled => closed
2009-12-09 16:17Berk GeveciResolutionopen => fixed
2011-06-16 13:10Zack GalbreathCategory => (No Category)

Notes
(0004914)
Kevin Hobbs   
2006-09-14 11:41   
The render window also freezes when I threshold the wavelet.

I am able to save the output of both the clip and threshold to a '.vtu' file and in both cases the render window freezes as soon as I load the saved data.

Could this be an unstructured grid problem?
(0004915)
Kevin Hobbs   
2006-09-14 13:15   
I changed to debug mode and did "make clean && make" and I get :

terminate called after throwing an instance of 'std::bad_alloc'
  what(): St9bad_alloc
(0004916)
Kevin Hobbs   
2006-09-14 13:40   
I produced a backtrace frome the 1.1 GB core that paraview produced and atached it in bt.txt.
(0004917)
Kevin Hobbs   
2006-09-14 14:32   
#8 0x00002aaaaf3c899b in vtkOpenGLTexture::Load (this=0x2a43e80, ren=0xa6b1a0) at /home/kevin/kitware/ParaView/VTK/Rendering/vtkOpenGLTexture.cxx:202

This line is :

 resultData = new unsigned char [rowLength*ysize];

ddd tells me that rowLength and ysize are monstrous negative numbers
when I put a breakpoint there. xsize is also a huge negative number.
size[0] = 256, size[1] = 1, and size[2] = 1 do these make sense?
(0004925)
Kevin Hobbs   
2006-09-15 14:14   
I did a clean build today starting with "rm -r *" in the build directory. The behavior is the same as yesterday except now ParaView does not immediately exit when I update the clip, and it produces the message below when I click on the render viewport.

ErrorMessage
# Error or warning: There was a VTK Error in file: /home/kevin/kitware/ParaView/GUI/Widgets/vtkKWApplication.cxx (999)
 vtkPVApplication (0x7024c0): TclTk error: Uncaught exception: St9bad_alloc

Uncaught exception: St9bad_alloc

    while executing
"vtkTemp2 MouseAction 2 1 274 87 0 0"
    (command bound to event)
ErrorMessage end
(0004932)
Kevin Hobbs   
2006-09-15 16:44   
I can now relate this bug to one of the tests that fail on gargon.hooperlab, that is :

http://paraview.org/Testing/Sites/gargon.hooperlab/Linux-c++/20060915-1537-Experimental/Results/__GUI_Testing_Tcl_CalcInput-batch-ParaView.html [^]

I copied the file "/home/kevin/kitware/ParaView/GUI/Testing/Tcl/CalcInput.pvs" and commented out the lines that source things and close the window :

diff /home/kevin/kitware/ParaView/GUI/Testing/Tcl/CalcInput.pvs CalcInput.pvs
4,5c4,5
< source [file join [file dirname [info script]] CommandLineOptions.tcl]
< ParseCommandLine $pv(vtkTemp43) $argv $argc
---
> #source [file join [file dirname [info script]] CommandLineOptions.tcl]
> #ParseCommandLine $pv(vtkTemp43) $argv $argc
49,50c49,50
< source [file join [file dirname [info script]] Compare.tcl]
< Compare $Application $argv $argc
---
> #source [file join [file dirname [info script]] Compare.tcl]
> #Compare $Application $argv $argc
52c52
< $Application Exit
---
> #$Application Exit

When I run "paraview CalcInput.pvs" ParaView is left in the same frozen render viewport state I've described before. I can see only the sphere. If I click on the render viewport I get the message below and ParaView exits.

terminate called after throwing an instance of 'std::bad_alloc'
  what(): St9bad_alloc
Aborted
(0004960)
Kevin Hobbs   
2006-09-18 11:43   
It looks to me like FindPowerOfTwo is returning crazy numbers. It proceeds through the for loop and sets size to 265. Then after glGetIntegerv, maxDimGL is a huge negative number so of course size is set to this huge negative number. I don't have debug symbols for glGetIntegerv, so I have no idea what it's doing.
(0004969)
Kevin Hobbs   
2006-09-19 09:37   
I noticed that the of the ClipCut tests run on gargon batch failed while the ClientServer and RenderServer tests passed. I tried my little test where I bring up the wavelet source and clip it in paraview with offscreen rendering on and it works. Without offscreen rendering the viewport hangs and mesa complains about the invalid context.

I just tried some things on our Beowulf cluster as pvserver and my workstation as the pvclient. I tried opening up the wavelet source ( worked ), contouring ( worked ), color by process id ( worked ), but as soon as I clip anything the pvserver dies with a segfault.

The pvserver complained as below but it was run with --use-offscreen-rendering. The server was built with mesa as the only GL.

Connected to client
Process id: 0 >> ERROR: In /home/khobbs/kitware/ParaView/VTK/Rendering/vtkXOpenG
LRenderWindow.cxx, line 482
vtkXOpenGLRenderWindow (0x936aba0): bad X server connection.


Process id: 0 >> ERROR: In /home/khobbs/kitware/ParaView/VTK/Rendering/vtkXOpenG
LRenderWindow.cxx, line 309
vtkXOpenGLRenderWindow (0x936aba0): bad X server connection.


mpiexec: Warning: task 0 died with signal 11 (Segmentation fault).
(0004975)
Kevin Hobbs   
2006-09-19 15:16   
It looks like the bug 3767 and many of the tests that were failing on my
computer were due to my use of the osmesa library. Changing only the
VTK_OPENGL_HAS_OSMESA cmake setting got rid of most of the errors and
the hang.

The bug is :
http://paraview.org/Bug/bug.php?op=show&bugid=3767&pos=4 [^]

The failing tests are :
http://paraview.org/Testing/Sites/gargon.hooperlab/Linux-c++/20060918-1915-Experimental/Test.html [^]
(0005059)
Kevin Hobbs   
2006-09-28 08:09   
According to Utkarsh Ayachit the root cause of this bug is a bug in Mesa. He has provided a patch on the mailing list that avoids the bug. It has something to do with the 'HAVS mapper'. Here is his message :

http://public.kitware.com/pipermail/paraview/2006-September/003783.html [^]