1. 23 Mar, 2013 2 commits
    • Philippe Pébay's avatar
      Replaced the IndexingMode ivar with an TransposedRootIndexing bool · cc99b31b
      Philippe Pébay authored
      This is cleaner and makes more sense since there are only 2 modes
      of indexing root cells: either IIK or KJI. It is not considered
      that any other one would be potentially useful for anything.
      So I am either transposing the default (i.e., VTK-style, that is
      x-axis minor and z-axis major) to the x-axis major and z-axis
      minor used by some applications.
      
      Change-Id: Ic79f23622402c9579eefd57ea09743fb72788aa2
      cc99b31b
    • Philippe Pébay's avatar
      Fixed ShallowCopy defect · 12667426
      Philippe Pébay authored
      Change-Id: I68bfb91b87ec0708b9b11517fbfd758411a8b40a
      12667426
  2. 22 Mar, 2013 3 commits
  3. 21 Mar, 2013 2 commits
  4. 12 Mar, 2013 5 commits
  5. 11 Mar, 2013 8 commits
  6. 10 Mar, 2013 4 commits
  7. 08 Mar, 2013 14 commits
    • David Gobbi's avatar
      Merge topic 'test-projected-hexahedra-size' into master · 274cb5e9
      David Gobbi authored
      79feb9dd Reduce TestProjectedHexahedra window size to 500x500.
      274cb5e9
    • Sean McBride's avatar
      Eliminated (undefined) negation of INT_MIN · 38f76b82
      Sean McBride authored
      clang ubsan warned running vtkRenderingAnnotationCxx-TestCubeAxes2DMode:
      
      Rendering/Annotation/vtkCubeAxesActor.cxx:1315:30: runtime error: negation of -2147483648 cannot be represented in type 'int'
      
      This happens when 'range' is exactly 0.0, and so log10() gives -infinity,
      which when cast to int (apparently) becomes INT_MIN which cannot
      be negated due to the asymmetry of two's compliment representation.
      
      Now, check log10() result with newly-created vtkMath::IsFinite API.
      If the result is not finite, return 0 from vtkCubeAxesActor::Digits(),
      which I'm not sure is the correct thing to return.
      
      Change-Id: I36fc832df98ccb96c1fdc951c42e7e2c6de9d4b4
      38f76b82
    • Sean McBride's avatar
      Changed computations to uint64 to prevent signed overflow · 07cd3906
      Sean McBride authored
      With clang's -fsanitize=signed-integer-overflow, the test
      'vtkFiltersGeneralCxx-TestYoungsMaterialInterface' gives:
      
      IO/Geometry/vtkAVSucdReader.cxx:302:34: runtime error: signed integer overflow: 16 * 504823808 cannot be represented in type 'int'
      IO/Geometry/vtkAVSucdReader.cxx:302:60: runtime error: signed integer overflow: 4 * 2019295232 cannot be represented in type 'int'
      IO/Geometry/vtkAVSucdReader.cxx:303:35: runtime error: signed integer overflow: 12 * -1340538880 cannot be represented in type 'int'
      IO/Geometry/vtkAVSucdReader.cxx:313:35: runtime error: signed integer overflow: 369098752 * 2019295248 cannot be represented in type 'int'
      
      This code tries to guess endianness, and when it's wrong it
      ends up multiplying large numbers, resulting in signed overflow.
      Fix is to do computations as uint64, which may still overflow,
      but at least unsigned overflow is defined to wrap.
      
      Did other misc cleanup too. Notably, the code could go into an
      infinite loop trying big then little endianness over and over
      again if a file is corrupt.
      
      Change-Id: I2b7d84dae275386dcd87fa496b7916945284bc30
      07cd3906
    • Sean McBride's avatar
      Initialize ivars to 0 to prevent access of uninitialized data · 743dc8aa
      Sean McBride authored
      With -fsanitize=signed-integer-overflow, the test
      'vtkIOXMLPython-TestEmptyXML' gives:
      
      IO/XML/vtkXMLPStructuredDataReader.cxx:62:35: runtime error: signed integer overflow: 591396864 * -536868871 cannot be represented in type 'int'
      
      The reason being that 'PointDimensions' was never set and so contained
      random values that happened to be pretty big. There probably should
      be some better error checking somewhere, but it seems a good
      defence to zero these things anyway, in case a client of the class
      (incorrectly) invokes a method that accesses the uninitialized data,
      at least we'll get consistent results.
      
      Change-Id: I77f5ac68d4edad632931667ddf5e87f930f48b3f
      743dc8aa
    • Sean McBride's avatar
      Eliminated (undefined) signed integer overflow · e29cc816
      Sean McBride authored
      clang ubsan warned running some of the 'EnSightBlow' tests:
      
      IO/EnSight/vtkEnSight6BinaryReader.cxx:2823:15: runtime error: signed integer overflow: -1358823424 * 4 cannot be represented in type 'int'
      
      I just upcast to 64 bit so that the multiplication doesn't overflow.
      The tests now pass with the ubsan check.
      
      I have subtly changed the behaviour here, but believe the
      intent is preserved.
      
      Change-Id: I0a35789c169204f04d2a683a029e992e9b20ca65
      e29cc816
    • Berk Geveci's avatar
      Merge topic 'KdTree_roundoff' into master · 2fc4a23d
      Berk Geveci authored
      9d80bf90 Increasing the bounds just a bit to avoid round-off issues.
      2fc4a23d
    • Berk Geveci's avatar
      Merge topic 'flowpath_filters_check_arrays' into master · 8710fd10
      Berk Geveci authored
      eab902ac Adding in warning if the input vtkPointData structures is different.
      8710fd10
    • David C. Lonie's avatar
      Replace the power-of-two scaling with a clearer implementation. · 40fe9ec0
      David C. Lonie authored
      This implementation is easier to read and performed using only integer
      operations.
      
      Change-Id: I97ea7e73e6bbaba14b7dc29d26e1db55e57136e4
      40fe9ec0
    • David C. Lonie's avatar
      Rewrite the PrepareImageData methods to check extents, not dims. · 41e61435
      David C. Lonie authored
      The new rendering code using the extent coordinates to control where
      drawing occurs. Simply checking the dimensions of the image is not
      sufficient.
      
      Change-Id: Ib8185affafdc345b760f6c2164584f8379982a13
      41e61435
    • David Gobbi's avatar
      Reduce TestProjectedHexahedra window size to 500x500. · 79feb9dd
      David Gobbi authored
      The previous window size of 800x800 seemed needlessly large, and causes
      test failures on screens with a height of 800 or less (e.g. some laptops).
      
      Change-Id: Id6abd1bd773f0a074ebedb8eb0f699a7c8a665b5
      79feb9dd
    • Bill Lorensen's avatar
      Merge topic 'reduce-test-checkboard-memory-usage' into master · a7d62db3
      Bill Lorensen authored
      96ae86c7 Reduce memory usage for TestCheckerboard
      a7d62db3
    • Bill Lorensen's avatar
      Merge topic 'reduce-tcl_test-checkboard-memory-usage' into master · 1e993fc7
      Bill Lorensen authored
      c250dadc COMP: Reduce memory usage for TestCheckerboard
      1e993fc7
    • Bill Lorensen's avatar
      COMP: Reduce memory usage for TestCheckerboard · c250dadc
      Bill Lorensen authored
      This reduces the memory usage for TestCheckerboard.tcl by reducing the
      Z extents of the images from 99 to 10. This reduces the memory usage
      of the test from ~170MB to ~37MB.
      
      Change-Id: I074012d39f017f03463c240e96635820c8abdbae
      c250dadc
    • Kyle Lutz's avatar
      Reduce memory usage for TestCheckerboard · 96ae86c7
      Kyle Lutz authored
      This reduces the memory usage for TestCheckerboard by reducing the
      Z extents of the images from 99 to 10. This reduces the memory usage
      of the test from ~170MB to ~37MB.
      
      This should fix the test on AIX where it was running out of resources.
      
      Change-Id: I3513406e35a88793eff96d36f07c8497e15401f4
      96ae86c7
  8. 07 Mar, 2013 2 commits