There are additional buglists which may contain important bugs and
unsupported features for Coin. For the Coin satellite libraries,
extensions and example programs:

        simage/BUGS.txt
        SoQt/BUGS.txt
        So[Qt|Xt|Win]/src/Inventor/Qt/common/BUGS.txt
        SoWin/BUGS.txt
        SoXt/BUGS.txt
        SIMVoleon/BUGS.txt

..and buglists less likely to contain important bugs:

        SoGuiExamples/BUGS.txt
        SmallChange/BUGS.txt
        SoGtk/BUGS.txt
        Cute/BUGS.txt


Also, a number of bugs deemed to be obsolete due to age or obscurity
of environment they can be reproduced in, has been moved out of this
file. So also for bugs of little consequence, or for other reasons of
being of very low priority. They have been moved into catalogs under:

        Coin/docs/bugs/


Now, for the list of (possibly) significant bugs in the core Coin
library:

=====================================================================

005 Free memory read on connected SoSFEnum fields?

    This is an old problem which might have been fixed now, but we
    should check it with Valgrind: Purify used to report "Free memory
    read" when deleting SoSFEnum fields that were connected to other
    fields.

    <mortene@sim.no> 20020521.

=====================================================================

031 Setting up the alpha channel for the offscreen renderer doesn't
    consistently (between hosts) produce a transparent background when
    using setComponents(SoOffscreenRenderer::RGB_TRANSPARENCY).  It
    works at home on my Debian Linux w/ GeForce4MX440 with the 3123
    NVIDIA GLX sources, but not at work with whatever gfx system I
    have there.

    Index: src/misc/SoOffscreenGLXData.icc
    ===================================================================
    RCS file: /export/cvsroot/Coin/src/misc/SoOffscreenGLXData.icc,v
    retrieving revision 1.17
    diff -u -r1.17 SoOffscreenGLXData.icc
    --- src/misc/SoOffscreenGLXData.icc     19 Sep 2002 10:08:04 -0000      1.17
    +++ src/misc/SoOffscreenGLXData.icc     27 Sep 2002 08:50:02 -0000
    @@ -199,6 +199,8 @@
         attrs[pos++] = 4;
         // FIXME: won't get an alpha channel in the context unless we also
         // request a particular ALPHA bitsize. 20020605 mortene.
    +    attrs[pos++] = GLX_ALPHA_SIZE;
    +    attrs[pos++] = 4;
       }
       attrs[pos++] = None;
       return pos;

    20020927 larsa.

    UPDATE 20020927 mortene: this could simply be caused by the GLX /
    X11 drivers lacking support for any alpha-enabled visuals
    (according to spec, it is not guaranteed that any such visual is
    available). Anyway, the right way to solve this problem is to 1)
    write our own weight-function for X11 visuals (like it's been done
    for MSWin visuals in SoWin's SoWinGLWidget.cpp), for better
    control of the selection process. Then 2) warn the user if we
    can't get hold of any visual capable of all requested features.

    UPDATE 20021022 larsa: this was the case - old XFree3 with old
    Mesa does not support alpha for offscreen rendering.

    UPDATE 20021121 larsa: The offscreen renderer GL context creation
    is nevertheless bogus (with GLX at least), and should be fixed so
    the best possible context is created.

    UPDATE 20060203 mortene: a *lot* has been improved about offscreen
    rendering contexts since the last update on this bug, but the
    selection process among the available contexts has still not been
    fixed, nor exposed to the user in the form of any public API.

=====================================================================

045 SoCenterballManip rotation off-center.

    When dragging the 'stripes' on a CenterballManip node, rotation
    happen around an off-center axis. (Off CenterballManip center)
    This happen for rotations around all the axis.

    NOTE: The bug is not visible if there is no connected geometry, or
    if the geometry is 'simple', like one cube or one cone.

    A simple scene-graph to reproduce:

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --
    #Inventor V2.1 ascii

    Separator{
        CenterballManip {}
       Array{
              numElements1 2
              numElements2 2
              numElements3 2
                separation1  2.0 0.0 0.0
                separation2  0.0 2.0 0.0
                separation3  0.0 0.0 2.0
                      Cone {}
       }
    }

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --

    20021209 rolvs

    UPDATE 20030408 mortene: there is also a second bug here; the
    scene bounding box is not calculated correctly, as can be seen by
    loading this into e.g. the ExaminerViewer. The near clip plane
    will clip part of the manip's geometry, something which should
    never happen.

=====================================================================

047 SoAntiSquish seems to trigger a bug in our matrix factoring code.

    From a coin-discuss posting by Philippe Debann:

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --

    I am planning to design my own Dragger class and have been looking
    at the Toolmaker Ch. 8 examples: TranslateRadialDragger and
    RotTransDragger.  I found an unexpected behaviour of
    RotTransDragger (top-left of scene): When dragging one of the
    SoRotateCylindricalDraggers, the dragger will switch from rotation
    to scaling up as the mouse position approaches and goes beyond the
    bounding cylinder of the (spherical) dragger.  In fact, the
    dragger should not do any scaling up/down!! This happens
    systematically when dragging the horizontal band left or right
    (Y-axis rotation).

    When I comment out the SoAntiSquish part of RotTransDragger, this
    unexpected behaviour dissapears (But obviously you want the
    AntiSquish's anti-squishing behaviour!).  Alternately, if I set
    its 'recalcAlways' field to FALSE, I get a different strange
    behaviour: manipulation of the RotTransDragger affects the
    rendering of the RotTransManip to the right of it: its
    antisquishing scaling factor and fluctuations of its wire-mesh
    sphere.  All this seems to be independant of the AntiSquish
    'sizing' field, e.g.  BIGGEST_DIMENSION or LONGEST_DIAGONAL.

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --

    pederb has investigated, and found that this seems likely to be a
    problem with our SbMatrix code that factors the translation,
    scaling and rotation out of a matrix.

    20021218 mortene.

=====================================================================

051 ExaminerViewer's automatic initial "view all" doesn't work on
    a scene graph with SoImage.

    Reproduce with the following source code snippet. The
    "aass_logo.rgb" is an image of dimensions 80x68 (I haven't
    investigated if this matters or not).

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --
    #include <Inventor/Qt/SoQt.h>
    #include <Inventor/Qt/viewers/SoQtExaminerViewer.h>
    #include <Inventor/nodes/SoSeparator.h>
    #include <Inventor/nodes/SoImage.h>
    #include <Inventor/nodes/SoCone.h>

    int
    main(int argc, char* argv[])
    {
      QWidget * mainwin = SoQt::init(argv[0]);

      SoSeparator * root = new SoSeparator;
      root->ref();

      root->addChild(new SoCone);

      SoImage * img = new SoImage;
      root->addChild(img);
      img->filename = "aass_logo.rgb";

      SoQtExaminerViewer * viewer = new SoQtExaminerViewer(mainwin);
      viewer->setSceneGraph(root);

      SoQt::show(mainwin);
      SoQt::mainLoop();

      delete viewer;
      root->unref();

      return 0;
    }
    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --

    Notice how the cone is a tiny dot in the lower left corner of the
    image at first. Then click view all (the eye icon) and see how the
    camera is then positioned where it should be. Correct behavior
    would be for the camera to be positioned there from the start.

    20030103 mortene.

    WONTFIX? 20030225 rolvs: In this example, setSceneGraph calls
    viewAll() before the window is 'show'-ed. The viewport region
    passed to (*SoCamera)->viewAll( graph, vpreg ) is therefore wrong,
    causing a malformed boundingbox for the image and thus a malformed
    view.

    UPDATE 20030225 mortene: is there any simple and clean way to
    detect and handle this with a) a warning message for the app
    programmer, b) using a decent "dummy" / default region?

=====================================================================

054 SoComputeBoundingBox engine probably fails with viewport-dependent
    geometry.

    The SoGuiExamples/engines/computexfbbox example fails with a small
    scene with an SoText2 in it:

    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --
    #Inventor V2.1 ascii

    Cube { }
    Translation { translation 10 0 0 }
    Text2 { string [ "hepp", "tjo-bing", "hupp" ] }
    ----8<---- [snip] ---------8<---- [snip] ---------8<---- [snip] --

    ..and pederb tracked this down to a failure of the engine's
    evaluate() to be called again when the camera position changes.

    Note that this bug is probably rather hard to fix properly.

    20030128 mortene.

=====================================================================

060 SoToVRML2Action failing with VRML97 nodes.

    Reported by kintel:

      "SoToVRML2Action fails [with this file]:

      ----8<---- [snip] ---------8<---- [snip] ---------8<----
      #VRML V2.0 utf8

      Transform {
        children Shape {
          geometry IndexedFaceSet {
            coord Coordinate { point [ 0 0 0, 1 0 0, 1 1 0 ] }
            coordIndex [ 0, 1, 2, -1 ]
          }
        }
      }
      ----8<---- [snip] ---------8<---- [snip] ---------8<----

      ...and outputs an empty file:

      ----8<---- [snip] ---------8<---- [snip] ---------8<----
      #VRML V2.0 utf8

      Group {
        children
              Group {
              }
      }
      ----8<---- [snip] ---------8<---- [snip] ---------8<----

      This is a special case as it is already a VRML2 file. However, I
      guess the action will also fail given a mixed scene graph as
      input. This is a bigger issue since we often have situations
      with mixed scene graphs and want to convert them to VRML2."

    We've also had at least one external report from a user who found
    this to be a real problem. From the problem report:

      After some tests I've found that the problem seems to be that
      VRML shapes instead of being copied through by the ToVRML2
      action (they are obviously already VRML2), they are deleted
      (ignored?). It looks like that src/actions/SoToVRML2Action.cpp
      lacks one more callback to handle nodes already in VRML2 format.

    This is a medium priority item.

    20030307 mortene.

=====================================================================

066 Viewer "view all" functionality not 100% correct.

    See the VRML1 file below containing 5 line segments as separate
    linesets.  Using this file, the viewAll button in the examiner
    viewer only _almost_ zooms out the camera enough: parts of the
    model is outside the view.

    To reproduce:

    o Open model in an examiner viewer.
    o Resize the window so height ~= 2* width. (at least.)
    o Rotate the model so you get minimal total height in the view.
    o Press viewall. If my problems doesn't show, rotate around the up
      axis and press viewall repeatedly until error occurs.

    ----8<--- [snip] ---------8<--- [snip] ---------8<--- [snip] ---
    #VRML V1.0 ascii

    Separator {
      Separator {
        Coordinate3 {
          point [ 372031 7.20114e+06    32.4351, 372031 7.20114e+06   -465.843 ]
        }
        IndexedLineSet {
          coordIndex [ 0, 1, -1 ]
        }
      }
      Separator {
        Coordinate3 {
          point [ 386603 7.19641e+06    32.1387, 386603 7.19641e+06   -348.644 ]
        }
        IndexedLineSet {
          coordIndex [ 0, 1, -1 ]
        }
      }
      Separator {
        Coordinate3 {
          point [ 384767 7.19186e+06    25.2236, 384767 7.19186e+06   -285.863 ]
        }
        IndexedLineSet {
          coordIndex [ 0, 1, -1 ]
        }
      }
      Separator {
        Coordinate3 {
          point [ 377824 7.20872e+06    32.4382, 377824 7.20872e+06   -329.9802 ]
        }
        IndexedLineSet {
          coordIndex [ 0, 1, -1 ]
        }
      }
      Separator {
        Coordinate3 {
          point [ 383236 7.18857e+06    40.1396, 383236 7.18857e+06   -371.016 ]
        }
        IndexedLineSet {
          coordIndex [ 0, 1, -1 ]
        }
      }
    }
    ----8<--- [snip] ---------8<--- [snip] ---------8<--- [snip] ---

    I tried the model in the SoGuiExamples/engines/computexfbbox
    example, and the bounding box gets calculated correctly. I guess
    that means there's a bug in either SoCamera::viewAll() or in
    SoGuiViewer::viewAll() -- most likely the latter.

    20030507 mortene, reported by kintel.

=====================================================================

072 Assert failure when writing extension nodekit in binary format.

    Running the example program below causes the following assert to
    hit:

      test: /home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:248: void set_current_writeref(const SoBase *, int): Assertion `rc >= 0 && "buggy writerefcounter"' failed.
      Aborted

    This points at some very nasty error, where some part of the scene
    graph is not tagged for writing on the first SoWriteAction pass,
    but is still attempted written on the second pass.

    It's really lame that we have this kind of bug at all, as having
    the two passes traverse the same structures shouldn't be rocket
    science.

    When writing as ASCII, the assert does *not* hit -- which means
    traversal is different somewhere for binary and ascii export.

    --------------8<----------- [snip] ------------------8<--------
    #include <Inventor/actions/SoWriteAction.h>
    #include <Inventor/nodekits/SoSubKit.h>
    #include <Inventor/nodekits/SoNodeKit.h>
    #include <Inventor/nodekits/SoBaseKit.h>
    #include <Inventor/nodes/SoGroup.h>

    class SoTestKit : public SoBaseKit
    {
      SO_KIT_HEADER(SoTestKit);
      SO_KIT_CATALOG_ENTRY_HEADER(A);

     public:
      static void initClass();
      SoTestKit();
    };



    SO_KIT_SOURCE(SoTestKit);

    void
    SoTestKit::initClass()
    {
       SO_KIT_INIT_CLASS(SoTestKit, SoBaseKit, "BaseKit");
    };

    SoTestKit::SoTestKit()
    {
       SO_KIT_CONSTRUCTOR(SoTestKit);

       SO_KIT_ADD_CATALOG_ENTRY(A, SoGroup, FALSE, this, \x0, TRUE );

       SO_KIT_INIT_INSTANCE();
    };

    int
    main(int argc, char ** argv)
    {
      SoDB::init();
      SoNodeKit::init();
      SoTestKit::initClass();

      SoTestKit * testkit = new SoTestKit();
      testkit->ref();

      char * buf = (char*)malloc(1024);
      SoOutput out;
      out.setBuffer(buf, 1024, realloc);
      out.setBinary(TRUE);

      SoWriteAction wa(&out);
      wa.apply(testkit);

      testkit->unref();

      return 0;
    }
    --------------8<----------- [snip] ------------------8<--------

    Problem reported by Tamer Fahmy.

    20030602 mortene.

    UPDATE 20050610 mortene: it seems someone (pederb?) rewrote the
    assert, and made the code "kinder", as there are now warnings
    instead:

        Coin warning in SoWriterefCounter::setWriteref(): writeref < 0 for <noname> <0x806d6a8>
        Coin warning in SoWriterefCounter::<cleanup>(): Not removed from writedict: 0x806d6a8, Group:<noname>

=====================================================================

073 Multiple complexity settings for same SoTexture2 does not work.

    Using the same SoTexture2 nodes with different
    SoComplexity::textureQuality settings is something which works for
    SGI Inventor, so it should be considered a Coin bug.

    Reproducible problem, example file:

    --------------8<----------- [snip] ------------------8<--------
    #Inventor V2.1 ascii

    Separator {
       Separator {
          Complexity { textureQuality 0.1 }

          DEF mytex Texture2 {
             image 2 2 3
             0xffffff 0x00ff00   # white   green
             0xff0000 0xffff00   #  red    yellow
          }

          Cube { }
       }

       Separator {
          Complexity { textureQuality 1.0 }
          USE mytex
          Translation { translation 5 0 0 }
          Cube { }
       }
    }
    --------------8<----------- [snip] ------------------8<--------

    pederb's comments on the bug:

        We handled this earlier, but when adding proper support for
        auto-caching, it was easier to store everything for a texture
        object inside the texture object, and then identify it with a
        single uint32. Then it's quite easy to figure out the current
        texture object after a display list has been executed (the
        render-cache must store the OpenGL state changes it does so
        that SoGLLazyElement is updated after the display list has
        been executed). I could add additional variables for storing
        the filter settings, but I just couldn't see any practical
        reason for doing this.


    20030916 mortene.

    UPDATE 20040714 mortene: an external Coin user (Nigel Murry) had
    an ingenious way to use a texture multiple times with varying
    complexity settings, which now fails -- so there are indeed
    practical uses for this. So we should really consider this a bug,
    and a fairly high priority item aswell, as this is a regression
    versus older Coin versions.

    From Murry's bug report on coin-discuss:

        I have been using textures to show how some parameter (like
        saytemperature) varies over the surface of an
        object. Sometimes I would like to see distinct color bands and
        other times I would like to see a continuous variation of
        color. So I have a SoTexture2 defined using 8 or 16 color
        levels. Then I change bewteen a "banded" and a continuous
        display using a SoComplexity node with textureQuality set to
        0.00001 for a banded display.

        That normally works fine, but when the texture is used more
        than once there appears to be some confusion as to how the
        textureQuality is used. In the attached little iv file, if the
        switch at line 7 is off (whichChild=-1) then the second
        (legend) use is correct. If the switch at line 7 is on
        (whichChild=-3) then the second use shows a continuous
        texture.

=====================================================================

074 SoComplexity::textureQuality changes must be made before
    SoTexture2 traversal.

    In the "Inventor spirit" of appearance settings not having any
    effect until hitting a shape node, it should be possible to do
    this:

    --------------8<----------- [snip] ------------------8<--------
    #Inventor V2.1 ascii

    Separator {
      Texture2 {
         image 2 2 3
         0xffffff 0x00ff00   # white   green
         0xff0000 0xffff00   #  red    yellow
      }

      Complexity { textureQuality 0 }

      Cube { }
    }
    --------------8<----------- [snip] ------------------8<--------

    ..but the SoComplexity node will have no effect in this case. I
    feel it should.

    Note that SGI Inventor doesn't handle this either, but I still
    think it is a weakness in Coin which we should fix.

    20030916 mortene.

=====================================================================

075 SoHandleBoxManip does not behave correctly when locking a
    translation axis.

    To reproduce, load this simple iv-file into an examinerviewer:

    --------------8<----------- [snip] ------------------8<--------
    #Inventor V2.1 ascii

    HandleBoxManip { }
    --------------8<----------- [snip] ------------------8<--------

    ..then hold down SHIFT while LMB-clicking on one of the sides (for
    translation). The translation will correctly be locked to one
    axis, but when releasing the SHIFT button, the manip will jump to
    the current cursor location. This last behavior is incorrect.

    20030922 mortene.

=====================================================================

076 SoHandleBoxManip does not behave well upon "deep Z translation".

    To reproduce, load this simple iv-file into an examinerviewer:

    --------------8<----------- [snip] ------------------8<--------
    #Inventor V2.1 ascii

    HandleBoxManip { }
    --------------8<----------- [snip] ------------------8<--------

    ..then move the box out to the left side of the view
    (i.e. translate in -X), before LMB-clicking the right side of the
    box to translate in the Z direction. The manip will correctly move
    in the Z direction when moving the mouse cursor to the right, but
    upon continuing moving the mouse cursor to the right it suddenly
    misbehaves by first jumping back close to the camera, and then
    starting to move in the *negative* Z direction.

    This is clearly bad handling of some kind of border case.

    20030922 mortene.

    UPDATE 20060208 mortene: this may be a bit hard to reproduce. To
    make sure to trigger it, one can first move it way back as
    described above, then do "view all", and then go ahead with the
    above description again.

=====================================================================

090 Scene graph copy behavior deviates from SGI Inventor.

    For the code example below, we get different behavior on
    connection copying between Coin and SGI Inventor. For Coin, the
    node with the slave field is not copied (the connection is just
    shortcut), while it is copied for SGI Inventor.

-----8<------ [snip] ------------8<------ [snip] -------
#include <assert.h>

#include <Inventor/SoDB.h>
#include <Inventor/nodes/SoRotor.h>
#include <Inventor/SoLists.h>

int
main(void)
{
  SoDB::init();

  SoRotor * masterrotor = new SoRotor;
  SoRotor * slaverotor = new SoRotor;

  SoRotor * rotor = new SoRotor;

  rotor->rotation.connectFrom(&masterrotor->rotation);
  slaverotor->speed.connectFrom(&rotor->speed);

  printf("rotor==%p, masterrotor==%p, slaverotor==%p\n",
         rotor, masterrotor, slaverotor);


  SoRotor * copiedrotor = (SoRotor *)rotor->copy(TRUE);

  SoField * f;
  SbBool ok = copiedrotor->rotation.getConnectedField(f);
  assert(ok);
  SoRotor * copiedmaster = (SoRotor *)f->getContainer();

  SoFieldList slavelist;
  int nr = copiedrotor->speed.getForwardConnections(slavelist);
  assert(nr==0 || nr==1);
  SoRotor * copiedslave = nr ? (SoRotor *)slavelist[0]->getContainer() : NULL;

  printf("copiedrotor==%p, copiedmaster==%p, copiedslave==%p\n",
         copiedrotor, copiedmaster, copiedslave);

  return 0;
}
-----8<------ [snip] ------------8<------ [snip] -------

    A test run:

    Coin:
        rotor==0x807d3c0, masterrotor==0x807c740, slaverotor==0x807ce18
        copiedrotor==0x807e248, copiedmaster==0x807c740, copiedslave==(nil)

    SGI Inventor:
        rotor==0x8058710, masterrotor==0x8051690, slaverotor==0x8057ae0
        copiedrotor==0x8059cb8, copiedmaster==0x8051690, copiedslave==0x8059df0

    (Notice the differences for 'copiedslave'.)

    This is a quite nasty bug, which can cause hard to debug errors
    when moving code from SGI Inventor to Coin.

    20031121 mortene.

=====================================================================

092 SoToVRMLAction outputs empty scene graph when given a Coin/Inventor
    scene graph.

    SoToVRMLAction only supports VRML2 scene graphs, not Coin/Inventor
    scene graphs.

    20031126 kintel.

=====================================================================

095 SoIndexedTriangleStripSet renders much slower with PER_FACE normals.

    According to pederb, this is not expected, so something fishy is
    going on.

    This is seen on a Linux NVidia box with v41.91 driver and
    GeForce2.

    Here's a model which reproduces the problem. Just load it into the
    examinerviewer:

-----8<------ [snip] ------------8<------ [snip] -------
begin 644 makrell.iv.gz
M'XL(",G[S3\"`VUA:W)E;&PN:78`Q%U=CYVW<;[7KSA`;EK`..!P2`ZG=ZX@
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MY.<W;^6O;@[RAXW(_<?3Z=WA/Z\_O7_$^?;5[[[_ZLN7KPXWINA>N=\+A``?
MWIU^N/K\X=/A'^Y.]_)?11"XJ\/[SW\^'3Z>[G[07Q$ECV`?;N_O__%X^/+M
MV]L[X_OI5N1$^9^^$$!A>?_^]O.'=Z+KT^%/)]/Q]NK^='P`>`1Z<SK=&,_#
MZS]<O[N^.OSV^N;S7P[O[JZ%WJ'`D>&+PZUR^9?35[=W;T]YCY%30@#@PX^W
MTALWCUI^NOKP^?38[D=#__]N=<&7M]:*JT\G?/R9_//Q5OKZ\!^'7N&8,B8Z
M8#NFCE#ID(^Y%LB=O_@_>1-$ZKV:8)8_5A&LK1/`S^72D7OOZ'+04E,Y;*7G
M_A0PK+E`@>Z"A1NH(%4&S+\X12SRVP>LQTZML<B1_(.I[@`K<76YBLD$L1?J
M95YSJ"G+&0J@$&LJV)B@JB`GKKWT'2)B*@^"U$6PMY8;E7G5H;:LIRB(J;KF
MBKUU0Z0D/;4?.="Z"Y8$J$.1<^7.;5YUK#'K.<(Q]V;](DR3_L8Q)4J)G^)U
M[J6X''70,2N3D'KJTXJ#35G-4/`:9U;!5*0W2`&!A,P>L<,F6#/W;`QE=7JZ
MY(RHCC1E/4/0196%83D*/P!40:ZRT.[&C0S039!3)FN*0,OJ,Z\ZUICU'`6Q
M5!6K.:/W"HJU\VYMZKJJFB`6+#9-F1JF/*\XUI35#!4/P!"QRSK23%!&6JM[
MQ-R-(4H34/?=7I'P`L6!AJSF)WAH,ZK(*,/<BHJ5FNE<0U+.)DBRT>IP[9"P
MEFG%L8:L9ZB(K7<5!)1]0P>7^`:-ZKXIXE-8YT%ALN$J>V_I9Y:1N.I08]9S
M5,2*HAK5D\"J'&6OS-SVC2E-%V,1I);0!!LGXGZ!ZE!CUG-\-#C*<IME<U3!
M+`OJF6$K"PX;HJRNK(*520;P):I#C5G/42=7!S)$"07,;ZH@NT>A_<"5_YF@
M[+>LD[5TH5OF-<?:LIRB;I.0T0$S%)U:*'(R"W>`9"N/`F)JJEF6IR0!Q+3F
M6%/64P0-X3*8(`,GG5FY]\9\QM$I9!P9<FOHT1\E+O.J8XU9SU$]7U#$HAX1
M%?5\`:DD.N/$)Y=KK623$V\CI3*M.=B6Y10?(QQ90SNZMR\>5-XW17X9##!+
MS(O5Y,2;ZM-Z8PU935#PL&?#DVBWR.X*1RX->;\-Z8_=D\B4A*$*%@FBYC6'
M6K*>H2#6[)J+S#T4.0GG$/+.C^:6W%TLA;+LP2J8.]&\XEA35C/4#(W$MBK7
M"A"I''66]72?"A`]YE:VVG0.BB"#+-#SFD,M6<]0$)E-L9"2T:=RB8KXR+O4
M7FJ535!VW6*`LC:?V9WCFF-M64X1CK)'5-(@5EQ`^;\("E,9G&?RF38215!"
MX6YR)&%OF=8<;<IRA@59O=YZ+`DELE?!D@KNW6/U%2RK(2$1;(*%:J=IS=&F
MK*=82F.C6#5R5SEJLMCNNT]\8TMJU"Y_<$'Y^7[(QC5'V[*<8BEL>3BB4HHJ
M)I)(ELX`ML0N*-Y%4T&)B]*9?HYKCC5E.<,B`]`2=BRN+B@BYRX[S'[@U$1]
M$R3;@("KL+E`<ZPIRQD6F4%LM836P%(T4*&?\7<3%O+J1"N,Q3PFW4SKM.9H
M4]93++F`"]*6OI)]0X+\??]AKJY:0ESS$BLC(UZ@.=B6Y11+3IM<AV*5%HEO
MZIE\:\H,7GJ3D+AJ,"#])S^L\YJ#35E/4;W"K8Q8*UI@4\49VL?M*=>&6[VQ
MNVH)#V7'F5<=;<QZCN&::+C*.J`ZV)@8Q](E@).`5%:)7#-:YD7&6NWGQ$K?
MY(KU2^:>VA.&+MC`!<7;<,!&]*3-+@@/B,S*4`));F<%@PSCBK,+%JB6@A`F
M[<FF,<APO>HP(F^"8A/K9W'-Q17X&ZH;9],LP$]"M4&*RS5'`2W9I(*ZH9D@
M[O+F/]?<N553C96PSU,<T.Q#I\F4-D`M/5Y`<+7B,!Y@V?!2;R8H\3'_C<G2
M,'E+2N5+"*[6&P84E]D%9<VS19NAR%;\5]>E1F!R6EIGRA<P7*TX#`A$+BA#
MP.0D.F[MW";@@`38FFM&[)<P7*TX"IB3KS4RWXNM-9V3^!D7:`XCYNQ3C]#S
MB;+40*?ZU_<`JH5\^ZFR[U_$<;GJ,&+V;S4R4?,%IQ,__0C#51.[H,S/LDT7
M>))4'^2X7'48<;..++*P62>W<TUA'[42DY)W(#;*[1*&BQ4'\1[$L*9M#F#%
M,W#B!;M@:9MG!1*>X@*"?U=S\Z[KI(DA\R:S!),74(QJSFEK2M<,LVH6'Q[P
M`HK+-8<!M_'?6RX>#(`PQ$LTAQ&!-^,P;8X+=[I(=1CQ0;!#VOQ3<13K"M5_
M%Q'P09`=L5"KEZD>1RSN-U4M@/T\XA1!*F3EA"I+DF:HCN)+-W&R]H(ULPO6
M3-D$BVP?=$8PU0=$_9,(-BX)SJC>Y"KJ,#I6$M?G:1EJ2'.X+06["[;63%"\
M]]TWQ6.J@VV):^Y;6XC($!M(N^`2B@^JN>GR+*K%J6[Y`M7AQH15-R]`5TCR
M]RI8>?>=V"#%[(#`:(`BCPSSFL-->1`$Y.::N["\2'.P*7'-==-LJ\512VJ4
MZ7+;0$/-_(IBIO3T(Z@1Q>&6Q#7WS8A%_MI4-_DA74!Q^Y931J+A==D!S\V^
ML.)P4Z*:M4)F@&3Y2M%<),2[Q#:E.<7\L.3(;H1/G*PQU=&VQ#4W_Q0W0^FV
M0%3Q?!$NH1A5W5/9!*GZT)&N[/T"CH5]B<^:%-XT,U^@.=R6L&;_**&*HP_9
M-5?F2XSSL*UI@@M=->*Y12>L.MJ6N.8"FV#V42NM*CE?0+$VEY-YYQ03I-XN
MT!QNR[#F9M]DZ'3I.>$%%*T8K((,R5TGPG,+;5ASM"F-7+!`P>H^6T_GO)>P
MYFA3].,ZTYRS`[:F7TM?HCG:EKAJ_\I=VF*_(JJ[N->76">LFOK6EEZM`QO6
M6NL%%..:=<<UP>2[?BOP-(8?HAC7W&`3A-I<,[>4+Z$85EWKHZ`'$)7.>O$A
MC@.55D[Z-9Y^[,#)X]X&/7''V;.=8<`X16S;\=.:T(Y8H@8?E6<I=L'SLYWB
MYX$"=CTG@7WV;&<<<+41PPQ+K@;86O%"G3CS5'*?/MD91UQNQ##%7+ME8"0X
M1@M193GFMC\'$S[:&4=<;\8PQT0/E04@5H=#O_UI^Z_9@H<[PWCKK1ADJ!4$
MRX$FT+A!Y4C"I]RFCW;&$1?;,,Y05D#;_V2SK4VW#/V\+>T_+0X?[8PCKK=B
MF*-^?FL?D7<N5J;DKH?6Z^3QSCC>>BM&&69/@J#$IVP,"VBP.GW`<P!QL0WC
M#!.8WY3U@R63DQ6%I@]W1O'66S#,L';/0R3FEKOO?0DJ3A_M'$!<;,4X0\LX
MX9$US:2*97'`GFGZ8.<`XGHK1CDV2'[>KY!_L*@A`9TY^A8]V3F`N-Z.48[B
M"/B1S:1G*-3O+>H*SY_MC".NMV.88R]^8E-]`CO/FC4ATJ>/=H8!UULQ3)$?
M#O%*9)KMNX.<,I\9.]&CG7'$Y6:,4NQ9MP%=)B"1C3'9-2KW^=.=8<3U9HQS
M+'6[CH#0(I>DR=6]=Q<]WQG&6V[$,$/Q='P+)+%RT9,AE"0RGS[>&09<;<(P
M05DAW-5A`=0C3RPQ6]_/J>CQSCC@<AL&&4K0XD9L28]JJB#(]@\X?;HSCKC<
MB&&*>:ML=]GI6563;!S[33)ZNC.,M]Z&489^5EN"4ZZ=[2Q?UH]7I@]W1@%7
MFS!.L!5+26F1P,YK-BA\YOA;]&!G%&^U`>,$R7+D>K4;0-$E1#9&V-]R%CW6
M&<9;;<$P/T3+)1+:^482]QX`9H]TAN$66R]*KVF^5'/D4,&/:589./N\0_@\
MYP#B8@M&"5H2JAT3BD.C%1/HK$>ZI@]SA@'7FS!.D5)VCOIUJ1U7U%3Y_,'4
M`<359HQ3Q.(42ZN\G7^LN-\6PX<Y!Q#7FS'*$;GVK=#7[02=GC3*^V@T?)@S
MCKC>CG&.P?+F^H+I>CL&$PK"B)-?ZR9K(AC'0C+;\ERL/H`7)N@^M@'*"-=T
M4$IG+I*,$A0/H'ERHOIL04)5/1FG#P"NMF`L3S9@P>5=$F;8"-PV2)Y)4!\%
M&"<SHB.(S\D1*VZWKR5;X8ESZU#GAR(F<D3Q:/U8%W"G?7`6S,@,(*Z?+K&4
M_X@9UW=,G&,I5IN35=M3HGKU7MK?SA7OF&I)5BV=)%\EQ(4'W!<18JFH$<#U
M8R=2T1HPXOIN"3.L_C%<!G1'56)ID9NM7`XA/AM#\LP19J[V*`&SGNBLTP.Q
M4;>!F.7O&UF.59:4.IEP'`%</56")?D!(Z[OECA'@N))4;&GEO+$]99-_\QG
M1=%^Z1F<HO@Y:#&JS%9IWV2:=01Q]=`)?FXR8,7U_1+GV)/+93_(*\-;XVV:
M[I>:[(O[HJFI3&KNWF3XPFS.?P!Q_=@)?DT5-^/R?HDSE"7!I@'W;)^NZVVX
MI>T=G7B_R'0R0?&EH5BRO+=R)L<<2JH/X"T?.&&&=;M7*E49,"J&6;94F,NJ
M#^`MMF#PT\B!0;-^&(8Y%@#TNP\DMH!FB*!;YG2OM(Y>(LA5O5JU=LYU[RF&
M2@DC>(N'3?";WP$3KN^4.$?]:LH*/*T66SA[PYSK;*=0]1MPA+Y?K2NK2:(V
M5T$9P%L\9H+?LL?MM[Q#X@QE&3`3MISLG$W*19P<GN\2(69X7:]#T.5?IE7M
M?:ID-(2W=LP$3VG$+;B\2^(,'TQ#7-FC<-FX4Y\]C3.`^(P<M21@B*P)7_7O
MQ7T7<[?)*L((XC-R1&[DJO5B#DO_"9U49N=?X^UY0'7JBP5H(&%NGBR(CB"N
M7B'"#.MVRQ&3.+E>^@)]YV"N'#H"N-Z(48JR(?J0%42T2WR$*^8R60T=0EQM
MQ3C%YO?>"]-BF58$67O/U;U#Q=`AQ/5F#'/,[LG*$D%V42]6S7'DR6+H".)Z
M.X8YAA?&]4OM8CM&"X1Z7BTG+QJ)/D*K.$ILT'BN)CH"&&*H2=/.;0.L]@6_
MN4<TSU!<Z>SO3,E6:F5@B>.!>:Y.-@"XWH:QBNB($9?W2IABY_SPY)D=R2&]
M&!/J9&EP!/#Y&,IP,D'4)=L?K,2R7^S"([&"W_F1Q`TBOY$/V[Y3@G7!`<#E
M4R56%QRPX?I>&>#8/&V<9(+8YTKZ&-;N:OZ1?J'$V>M?'?S@1Z^IE=E"V0CB
MZK$3+(@.F'%]QPQP)/#W6'-*NJ<?Y3_US$X>K`X.(3XC1TZ6X4;UT`VQYD;[
MJ"0^'+M>F&MEO\)V+A52U?NC)NN#(XC+ITRP/CA@Q_4]$^?(V3]S*/J8M)7`
M$LDTF^X8_7VO#S9JE@$2+PDRS=8'1Q"7#YY8@3!NQ>7=$B.H2X)=$RZ"6J[3
MTA$TNZ!]OE<R;E?QMI+84Q>YMSY9'XP#KA\WP7+9@!67=TN<(J#?X<4225KQ
M"+1B?Z;X$>X7_5S<;PB155/MW0FI],E2]0C@ZJ$39J@%(ZO]R9[:]"A;U\NQ
M.\T5"4<`5]LP6",<&#CKAV*<8V:R="FPG]LY:IJ=RGR_=(3MQ8C2J[WQ*HI3
MFRT5CB"N'CJQ4N&`$==W2YAB(7\$J0!FNT\5)/J$EB_H%LZ6N].RE%NQ@NRG
M=:Y<.`*X?.3$ZH4#5ES>+6&&LG%[&5P"-[O\.TEH5WF^5S@AN6)Q[>QH?:=S
MJ=Q8#6D$</&P"=9'!VRXO%/B%#OY]Q@DOHV]\"6F;LPT61X=0'Q&C@#=Z_]=
M1IB5[$C/H.T_(0_6%480GY&C[/5>M`,29UM3@K6+36%^"O;M$6/2%[K)`_A:
MVVR-=`1Q^3(1IM@LE26"W/4;)K$WVJ'&R2+I".)Z,\8Y(OFX3:W:BWS89(Q1
MFRR2CB"NMV.<8P+G"%CM=2?4+Q]ZGBR2#B$NMV.8H\1JN"T49&][%O%V$O;)
M(ND0XGH[ACE&E\?U"^YR.P:KAGBT](M?Y06ZH&B9;;9,&D(;X588O5[(]C:S
M;H$DOD*?YT?VXH96G*"P'_NU*T0FRV<CB&MM&*J0ADVXOE,&*")W\*.JW9^&
M[!)7[UZ!"9<+AQ"?CV)U<\LZE[I==,NYG<E"AL<A=T]?:\["WO;4BX!@GWJ-
M%0M'`!=/E7"Q,&K%Y=TRQM!?+)._M%,A6=\@+;.=HH&I5\QDY['3OJB9]$J3
M%;,1Q+7C)EP<#=IP=9\,\6OX4-O*5CB2@)73;,UQ!/#Y&&[/D9B?HMDFE)FS
MSZG$AZ&X>>2'5'NRR$:WH31;'AP`7#Q1HO7!J`V7=\H`P9+`;%U('R[VW+$L
MXGF^4V1G\/LUB7/RZQ2HI=8FZX,#B.L'3JP^&#;C+]$O<8KDCP\V]B><C[ED
MO9UJOE]D1?7Z8*WZ=7762ZYD"I6Y^N`(X/JA$RN6#9AQ?<?$.4H$ZH*@-R8J
M8M/'A_MTQW1_1HKTX6(SMRQ2Y<R4CI5%!P"7CYPH0T[%KRO@)C;VSQPD!N]U
MLD`X@KC:BL$"X<#`63\4XQPI>WY4=C[U!?5]"6*>[Q9@SQT#Z].W-L3*[O*[
M<'UP`'#YP`D6",,V7-TG<7[=2V"H3Z_;BHABE[TW%.^37)IIQB)31:]3D(U8
M=EB>+!`.(2X>-\$"8=R*R[LES%"<+T_PBBV[O3*N#S\BXR7=XAEUO4_6KK]+
M^M$NY;F2T0C@\I$3K(D.F'%]Q\0YYLI6K6J0`3S"J;)8X&1-=`#Q.3GJ$\A^
MIC9WL&O,M*B99\L((XC/R-$=$[W'FLDN)JPDODR?GX1`VZN'LF6"W5`K!+!.
MED0'`->O$E&&R<*^:J4"^W)"C(V8<;(@.H*XW(I1BO+WX)HE%K*9*EXOG"D&
M!>NA0XC+S1CGJ/4&FRY5HA9[54G?VLF3]=`AQ/5V#',$H&V1Z%XK$W>G(D_6
M0T<0U]LQS#&\-*Y?;)?;,5@F+!H?U.Q'R$I/8*<%FQYXFZB+CJ"%^;7MD)*X
M"G9@G%-.?;]5QNB)\2U\%YG.?@)6(N.<)VME(XAKS1>KB`Z8;WE_A!G6[)?+
MR<BN6Q4E\>[Q@6AQ<`CPV2B6GBT!"?JNKCG[K<.YFD)T'$I4T/V64<ALUTG4
M7K1".%D?'$%</U<B%<(1*R[OEBC!K68EH5U'"]5JJF<<EWBG@#])C'K=H4BQ
MA@4X62N+POT"0R94$!TQX.H.B1/,VXF[BIJ%R^*LI7VU*E8;',%[-G[-;PJN
M1?./$E6P7@HZ/P(1O!0C2X,[/KGMW]B.E@4'\!9/D5A-,&[`U1TR0%`]52W7
M%`TQ%9'UQ2N<[I&2L3I#KI;#D+%8,\Q6IT<0%P^:8#EPP(CKNR7.L69_;).T
MLJ%#1]^*QM3F^Z5R]HJ@;!,-?4Z)X&QU>@1Q^=@)5L<&[+B^9^(<FY=JN\32
MND/K*>R&^Q18O&,>WO"478?L:A2]7.A,E!XJ@P[@K1\Y48:<>+,UZ1U`,L)8
MW-?]5Q_!DN``XFHCADJ"\4&S?!1&^1'X\='4**4-+]7]LAWN$0EE/2^?K'9N
M/5(3["^3B]4#1P"7#YI807#`BJM[)4RP;]DV62L9?364@<0PWRF5K;B?0;S_
M;/,$M/HV60T<05P];H+%P`$K+N^6.$6UB1\'1R,A2S:)%]POZ)="#X*$X`^%
MR5BDR1K1&.+BL1.L@`[8<7W/A#EV*#UO%?GN%RAP[VE_L#Y8`1U!?$:.J%D1
M'68RN>PV@2RK.,Q6(@8`GX]A8[__NS:QNYWQ:9SV7UJ&9V#GUJL/6CNKJ'=;
M9<R4YRJ@(X#K%XDHQ9Z\/%1+\XL[6U.\V0+H".)R*X8ITG8[B(1GG55SDSB]
M0IZL?PX@_F][W[(KV7);.9:^XGZ`40A&,%[?H88GA@>-AF`(<`L-M3PR].^]
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M5TKFQJ./<\>0Z:Q-RD'-CNPC91PGLED`+3]N7,*HG\!TC[@-U#L21=!H2S7A
M;Q@3L5:=9TIA#/!M%MJMYRPOZ2K@DD7)/@S".I?=/-PZTCF)0AUK^4PFC`"F
M=A*G1!C@+]TA$1.-PC[DOD4T9%F7/%B5>WW2NIT_!LE=][",Q:,]B!F?0!A!
MS`X;ISX88#'?+WX;D:KT2Q7=\B>YF)#'?L$QO$SUAW^0.8EC:E_E@6-\&F$$
M,3UX?)I9@,9\Q_A-7+;.MUO*\=/E7?EZ[A7D#6RGN!OKA2.R8W4>:M4!O/RP
M\5HXK_!:57<%V4[+NWMTW"IA!#&91*=,Z(Z:Y"#TFR>WV2K3+`?7%+$+"><N
M6=<%"NAT4R]0$)>4W@Y%P@!B?M@X5<(`C_F>\=NX.AGA?4G][)L<Q7VPRS3@
MF#JUXEJX;+VZJ\^!Q*4?BH0AQ.S@<8J$?AK3_>*W<+.])D\-Z:;DX2P[GN=\
MP2]$%^*2G1YR_@-+N]H.M:,08G;L.(71`(_YGG';N,M:=@>`'#TCC3)\QP.%
MRR6-1@#?9Z*]3@'`)4_$R7*?ZFKC3$\(X+W+OD[VO;*GA?5\`&:U<5\Y]O>^
MJ?NU9&Q"/]2ZNKRZ=RQ7!Q#S1PB_C5CE;O.+5>&V-+L_QN:410.`Z2SZ3>QK
MVU,?&Q%CC\VAXW8^5$4CB.DT^DT<7,W$Z[CB'A,T\J$H&D),I]%OHSWJ@E&"
M-"??6'9TYD-5-`*8SZ+;1.?0F#[6)K/HU@>'#"*7I(;A<.CU;[UB+&V'HF@$
MT6VC7(YHB&BH4E`MC2J?VPB^#5'N35<5:B/+J^U0*8L@YO/H4T4C/.9[QF\C
M#3N\@J5L540YS_)@]>O3!T.([[.1UU46;W8'8\,2CQ:=A^-B.U_3,'9*RH,H
M&V6M0XW0#_@[=!B72!C@,-TI;@.GG5-%EJ8O:7UKJ_9U#^AWRF[;5)+29S6R
MY<!</Y/-(H#I<>,31@,LIKLE8.+HABB;J:4AR];!!_LP?%IA#/&--EH=M".+
M)97#&*G\/@Y&X'PWL1+K)>K4!\U]*!9&$+/[BU,K#+"8[Y>(C3;_29?6W0X\
MI^BY+SAFVJR!'U.5QR5[C!_DM#ZM,(*8'CQ.L3#`8[YG`C9>!QJQ.!I:9$32
MS6N,<\_0LJO%M:*J-\VT\5!8\4F&$<3TZ'%):`$6\_WBM;`WXW".0K+_X5MO
MF"$>5-+<7JE+9Z*!/&9?NVVI/5B"NE32`%Y^V'@M;-.$5^`-?=IE8I&[[O<V
M.B7#"&(RB4[!,!`WV7$8L'!?9&-1I]D:6!UUCW.O(+71KT;&*,.Q>*5B+IB'
M>F$$,3UR?()A@,9\Q[A-Q.1G=//2+!3AWGB6_8)?NNU6WJU1J]9?UMTCEVZ]
M,(28'3M.O3#`8[IC_";.2P^79TELH;,Q2-`KCL$PJ`VIR-TD50^*ED>;`%Q2
M4@@Q/7A\*FF`Q]_!,VX;T=JB;#:NI*M?V4JXSZ32".#;3$2R435VJ$K=5;Z9
M8$&G0XDA@O@^$VNQ!U>Q&IO:D.3*_[W.NV!;=ME]D3<K5"/""+1.)>P`8/X@
MX381:8KY1;:VZ,Z".=9=?=.KE$8`TUET6WAMZR%\W=(B(W)SWJ<Z:00QG42_
MB6V-*\1VUU=/T!.YCD.A-(283J/?1IKM&B/:KG898!T/SN&YE-(08#Z-7A.]
MXV+^2)O+HELQG-_*:/527ZKF">!]]%+Y4">-(+IM[-,V@K?&>I'*-_SWI/F*
MC8L-D?NPR;</$3`.Y;,(XN_`HTLGC?"8[QF_C5CM&H_(S'0'[D:?V/-,-HP`
MOM%$FM=-D@4-;%F[YGWW]T>CW"QAY[^FWD,MN^KQAT/=,`*8WV%\PF&$QG2_
M^$W4@&$1%G2:TX<_N="Q7^121@/LFCE1[UCKG4K8`<#TP/$II1$.\[T2L+%>
M[[J6-51?6!CL'^K#'M4PAOA.&R\YI_=B8DEI#[8SN8-QETK7V4;589`$RSRS
M#T7#`&)^A_%IA@$6T]T2L-"&;M$,9"3])LH!]?&"6WA/,Q&&Z54&\N9\'8>*
M80@Q/71\FF&`QWS/!&R\7J@>1/J<MCP66._*W@'/$):P9F.O8VB*/C#LT#P4
M#B.(Z='CU-$"/.9[)F!C[RIQ#$R2>C-\EQ<2[[?L!3RS;34($G6K)Y;6?95Y
MII<&\/)#QVLA#4N=X)*F'!;"<%%/5>P(8C*)3N$P$#?YD>BW$=^X+KJG'O?K
M6,C71>=^P71AMQH,>./:83Y7J8?*800Q/7:<TF&`QWS/!&SLUG#)(UM51\:V
M5Z<7/%/MLA)1`>?W/G,WGKBEPQ!B=O0XI<,`C_F>\=N(U%Q_M1RFJ+KB87FF
MM;_@&;)KM>0)^ZX5C+G[6'2H*840LZ/'J9@&>,SW3,#&W8V>6ILF,KSE;?)Y
MJ)D&$-]IXZ8R[)T&#.>:T$LJPOU0;8@@OL]&+-8L>O;D,K0\N,L"E>>]D*X=
MR+L5NMZU'_/NKF&O;!H!S!\HO"9BJ6?1B!%ZZPL:7-!Q3W73"&(ZBP$3B\U&
M8[,&HT1O&^M0-@TAIM/HMW%:)H/4MDQ[$J=CR3P/==,08CZ/;AO[OH8)Y`=;
M11NY=[@=*J<AQ'P>W39Z!\?\X3:91[>&.)$6%,L>F:3&^FME>QXJIQ%$MXUC
MVK%CIFH+7`;>>,7$9CO.68XOLXJ-L(#GH9X60LRGT:6<!FA,]XO?0K[N*:V%
MV4JRG4NOAR)B"/&--C8+,FY(VYHM_G=;KT3C=8R$16N7)4\;R&X?E`E\0F($
M,;_'^(3$"(_YGO';6*]B^6!]#.);IR$7I)U[9ET52AY#"_'HME@;/2C_N32U
M"&!^\/C4TPB-^8X)V#@L>-:TX\R]U7:H3(;@WFI>+Q8_2-"XW5U`$0G#0?6R
M3T\^(DU&!C?'H80804SO*3X%,<)BNEL")BX[HMCJU,?F.I+I,E]PR[J>@I6K
M`MGN.MAWN:I;/PPAIH>.3S\,T)CN%[^%UV:-CG'>`&=C>N1IKU]VL^U-G9NM
ML:J\9M/VH8`80/P=8L>GID5XS/>,W\;K5NB!*6`IXD+>W>L+GIFV(Q>#I/T8
M)`;(%_A,.`T!9@>/U\11:%PL`EOK3UV^>!\JB!'$;!:=`F(@=/*#,6#CM3<5
M@4":2P/Q[JKID%_X\HO<OD_F%Z0^]5`_C""FQXY3/_33F.X7OX6SEFI7"8QM
M>R'DL?7[>3K@EVXW,<U2;0&%#M,>J+!.]3"$F!T[3O4PP&.^9P(VZ@%N(+;6
MNTV7M7%_P3%\/8T@3R)H27F3K!`/%:408G;P."53/XWI?O%;N,KUTLEH5*Y,
M=2X^U$L#@&^T<-H]%',-KGH[P6*L2.E08@@@OM/&W8OM9-M8D6EIE7I;XX7^
M5\?W$;21+N8;8:+9AWII!#%_C/#:B-63E>7GVGKWEMRN-EJA0[TT@IC/H]_&
M:5O.Y[Y4XB8WX3T*1Y=>&D/,YM%OX[`[$^=NI&&&;YZMMD.]-(28SZ/;QFX%
M?(06+;-1WNFB8[TT@)C/H]]&W_"8/MZFL^A3#M<WPB+H.H[#^B>Y17`_*M_[
MU-((HMM$OI[<`^#4;'7L4>_W:41,M!(^[*O7FEI.W<U#"2V$F$^C2RZ-\)CN
MF(")UV_NC2V%J=3O]WTX9<,(X!LMO&ID!7/;56I<2/9>B<7%-P569)!PT:F(
M'4),[R\^U3#"8[YG#FS$E&YEFU'*J"]X9IB^(ALS3:7J6`.<"FD!P/S8\<BE
M$0[SO>*WL!J%M)4:.8+;]JET>(+X#@NM(;4]#!$+[_M"8R00^Q6(6`9J'&*Z
M>:"R^X3#"&!V5W'IAA$.\[T2-W$T79OW*1>,[U>\LJY09+TK2[X9?SD/9<,0
M8G;D^%3#"(WI?@F8:.L<_&(3Q1?51_.SWRWS>@L8,X(FLUUV'=^+)4[1,(*8
M'SH^`2U`8[I?`A9>@TG]OA);F!0>I9Y^OUQ7;%=BW5<O.C'?[WQR2J4AQ/38
M\=NXK%@+'N6E`$%$9USK4#4,(/X./+IDPTCPY(>CWT0K'2+AT'JIXMUO0(JX
MI9E;FEPN9FY!DM,.1<,08GKH^$1#/XW9;@D8^#VC;*US,T2LW/HK;F&Z$*N-
MM5U.RIQ*AB'$[-#Q:88!&O,=$S!Q7/OU:%KH%"KW1=43MXS1S2V\:FF'2E((
M,3MT?$JIG\5LK_CM&^4*FTW%XFNV1^F:2R2-`+[/0*OM=R9NMGVKPIAUIBI$
M`-]HHCS7JX"U6D<9`Q%.KW2]6B^W*+8\%<OS6"`-(>8/#VX;KQ4@T[HZ*7IK
M[X?ZJ!\PG\6(B7;E%&*G6%X^F!J/0WDTA)A,8\1$VPX&$TWTA(E[E'JHCH80
M\VGTVV@/7^DP87)!G^OAN./11T.(^3SZ;?0-CNFC;3:-Y=M>2[26\:W(:QER
M*3S&/G[0D)?68-`0`5/DLKO:\/,^OH!,`&H26V@HKYYK0_2R^K&A&S%B8Z]V
M,\:>>F]PE4)ZPS^Y0^S37NY>Z#I%GNF9#3WFXV(B@OA&&^6TKC9L<MY-&A:Y
MV''>(;9FK\$@,<"P+Q>'H`^-^0+B^VQ$J!@BIG1YU[W*1EPLB>]YU#O-Y)5V
M+I*SR6,/F"?V.$=\HXUS;D7LZ"O+KIQ$VD;M8\.U%[.]`3`7Z6,/LB6]K&/$
M=]K8I]T2(?5AO3R7]NX?93$`+JKV)!,6.DNB;$\,/JT>`[[/0EFX:,/!LA05
MMC>^_&YP7$/ORY8IH=2I)@*ZS76.^$8;*UFY'60O;4?(=OH]C5+.LD4KRZX!
MA-A&6ECJ,>#[+"PR,JC"<+T-B(F^W%.(:<,>-][ROH,VNSM!%$![DW67OLW,
M96HSXG%W5;[@%4M*ZE0E51J6UC]6-T*(;[/0-I4W4:N[W;<X[N[6!"#K!5?\
MC3"`:`3*?7CKOC.Y`=]F894GI?2;J39[G%ETG37N$8<,KPT#S"AZ02&Z89E[
MO8#X1AO;UB5WX=5T!3@:QL)U#[AEL0_`0G+=3-4$KK9]#O@^"P==W[Q(;Z+K
M6.FL^J"W++DPHV&<!:*PS0MK1SX'?)N)&"%D1`0@ED':;+2[0^F`FT3=X$"U
M?"]&FX*4X!3O7>9AFC.BF4A7^`TY+8T'JP:6:X30D.K0M6Z=,E6\@/A&&UOI
MUG!-6]F(H'<_RV/RU])"&8.K-,/*"</I,=[[#.RZN&=Y*+3I8Z9$&%4>I&?Z
MHB:&3:3Y^K08)IG.ZQCO7?9A<K#A?Q19&*`=1@=:]UG/]V<Q6IU8VO]"^!M&
MO/$YXOM,1$MK.#'5L33$"K[0W4R_1['=+&!8'C:@;[*]=LYSP+>9R&5;]8]W
MD<.;:#>QI&WW+*YFBNWHHU="P[D)@VX]1WR?B8U-`Y8+!\JTAGNW^UI=&7;.
M<V$>9464XP\/*D(^Q/?:V)?=!+'+PFBH#7OE/N\KE.O[-9:M+K23MS%&?0'P
M?28"QNY*9<GHT`Y+NOVQNP`/:T0MU+'<_=*E(:*RKWD,^"X#D=4L>RUS8PS1
MAEVF@ONXZ:U;&6HUV4:*AA-_?Q^)?L0WVHCI3MN-2K-*N\5M/RB^=Q6_.W*X
M*5(!B?YW?T]R!/!]%O:M);7)LEE>&DZ0^#'$2&^ZLXIZQ])8I@S:G>]K;P'`
M]YE8N]98,1/L4;7A+/.^MV`-9&K#8'G!"<NAABEV]V/`=QF(-%V*Q6@H=WMU
M70$REX^*K2"J,H:&LVY=*78Y'M1>`'R?B7V;0E;ERE5I*,-DNP^;"G8-<*\N
MZ0"\@K_LYXAO-%$O^A"UCZS@0O)\XL?W4`51-_ZJ?K@TFZO(^O@^8P\@OL_&
MM=NE<7;2'+;2&*7,%U13-^(;;70JNZXTEK=6)'55B:%&BM"MU5KG61W`#^<U
M;G6[\`(S8-&$KK)DJ7QNWRAV,`+VE2$5/3D$MF<_J?:$\)(9=)K7KQ.!^,JI
MM?F.I&_>5ZU\Q;P(8C:!`0N+[4G#8@DCL-1B"R.2CFJV(;QT"CVEF5`W2>]W
M3@L7Z?LT#!PY@RSML,HI+_C$]A"W6JC)M#-6PYIJGFD1(<#<H/&5H@(,9GO$
M;V`CHQH\(S73AO7N-<R02U:Y;DQ>6W_);)@\6SM2V2*`Z5'CMG!<CV'6N<E>
MD)L5"="9EAH"3*;05VT,A$UZ'/HMQ#+].AU=Y84_J;#*=N;Y@E>F2:F#]?Y2
M:;CZHTG^^1:!$%QRU/C*JR$*TYWBMW'4:U\GUBG$6E/&V%Y/G3*W[1(755%3
M6Q'=RJ"C;2\AP-RH\563`P2F>\1OX;JTDMF6O,L*0'FVX8%6XO7)'L7>BUNM
MZ6I]X3>-.L[V<X40L\/&5T"/L)CO%[>-2H;=&8@HTSJC$$9T5$0/`;[/Q*9/
MNDF[.=>V.C'OXU!<I!7YNN&,H5M-ZUH/MD5X]F<&X-)[BD\R\!.8[`^_>5V?
M<>NR5X--&)K<:^_''FEL!O*6YP!UHB!@'FX[#B'F!HU+(8E0F.\4MXFC7JH0
M;:F,D)PVGNW!%F&W4SK9[G(@-]U3/[=<0MS.]M2'$-,#QZ<+17A,=TS`Q&E?
MC:Q1J@HR/\JMK_O<,7/:F#CU=AS)1V'LF&=G1@*`^:'C-G&7:0\^KXVI5PN-
M>R)E&&?'@T*(V33Z361-V-$0_V28#+!F?6"B2R\((:;3Z%(^(_TEOP?Z3>1Z
M103IXF^W@G%EGPF?`<#W&=C+U)%[(RSTO`/A#VL?ZIXQQ/?9R'KG"Q!;E_6J
MJ+V8K\8^TSUCB&^TL5\\,FU#Q(JVW5>U`Z-.[=>@8T<[,>#`18?2;`@P?5ST
MF^AE\7?P2R:-;J$(#9N6FFH%L)2:,/CL^U6@0_IT0[D-6]6VDN*K6740(*^/
M4X#?N&ZGOVHK8D"5:X$7?;C%R"U[1@`3N?/;5NQ]XSJT&E7UOEGJZTSQ#"&F
M\^<VD9==C=EH=8TL9#3CH^CCE0!C@-DLND3`4"=)[W9>$[E<HH6\45>U&2$E
MW\=>Z==-J8R%@B[9^QZ[]B,),(27'#4N/2Q`8+9#W/95OBZWGCRFMBMH]JB;
M>#UBHV''Q*X'T">6CZVM,\TS!)@=,FX#K[T!H]6I9X)D2/EX5:%7``P!9G/H
MT_\B89,>AWX3N32[1F?TV;HV++,\V.?C=<L@D_:$9)/S,>&O>20!AO"2P\:E
M`$88S/>)V\3>["+:@0AG:S=WK>/8*9/6]9#<9%WTXV>,]F`\="F`(<3DN/'I
M?Q$2T[WB-W%V>TIUR3[PI@TWDNY^[I8U6$L[!+MT3%R8!=J#/6<N12R$F!XZ
M/M4SPF.^9_PV[GD]V[7+DN"9\I;.?3N/$AB!>Y-]G:SP2P,+.M7$M@CZYW&(
M,;:9>$:D*TGIHSR/E.@`7'X_\:F``0J3/>*WKPVK)($XA(HBEKX+G;JDDIUE
ME-VRI+J0K/#J/!,`0XBY0>.3_R(4YCO%;R-?]3TI,191"V9%Q&T^=PML5,&\
M[P4@F>H7]WD_X_H$P!!B>NCX9*$(C_F>"=BH+R"@H6R7U5D/<]Y^P3&C5&TX
MZR)5/>67[/M%@4OU#`'FQX[71+N#4F2SAG1/M4RL)FB=B9X1Q'06W28B5DTJ
M1`*D.7ME1A#QF402`4PGT25[!CI+>N_S&\C%O#SQ@]=UW4=;AYIG"/%])K92
ME<35Y]*KQTB.`^Q])GF&$-]H8UU-.SY6H<W.$^\Y#^7$"-[;[+OV$"&-*W9C
MT28LN-OY>%-KK=>`,PJKR-"X43N5.P.`V4.BVT(WB?EN22;1K1&5U>T1,]VM
MJ=>9%>)Y*'_ZT=SV;=EBH-7O.8FT89&']<[LFYWMQ=E"JPG1"#8DLN-4^O0#
MYM+G-4^VUBEY>U95>O?]_;$1U=,/F,V?W\)F#P@A^]]:H9`7%(C/Q6@_8C:+
M7AW0WTG2NYW71"(NUZN86)CJ_7*8\HA?\`K94U8+JP95`AOQ'J="H!<N/6C<
MFIB7P=<]\H<__.&I;;^VH3HO00I#5:FF_769K7]M\KD/?FU21C'182!S9\W`
M%CI'[S<PG^F;01B'-9\$0]`6>SYCHJOI*D_NVZ.Z;F`^5?."."G4_$"YB_@[
M*6Q<UK3O#P@B8Z*F;385*0X%2;;;+U>K53?WS=%D`_1O83Y5Y$(H21[_@?H6
M(B>+9)<]O*['X7J1E3UPBKR.LZ,LLXW!>[?6[`KQ(=_U6YC/1;483HK3?R"?
MA<C)(MEECUR\KVVF'MI#&RP%=ME1EN?U;@/U.50&Q0I/WB[[)\KG4E$$)<GC
MGPM_(6JR*/:8,Z\Z69%36'K/Y^AC[-^V^5SM"L%\H36[#`N*N<I0H1:KZC)6
M-'!6N=Y*(7E,6(6+67G<</RICA6"R8GC'RA6(7*2./:84\MW<AB,&`XZ8^M!
MCC>1<L,#ZP6[8Z..<8/RN0P5@LEQ^`\$IQ`W61S[[&E67<-<0DUQ]#')$609
M285^5>]#MU,CFI>>_/PGRN="4@0EQ^.?:PDA9K(8=IE#S;@!*2Q?M=JH6A.,
M48Q$U8Z7CK:59*S$U[R9%W^@?\5P<ESNL8:1X-@Q13LSRUBV\FTR\[G`%8+)
MXL9C#E(R^ZY5JA:0:I<<]';N_*SJ'8))8N<'@E4HDI,ZA,<<VNV2'W?526^/
MU6^GO$_EJ`#(EUEB+^5V769IAD*]R;N/O\7Y3&Z*H7RA-:/9(>2)K]!+D)$E
MEU)O4#[55H(X7V>-/D(@.`-K"NF6I8]Z,_9Y>O<LU@_0NXLF.G+17NE\8\ZG
M.E$,)V>P\9GC8">+Y21Z?E`EO]HLI'1]7Z_H]JV:3MWX4[FI-WZN^\1P//:T
M;DFG/+:G]4^TV;?539<]=+TN@<7-TEP:<<&UWF8[GZH[,9PL?CSVE'WQ@[R=
M22_?!5&W];G/Y)L02A8Y'FN0#6K="`.8'CKH%9'"-['\N6`1Q,D@YZDXX0SD
MI`[A,:=?.\M%69LZ-A&2_QKGN"Y[3!M,ZLGU+GNN-P?T"2]*CL>?EO)]Y"1Q
M[+%&GD56<K!PUY(M^IV\4A^D6+9R"8K<%2GV8@'1]=6"@.CBQ<EQN,L:ZE;.
M-IBQY,*#'I(GO#`IW#Q7)WP>3PH<ESFSFD0K^U)U"Z24AE?C4Y8W4;-3@[2H
MCQ61)[PP*4Y_KD[XV,EBV67/*M9&9@I[AJ8U*HNC+,N35%(8YU&+E3Z6[`(-
M"11>G!2G/Q<H?.QDL>RR9U=;*V)E45701$_M.Z@#"3^+-=O%XDPG/GS/J'4'
MBOD^D"27/]==?-QD<>RQ!ZMSRQTQNK2M91]J6-_&1`HGSE?:0Z-H?HG&78_0
MRDFTC[G(T^"AOH9)$-T>"`)(9PJI%#Z8C$!^KE#XB,DBV&5/E3*UM*F[ZWW:
M4Z)BU2C'5>]>[]]8]G>P==!.-\+R4Y7""9/C\N<JA8^=+)9]]MCUQHS&50O`
MR$3Z3<740W)K9C%6)H-T(3/6OM%@GVD4+I`D?S\O>[N(2>+79\U5@>N5BVU2
M`?!M/OV<8+[*BEAWLFZTE&ON^UPAV<6)D^%NES%C74>?9!6MA0_\Q>`:$5V<
M,%G<.,P1ARO,GI85U8&HN:F//Z\2.W&2V/FQ\N**X:RN\-043,KF)30M6A<I
M;7[X/4]E%R?.EQF#'-M\O623J&HJ%>2,D.[BQ?DZ>[`(&=\9'-L>=:P897=(
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MLW=!+DSZ5<A_=IQB>WM!DBG+0+%`Y9A0$8-YS=]/50HG-5D4N^R12PKT9V&8
M:9:S+7W@(DCRM.P08YH=ZAXR18Z84N'%R7'Z4Z'"24\6S1Y[R(JE&"Y%#R*Y
M[(P^[E%RL-R+/;560:"^6#IVQ8*R1ZKZ7I@<IS^77WSLY)#LLJ9>CW01\=:P
M(*1D)29Y^&"^T)IV76^"A>:J4ZM6#4-AC4:.59,;4.R<M>@`?49T"A]*3A0_
M5RE\U&11[+*'B>@ZR'2=!T7LKRC'Z`WVHN">K2H[4@)O*R15.'%27/Y4J7!1
MD\2PSYAM*%-F!-T`(;)Z"S-L!`.>B6VUN=>8$:G"AY+D[J<%<!\W211[K&G;
M!A.6NQ1T$!VRQWP%*9X7C&QF')+EX?_O63@DOSAQ<CSNL0;&V%6<6(7J,,J\
MVX<M<L_4%Q],%C<.<P8U8Z?+%?=:8D'"]MOCEYXZL1,GAY[G,HPODI,ZA,^<
MZUG87O5H,MH,#"DM),+X8+[0FHH$T31`9DWMD53W=;O;_:D$X\/Y4GOT)5+%
M6;J)OQ(&U=U"2H43Y@NMF6SDR),+JIMPJ=Q7N)]W2V70SYOV3[GOJ]$.ZB\^
MG*1QQV6/AY\LGG/X\92^FT8/B^U33Y!7V=W1:E!^\<`XK.%"MNU]2^'-?KFD
M)Q0V!YW$[-GHXR*ML)ZY"<HO/IP<>CSFK$LLVU3M#?NV>Y\W2Q6/_N+"R:+'
M94]K5AUJO9!.Z@VI_(C*7#Z<)'Z>*Q6N:,[J%0YSJ-@#:[*$(KV8%>O'FZ63
MCV2RDU%[==91CB?2BIORD$>J<.%D.?UY7=_#3@[''EN&W1$G1>]BA\FEMMK"
M%%L93Z8&:2#OO9:@P.7`2/*UPY:YKSI?'U9%Q@H5BX"@3N&#>9T9CT;A<G5.
MQ+C,J96JR09VXK_LW7][XM])\)J&LO5[$"+\48)\IDYX,#(<[5$F'*PDD>NS
M9MA!_\*KZU43\'-O/<ROE83D,FT5D;!RQG]&4)?PX;SN;H<HX2`FB5^/,:V1
M%=X+VZ.%15Z1J3W*[RKVF@_XZG9O,GX"KZB<Y<))\K=#:_'0D\2RRYH]^/I1
M51\&05@4XA64)5PX7V@.]Z+L5*2(^A(4I@>YFRL<.]RN8S6-=<V,")(0B0D3
M/IB<4'9($QYVLEAVF;.F*094JUZ?AF_<:XXHR<N.\L&87JU_/MAY^%2<<.'D
M^-RC3GC822+99\XNUZV$^$XMD191A:(24).M$EI`IKVW+C.1>M+'[:_/]`D?
M3([3/35O#SU9-/OL&7;:3+:9ZRS!4AR>,9;Y>Y&9QY@Z+/?"'X9NC^[BPDER
MNL<<N=Q!V\@DHPDTU46+8[*+"R:+'8<Y&(&M0"KG/[2PLN3AB6!MV`&3Q(U#
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MPG5S^9Z/XF%[KY$\6%F1D6+TWH+2A`\GQ>6>0KZ'GB26?>9(\J_L2,54=ZD4
MHK;"+-OCO%+BT_MT92Y$#XA(+BZ,+(<_M86M9U*?1><\.&#B/T&%PH630(Q#
MGO"X.BMD/.;,:DMHDAL?=7\4TN;%*\XP6;U5G@!4=C!%E)MUZU.!P@>2XVZ'
M1.'A)HEBES5L.]D(,Z[&'_Z`G[=C#-O75"2#TSKNV..VW.A0)WPPKSO;HTUX
M>$FBUV?.NF(/>:CN92)YB*W4,,5L:4.=L-AV\@ZY6#Q8Q7?AY+C<H[EXZ,FB
MV64/@H^NWT5LSUGB5]UF50YUPH7SI?9T6T6"E78-3/*J2`^'S^C+SK"4;<&S
M;E?/'GG"@Y(3RAYUPL--%L<^>R8;.[3GN'`D#L(L;SL9U@8-W<R+A2M3I:`\
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(XF.\,Q9<`@``
`
end
-----8<------ [snip] ------------8<------ [snip] -------

    pederb has a patch below that "fixes" the problem, but this is not
    a proper fix. Somebody needs to investigate to figure out what is
    really going on. A good start would perhaps be to test on a
    variety of different OpenGL drivers to figure out whether or not
    it is a weakness with NVidia's drivers.

-----8<------ [snip] ------------8<------ [snip] -------
Index: src/misc/SoGLnonindexedTristripSetTemplate.icc
===================================================================
RCS file: /export/cvsroot/Coin/src/misc/SoGLnonindexedTristripSetTemplate.icc,v
retrieving revision 1.6
diff -u -r1.6 SoGLnonindexedTristripSetTemplate.icc
--- src/misc/SoGLnonindexedTristripSetTemplate.icc       14 Feb 2003 17:31:28 -0000       1.6
+++ src/misc/SoGLnonindexedTristripSetTemplate.icc       20 Nov 2003 12:43:03 -0000
@@ -67,8 +67,12 @@

     glBegin(GL_TRIANGLE_STRIP);

+#if NBINDING==PER_VERTEX || NBINDING==PER_STRIP || NBINDING==PER_FACE
 #if NBINDING==PER_VERTEX || NBINDING==PER_STRIP
     currnormal = normals++;
+#else
+    currnormal = normals;
+#endif
     glNormal3fv((const GLfloat *)currnormal);
 #endif
 #if MBINDING==PER_STRIP || MBINDING==PER_VERTEX
@@ -80,8 +84,12 @@
     SEND_VERTEX(idx);
     idx++;

+#if NBINDING==PER_VERTEX || NBINDING==PER_FACE
 #if NBINDING==PER_VERTEX
     currnormal = normals++;
+#else
+    currnormal = normals;
+#endif
     glNormal3fv((const GLfloat *)currnormal);
 #endif
 #if MBINDING==PER_VERTEX
-----8<------ [snip] ------------8<------ [snip] -------

    20031203 mortene.

=====================================================================

099 View frustum culling is not done unless there are cached bounding
    boxes available.

    E.g. if an app programmer calls setAutoClipping(FALSE) on a
    standard viewer, SoGetBoundingBoxAction will not be invoked each
    frame, and there is no other mechanism in Coin which triggers the
    calculation and storage of bounding boxes.

    This is not a bug, but a performance enhancement suggestion. It
    can have a dramatic effect under certain circumstances.

    pederb suggests this can be solved by letting SoGLRenderAction
    invoke an SoGetBoundingBoxAction on the root node before
    rendering.

    20031211 mortene, reported by pederb.

    UPDATE 20040909 mortene: the above suggestion for a fix may not be
    such a good idea:

      On Wed, Sep 08, 2004 at 04:27:55PM +0200, Karin Kosina wrote:
      :
      : [...]
      :
      : And: why don't we implement pederb's suggestion (applying a
      : GetBoundingBoxAction before rendering)?

      [pederb:]

      I feel it should be possible to toggle this feature on/off.
      Other applications might apply an SoGetBoundingBoxAction before
      they call SoSceneManager::render() (or render the scene using an
      SoGLRenderAction directly). If you have a scene with lots of
      animations, SoGetBoundingBoxAction can be pretty slow.

      If you have an application that does multipass rendering, and we
      apply an SoGetBoundingBoxAction every time
      SoGLRenderAction::apply() is called, that would be pretty bad.


    Note: when this bug has been fixed, remove the warning note on
    So@Gui@Viewer::setAutoClipping() which mentions this problem.


    UPDATE 20060208 mortene: pederb added support for automatically
    applying an SoGetBoundingBoxAction, but it is at the moment only
    possible to activate with an envvar -- as it is basically a policy
    decision whether we want this or not.

=====================================================================

102 Lack of control over OffscreenRenderer context setup

    There is currently no way to enable stencil-buffer and other GL
    features when creating the SoOffscreenRenderer context.  There
    should be a way to control offscreen renderer features in some
    way.  Requested by Josh Flowers.

    20040203 larsa.


    UPDATE 20040203 mortene: my suggestion is that the API should
    present the possibility of setting up a callback weighting
    function to run the available GL-capable contexts through, because
    this will give the API client full control if he/she so wishes.

    Something like this has already been implemented internally for
    SoWin, see SoWinGLWidget::weighPixelFormat(). I.e. default
    weighting algorithms like that function should be available inside
    libCoin, with the option of overriding them through the API.

    Designing the API so it manages to abstract away the peculiarities
    of each window-specific (GLX / WGL / AGL) context struct format
    might be challenging.

    UPDATE 20040922 mortene: there have been several more requests for
    this feature. Lacking control over the GL context capabilities can
    be a serious show-stopper for programmers wanting to use e.g. the
    stencil buffer, so I consider this a medium-to-high priority item.

    UPDATE 20050510 mortene: something like this has been implemented
    by TGS, with among others the SoGLGraphicConfigTemplate class.
    Consider using the same design and API.

    UPDATE 20060224 kyrah: mortene has implemented a hack to allow
    requesting a stencil buffer via the environment variable
    COIN_OFFSCREEN_STENCIL_BITS. Just noting this here so we don't
    forget to get rid of this workaround when this issue is finally
    resolved.

=====================================================================

108 SoVRMLBackground (and SoVRMLImageTexture?) texturing doesn't work
    with SoOffscreenRenderer.

    Most likely because SoVRMLBackground uses an SoVRMLImageTexture,
    which loads textures in the background (if Coin is configured with
    platform-independent mt-support, which is the default). So on the
    first frame, which is what the SoOffscreenRenderer grabs, textures
    will not have completed loading yet.

    Should perhaps best be solved by using a new element
    "SoSynchronizedElement", or some such, which tells traversal
    operations whether or not they are allowed to run in parallel
    threads. SoOffscreenRenderer should then add such an element
    before invoking the SoGLRenderAction.

    20040427 mortene.

    UPDATE 20050505 mortene: Colin Dunlop reported another effect of
    this; fractured textures (i.e. SoTextureScalePolicy::policy set to
    FRACTURE) will typically render in low quality. The fix outlined
    above should also be the underpinning for taking care of this
    problem.

    When this problem has been properly taken care of, remove the
    following hack from src/misc/SoOffscreenRenderer (which doesn't
    seem to work):

      const int bigimagechangelimit = SoGLBigImage::setChangeLimit(INT_MAX);

    (And check whether or not there are any similar hacks to also take
    care of.)

    UPDATE 20050512 mortene: correction; fractured textures are not
    scaled or otherwise prepared in separate threads, but gradually in
    multiple render passes. The SoGLBigImage::setChangeLimit() is
    there to make sure that all necessary preparations are made on the
    first pass. This seems to work correctly upon some slight testing
    with the following iv-file:

-------------8<---------------- [snip] --------------------------
#Inventor V2.1 ascii

Separator {
   DirectionalLight {
      direction 1 -1 -10
   }
   PerspectiveCamera {
      position -0.47643667 0.5896858 6
      nearDistance 0.1
      farDistance 100
   }
   Separator {
      TextureScalePolicy {
         policy FRACTURE
      }
      Texture2 {
         filename "/tmp/M245-m040904.png"
      }
      Cube {
      }
   }
}
-------------8<---------------- [snip] --------------------------

    I'm therefore not quite sure what the problem is at Colin's end,
    so I've asked for more information. (My best guess is that it is
    downscaling of tremendously large textures to not-so-large
    offscreen pixmaps which may look bad, due to scaling algorithms
    tuned for speed, not quality, in SoGLBigImage.)

=====================================================================

114 VRML-nodes are erroneously renamed upon import.

    VRML-nodes can legally be named in ways which are not ok for
    ordinary Inventor/Coin nodes. They are therefore sometimes renamed
    in error upon import of VRML97-files, from within SoBase::setName().

    I suggest this is fixed within SoBase::setName(), by making the
    code check whether or not the class is internal and has the
    "SoVRML" prefix, and if so; do the Right Thing wrt the VRML97
    spec.

    (BTW, do we have the same problem wrt VRML 1 nodes?)

    According to kintel, this is probably quite a bit of work to do
    properly, one will for instance have to read node names in UTF-8.

    20040521 mortene, bug reported by kintel.

=====================================================================

121 SoOffscreenRenderer tiled rendering causes problems with
    subsequent viewport changes.

    The SmallChange ViewportRegion node sets up a sub-viewport
    according to the SoGLRenderAction's viewport settings, and does
    not know about whether or not the current rendering is part of a
    tiled rendering. This causes artifacts for the commonly used
    "sub-viewport + orthocam + scene graph" trick for rendering
    overlay graphics.

    This needs to be fixed partly in Coin (there is now no mechanism
    for passing on information about whether or not we're doing tiled
    rendering), and partly in SmallChange (ViewportRegion must set up
    the sub-viewport according to the current tile).

    As for the Coin part, this could perhaps best be solved by using a
    new element "SoViewportTileElement", or some such, which should
    contain information about tilesize (absolute and
    relative). SoOffscreenRenderer should then add such an element
    to the state stack before invoking the SoGLRenderAction.

    This is a tricky issue, of medium priority.

    20040705 mortene.

    UPDATE 20050520 mortene: bug item #191 depends on this bug.

    UPDATE 20050520 mortene: As a side effect of this fix, SoCamera
    could then also do the view volume narrowing itself, removing the
    need for the current ugly hack in SoOffscreenRenderer to do this.

=====================================================================

129 Dragger bugs. This is a collection of bugs mostly related to
    scaling of draggers.

    JackDragger: Scale down as much as possible. Use viewAll to keep
    the geometry in focus. When a certain limit has been reached, it
    is no longer possible to click on the drag tabs. Might be because
    JackDragger uses childdraggers to do it's job, and child draggers
    currently do not have a good enough minimum scale restriction.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    JackDragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    Scale2Dragger: Scale down alot, and you will have problems with
    viewAll. Also had problems with grabbing the dragger. The geometry
    of the dragger does not get scaled properly. Some of these
    problems are probably related to the lack of a minimum scale
    restriction when using child draggers. Others are problems
    relating to the way the dragger geometry/tabs are scaled.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    Scale2Dragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    SoTabBoxDragger: Lack of minimum scale restriction because it uses
    child draggers (SoTabPlaneDraggers). The dragger geometry does not
    look correct with extremely small scale.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    TabBoxDragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    SoTransformerDragger: Geometry does not look correct when having
    non-uniform scale in the scenegraph. Load the following file into
    examinerviewer and click on the rotation tab on one of the sides,
    and you should see the problem.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    TransformerManip {}
    Text3 {
      string "b"
      parts ALL
    }
    --------8<------------- [snip] --------------8<------------- [snip]

    Also, free rotation (holding shift while rotating) does not behave
    100% as expected when the locater position goes outside of the
    projection sphere. Using the same scenegraph, press shift and
    rotate the center tab, and you will see the problem. This problem
    happens when having non-uniform scale.


    SoTransformBoxDragger: Unable to grab hold of the dragger tabs
    when scale is small. This is probably also realted to the lack of
    minimum scale restriction when using child draggers.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    TransformBoxDragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    Scale1Dragger: Impossible to grab hold of dragger tabs again after
    scaling as small as possible. Tabs should have a certain dimension
    even though the dragger has been scaled down.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    Scale1Dragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    Scale2Dragger: Difficult to repick dragger tabs when scale gets low.
    viewAll does not work properly.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    Scale2Dragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    Scale2UniformDragger: viewAll does not work as expected when scaling
    down alot. Also the geometry of the dragger gets strange when having
    low scale.

    --------8<------------- [snip] --------------8<------------- [snip]
    #Inventor V2.1 ascii
    Scale2UniformDragger {}
    --------8<------------- [snip] --------------8<------------- [snip]


    General observations:
     A general problem is that it is difficult to separate and click on
     a tab of choice when scale is small. This should be fixed for
     easier use. It would be nice if the tabs are spaced so that it is
     possible to see which is which. Dragger geometries should also be
     fixed so that they are intuitive and easy to use even when scale
     is small/big, etc.

     OIV does not have a restriction on the scale when using child
     draggers, but we should since without there will be trouble.

     viewAll does not always work as expected, not zooming in enough on
     some draggers under certain circumstances. This should be checked.


    Priorities:
     1) Fix minimum scale restriction when using child draggers
     2) Fix scaling/positioning of dragger geometries so that
        they are intuitive and easy to use at any scale
     3) See if the viewAll works after these bugs have been fixed.
        If not, find out what's wrong and fix it.
     4) Roam all the draggers, have fun with them and look for
        problems to fix
     5) Fix all other dragger bugs in this list :)
     6) Fix the rest of the bugs in this list (:


     20040810 jornskaa

=====================================================================

133 SoExtSelection selection area rendering bug

    SoGui toolkits needs to set the window id element so the
    SoExtSelection node can know which window id it is being
    manipulated in, so it doesn't render the selection area rectangle
    in all available views.

    You probably also have canvas refresh issues related to this -
    make sure the inactive views does not refresh their displays
    during interaction.

    20040831 larsa.

    UPDATE 20040831 mortene: This is a high priority item, as it
      causes Coin to be unusable when using multiple views and
      SoExtSelection together.

    UPDATE 20061026 mortene: we have recurring support requests for
      this bug to be fixed.

      The above suggestion for how to fix it may be be unnecesary
      roundabout and complicated, BTW. I'm guessing it should be
      possible to just store the GL context id (taken from
      SoGLRenderAction::getCacheContext()) when a selection is first
      activated, and then only render it to that context.

      With regard to picking up events from only the same context, I'm
      unsure as to whether we're currently able to detect which
      context an SoHandleEventAction comes from, though.

=====================================================================

135 Elements used by <node>::doAction() functions not automatically
    set up.

    When writing extension actions, it is a major hassle that one
    needs to explicitly enable all elements used by nodes' doAction()
    functions for the new action. It is a hassle for many reasons, not
    the least that one can not find out which elements are actually
    used without reading the source code. In addition, these are
    really implementation details that may change without notice,
    rendering client code buggy without any warning.

    So this needs to be fixed. Would probably best be done by changing
    the internals somewhat so that extension actions automatically
    inherits a list of enabled elements from SoAction aswell, and not
    just from the other actions, where this works as it should
    already.

    As an example, it should be possible to do something like this for
    e.g. SoLevelOfDetail:

    ------8<----- [snip] --------------------8<----- [snip] --------------
    SoLevelOfDetail::initClass(void)
    {
      SO_NODE_INTERNAL_INIT_CLASS(SoLevelOfDetail, SO_FROM_INVENTOR_1);

     // FIXME: these elements are used from doAction(), so they should be
     // enabled for all actions.
     //   SO_ENABLE(SoAction, SoComplexityTypeElement);
     //   SO_ENABLE(SoAction, SoViewportRegionElement);
     [...]
    ------8<----- [snip] --------------------8<----- [snip] --------------

    Medium priority item, we've had support requests / problem reports
    from at least two external developers about this. It is a
    non-trivial issue to fix.

    20040902 mortene.

=====================================================================

140 Rust in the SoVRMLParent machinery for keeping both an SoMFNode
    and an SoChildList of its children.

    Jochen Stier reported a problem with this on coin-discuss, and
    provided the following code for reproducing the problem:

----8<----- [snip] --------------------8<----- [snip] ----------------
I have a little piece of code that demonstrate the problem. As far as
I understand, all the print statements should output 1. But that is
not the case..... The refcount of the children of lGroup is
incremented during the copy, but never decremented :(

 SoVRMLGroup* lGroup = new SoVRMLGroup();
 lGroup->ref();

 lGroup->addChild(new SoVRMLGroup());
 ((SoVRMLGroup*)lGroup->getChild(0))->addChild(new SoVRMLGroup());
 ((SoVRMLGroup*)((SoVRMLGroup*)lGroup->getChild(0))->getChild(0))->addChild(new SoVRMLGroup());

 LOG<12> () << lGroup->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)lGroup->getChild(0))->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)((SoVRMLGroup*)lGroup->getChild(0))->getChild(0))->getRefCount();

 SoVRMLGroup* lCopy = (SoVRMLGroup*)lGroup->copy();
 lCopy->ref();

 LOG<12> () << lGroup->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)lGroup->getChild(0))->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)((SoVRMLGroup*)lGroup->getChild(0))->getChild(0))->getRefCount();

 LOG<12> () << lCopy->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)lCopy->getChild(0))->getRefCount();
 LOG<12> () << ((SoVRMLGroup*)((SoVRMLGroup*)lCopy->getChild(0))->getChild(0))->getRefCount();
----8<----- [snip] --------------------8<----- [snip] ----------------

    At the least, this can lead to memory leaks as some nodes will
    never be destructed.

    The problem is related to the fact that we need to keep two copies
    of references to children nodes for an SoVRMLParent: one from the
    "children" SoMFNode field (because that is in the public API) and
    one that can be fetched from the SoNode::getChildren() method
    (also part of the public API, and used by e.g. traversal methods).

    The issue of keeping these internally in sync at all times has
    been a long-standing source of bugs, and this bug report indicates
    that it is still not working 100%.

    Consider this medium-to-high priority.

    20040908 mortene.

    UPDATE 20060208 mortene: see also items #162 and #196, which are
    closely related.

=====================================================================

141 OpenGL context store/restore "stacking" for
    e.g. SoOffscreenRenderer will fail if different window system GL
    bindings are used in libCoin vs the main windows system binding.

    [We used to run into this on Mac OS X, where Sc21 uses
    NSOpenGLContext, and libCoin attempts to check (and restore) the
    current context with aglGetCurrentContext(), but this is now
    fixed, see update below.]

    We can in theory also get the same problem for other combinations,
    e.g. SoXt for a X11/GLX binding for the application on a Windows
    system where libCoin will use WGL internally.

    20040915 mortene, reported by kyrah.

    UPDATE 20040915 mortene, this is kintel's write-up of assumed
    problem areas and issues to cover:

        Cases in which this has potential of breaking Coin:
        o Offscreen rendering w/Cocoa [fixed]
        o Offscreen rendering using a window system layer on top of
          the native system (as Morten also pointed out), e.g. SoXt
          under Windows.
        o Offscreen rendering of a scene containing an SoSceneTexture2
          node
        o Offscreen rendering done from a callback node
        o Nested SoSceneTexture2 nodes

        What I have here is not a solution but rather requirements for
        a solution that will help implementing this properly. The
        basic mechanism for solving the problem should be the same as
        kyrah suggested initially (i.e. using callbacks into client
        code):

        o Must call back into client code (we cannot e.g. execute
          Objective-C code from Coin)
        o Must have a stack of these callbacks (to take nested
          offscreen renderers into account)
        o Must take into account that multiple threads can
          simultaneously render (by having one stack per thread?)
        o Should be able to share resources between on-screen and
          off-screen contexts:
          - Imagine rendering a huge volume with SIMVoleon and wanting
            to also offscreen render that every frame. Without texture
            sharing, this is not going to be very fast.
        o Should also support sharing between "layered" and native
          contexts:
          - Using e.g. GLX on Windows or NSOpenGL on Mac, we should
            make the offscreen context (WGL and AGL respectively) able
            to share from the layered contexts by e.g. extracting the
            core component.

    larsa's feedback:

        It must behave like a stack, but it must be implemented
        non-stackish.  I'm trying to figure out how this relates to
        how BeOS did things.  There you would have to lock a GL
        context when using it, and unlock it afterwards.  You couldn't
        lock another context while one was locked already, so you
        would basically have to unlock the onscreen rendering context
        while doing any offscreen rendering.  So that's a model that
        should fit below the surface of whatever design we end up with
        on this.  I'm not sure if this is a problem at all yet though,
        just that you wouldn't be able to translate makeCurrent() into
        Lock() calls without pairing every Lock() up with Unlock() of
        the previous context...

    Update 20051108 kyrah:

        Part of this problem (offscreen rendering in Sc21) has
        been fixed by adding a callback interface for providing an
        external offscreen rendering implementation (see gl.c:
        cc_glglue_context_set_offscreen_cb_functions()). Note that
        the other issues mentioned above ("stacked" offscreen
        renderers, GLX-on-Windows) still remain to be resolved.

=====================================================================

143 Warnings typically triggered from faulty input files should be
    easy to hide.

    We should provide a mechanism to hide all the warnings typically
    thrown out when encountering a buggy input file, as a service to
    applications that have little control over their input sources
    (like VRML viewers, for instance).

    A couple of ideas:

        * Provide an envvar, "COIN_DONT_WARN_ON_SCENEGRAPH" or some
          such, that can be set to turn them off.

        * Use a standard "tag" or prefix to those messages, so they
          are easy to filter out for someone installing their own
          error handler on SoError / SoDebugError.

    Both the above suggestions could be implemented. In addition, one
    needs to audit much of the Coin code to find the relevant
    invocations of SoDebugError::post[Warning]().

    Fairly low priority task.

    20040924 mortene.

=====================================================================

157 SoFont does not support styles under Windows.

    When adding a style to a fontname, the fallback font is used
    instead (Arial). This wrl file recreates the problem:

----8<----- [snip] --------------------8<----- [snip] ----------------
#Inventor V2.1 ascii

Font { name "Times New Roman:Bold" }
Text2 { string "test" }
----8<----- [snip] --------------------8<----- [snip] ----------------

    (This works fine when using FreeType on Linux.)

    20041102 handegar

    UPDATE 20041102 mortene: note that it works as expected when
    removing the ":Bold"-suffix, which indicates that the string we
    pass to Win32's CreateFont() is wrong. Most likely it needs to be
    broken up properly, and we should probably pass the information
    about "bold" weighting as one of the other arguments to the
    CreateFont() call, and not as part of the font name.

    When taking care of this bug, please also check how "italics" and
    other styles works (or not works).

=====================================================================

162 Inconsistent traversal behavior upon traversal of VRML97-nodes
    with SoSFNode and/or SoMFNode fields.

    According to pederb, only some VRML97-nodes forwards scene graph
    traversal to the sub-graphs in their SoSFNode- or SoMFNode-fields.

    E.g. SoVRMLText does not traverse its fontStyle field, but
    SoVRMLShape and SoVRMLAppearance traverses their SoSFNode-fields.

    Is there any good reason why we should /not/ let all nodes
    continue traversing down into SoSFNode and SoMFNode fields? If
    not, this should be fixed.

    20041104 mortene, reported by kintel.

    UPDATE 20050711 mortene: see also items #140 and #196, which seems
    to be the same issue.

=====================================================================

168 Memory read/write errors for VRML97 Group / Parent / Switch nodes.

    Valgrind reports

        Invalid read of size 4
        Invalid write of size 4

    for the local WRL file ~pederb/store/vrml97/sony_cam.wrl upon exit
    of the viewer it has been loaded into. The callstack backtraces at
    the point of the errors are:

----8<----- [snip] --------------------8<----- [snip] ----------------
==31646== Invalid read of size 4
==31646==    at 0x408BCA83: SoNodeSensor::dyingReference(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:102)
==31646==    by 0x406FD3E7: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:436)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40954FD6: SoVRMLSwitch::~SoVRMLSwitch(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Switch.cpp:179)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40949F72: SoVRMLParent::~SoVRMLParent(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Parent.cpp:165)
==31646==    by 0x4093CAD1: SoVRMLGroup::~SoVRMLGroup(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Group.cpp:291)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    Address 0x45436CEC is 44 bytes inside a block of size 48 free'd
==31646==    at 0x4002CDAD: __builtin_delete (vg_replace_malloc.c:244)
==31646==    by 0x408BD037: SoSensor::~SoSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoSensor.cpp:128)
==31646==    by 0x408BC3A7: SoDelayQueueSensor::~SoDelayQueueSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDelayQueueSensor.cpp:80)
==31646==    by 0x408BC1B2: SoDataSensor::~SoDataSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDataSensor.cpp:87)
==31646==    by 0x408BC9B5: SoNodeSensor::~SoNodeSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:66)
==31646==    by 0x40747E0F: SoProtoInstance::~SoProtoInstance(void) (/home/sigma/mortene/code/coin-head/src/misc/SoProtoInstance.cpp:130)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40748372: SoProtoInstance::sensorCB(void *, SoSensor *) (/home/sigma/mortene/code/coin-head/src/misc/SoProtoInstance.cpp:271)
==31646==    by 0x408BC285: SoDataSensor::invokeDeleteCallback(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDataSensor.cpp:248)
==31646==    by 0x408BCA7F: SoNodeSensor::dyingReference(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:121)
==31646==    by 0x406FD3E7: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:436)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40954FD6: SoVRMLSwitch::~SoVRMLSwitch(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Switch.cpp:179)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==

and

==31646== Invalid write of size 4
==31646==    at 0x408BCA96: SoNodeSensor::dyingReference(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:103)
==31646==    by 0x406FD3E7: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:436)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40954FD6: SoVRMLSwitch::~SoVRMLSwitch(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Switch.cpp:179)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40949F72: SoVRMLParent::~SoVRMLParent(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Parent.cpp:165)
==31646==    by 0x4093CAD1: SoVRMLGroup::~SoVRMLGroup(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Group.cpp:291)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    Address 0x45436CEC is 44 bytes inside a block of size 48 free'd
==31646==    at 0x4002CDAD: __builtin_delete (vg_replace_malloc.c:244)
==31646==    by 0x408BD037: SoSensor::~SoSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoSensor.cpp:128)
==31646==    by 0x408BC3A7: SoDelayQueueSensor::~SoDelayQueueSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDelayQueueSensor.cpp:80)
==31646==    by 0x408BC1B2: SoDataSensor::~SoDataSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDataSensor.cpp:87)
==31646==    by 0x408BC9B5: SoNodeSensor::~SoNodeSensor(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:66)
==31646==    by 0x40747E0F: SoProtoInstance::~SoProtoInstance(void) (/home/sigma/mortene/code/coin-head/src/misc/SoProtoInstance.cpp:130)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40748372: SoProtoInstance::sensorCB(void *, SoSensor *) (/home/sigma/mortene/code/coin-head/src/misc/SoProtoInstance.cpp:271)
==31646==    by 0x408BC285: SoDataSensor::invokeDeleteCallback(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoDataSensor.cpp:248)
==31646==    by 0x408BCA7F: SoNodeSensor::dyingReference(void) (/home/sigma/mortene/code/coin-head/src/sensors/SoNodeSensor.cpp:121)
==31646==    by 0x406FD3E7: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:436)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==    by 0x40766885: SoBaseList::truncate(int) (/home/sigma/mortene/code/coin-head/src/lists/SoBaseList.cpp:141)
==31646==    by 0x40701316: SoChildList::truncate(int) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:176)
==31646==    by 0x40701B95: SoChildList::~SoChildList(void) (/home/sigma/mortene/code/coin-head/src/misc/SoChildList.cpp:84)
==31646==    by 0x407E803D: SoGroup::~SoGroup(void) (/home/sigma/mortene/code/coin-head/src/nodes/SoGroup.cpp:239)
==31646==    by 0x40954FD6: SoVRMLSwitch::~SoVRMLSwitch(void) (/home/sigma/mortene/code/coin-head/src/vrml97/Switch.cpp:179)
==31646==    by 0x406FD43D: SoBase::destroy(void) (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:447)
==31646==    by 0x406FD99F: SoBase::unref(void) const (/home/sigma/mortene/code/coin-head/src/misc/SoBase.cpp:634)
==31646==
----8<----- [snip] --------------------8<----- [snip] ----------------

    20041105 mortene.

    UPDATE 20050520 mortene: this bug may be fixed, at least I
    couldn't reproduce it now (with Valgrind 2.4.0.rc4).

=====================================================================

191 SoOffscreenRenderer bug with SoOrthographicCamera in multitile-mode.

    The following iv-file results in a faulty image when fed to the
    Ivtools/renderos/renderos example, using a tile size of 512x512:

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

OrthographicCamera {
   viewportMapping CROP_VIEWPORT_LINE_FRAME
   nearDistance 0.1
   farDistance 53000
   height 6393.33333333333
   position -15206.092773 9440.072266 17456.371094
   orientation -0.536894 -0.843216 0.027043   0.798199
}

Cube {
   width 500
   height 100
   depth 50000
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    To reproduce with a forced tile size of 512x512, set the following
    envvars:

        COIN_OFFSCREENRENDERER_TILEWIDTH=512
        COIN_OFFSCREENRENDERER_TILEHEIGHT=512

    It appears to have something to do with the SoOffscreenRenderer
    camera view volume narrowing to not work correctly with the
    cropped viewport, as commenting out the "viewportMapping" field of
    the camera causes the bug to go away.

    20050519 mortene.

    Yes. SoCamera will render a frame in the viewport when the
    viewport mapping is set to one of the CROP modes. To fix this we
    need a new element called SoViewportTile or something to make it
    possible for SoCamera to detect that the SoGLRenderAction is
    actually rendering a tile in the viewport, and not the entire
    viewport. This is also needed to make other viewport-based
    rendering nodes, such as legend, work.

    I don't know if SoViewportTileElement is a good idea though, since
    it will complicate the CROP rendering code in SoCamera quite a
    bit.

    Another idea: Is it possible to use the WindowSize-parameter in
    SbViewportRegion to specify the entire viewport? This must be done
    _after_ the viewport is initialized in the GL context though, maybe
    in a pre-camera callback or something.

    20050519 pederb

    UPDATE 20050520 mortene: after discussion with pederb, we consider
    the Correct Fix for this to first take care of bug item #121. This
    means the SoOffscreenRenderer adding an element to the state stack
    upon traversal start which identifies the current tile to be
    rendered (i.e. containing row and column index, and ditto
    dimensions), and then leaving it to SoCamera to render the
    viewport cropping correctly. (As a side effect, SoCamera could
    then also do the view volume narrowing itself, removing the need
    for the current ugly hack in SoOffscreenRenderer to do this.)

    UPDATE 20050714 mortene: there's something else odd going on
    here... if the tilewidth and -height is set to 640 and 480,
    respectively, which is the same as the default size of renderos'
    buffer, it also comes out wrong -- but only if tiled rendering is
    *not* used. If we on the other hand force tiled rendering on (by
    using the envvar COIN_FORCE_TILED_OFFSCREENRENDERING=1), and the
    tile size is at least of dimensions 640x480, everything works..?
    Strange.

=====================================================================

194 SoOffscreenRenderer in tile-mode asserts on VRML97 scene graph.

    The following scene graph attempted drawn with an
    SoOffscreenRenderer (use e.g. the 'osgrab' debugging backdoor in
    an examiner viewer to reproduce) causes an assert to hit in
    SoOffscreenRenderer, due to the tail of the
    SoGLRenderAction::getCurPath() being NULL at one point. The tail
    of the path should likely never become NULL.

----8<-------------- [snip] -------8<-------------- [snip] -------------
#VRML V2.0 utf8

Group {
   children [
      Shape {
         geometry Sphere { }
      }
   ]
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    20050606 mortene.

=====================================================================

196 Bug with VRML97 scene graph searching.

    The below scene graph (with a VRML97 PROTO) demonstrates a problem
    with SoSearchAction in VRML97 graphs: one can not get paths to all
    nodes, which causes e.g. PROTO-connection to "miss out" on certain
    connections it should be able to make.

----8<-------------- [snip] -------8<-------------- [snip] -------------
#VRML V2.0 utf8

PROTO AnnText [
 exposedField MFString sText ""
 field SFFloat sFontSize 0.1
 field MFString sFontFamily "courier"
 field SFString sFontStyle ""
 exposedField SFVec3f slocation 0 0 0
]
{
   Transform {
      translation IS slocation
      children [
         Shape {
            geometry Text{
               string IS sText
               fontStyle FontStyle {
                  size IS sFontSize
                  family IS sFontFamily
                  style IS sFontStyle
                  spacing 2.5
               }
            }
         }
      ]
   }
}

Transform {
   children [
      AnnText { slocation 0 0 0 sText "hepp" }
   ]
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    pederb suggests that this should be fix be extending the
    SoVRMLParent etc hack by duplicating MFNode contents into "hidden"
    SoChildList instances. pederb has a framework for this in:

        void SoVRMLParent::updateChildList(SoNode * nodewithsfnode, SoChildList & cl)

    Note: is there some other, more fundamental, way we could fix this
    without applying this hack?  It's not very nice, and is giving us
    headaches now and then with bugs which are the results of these
    SoChildList instances not kept in sync.

    20050609 mortene.

    UPDATE 20050711 mortene: see also item #162, which seems to be the
    same issue.

    UPDATE 20060130 mortene:

      This is a /really/ tricky issue, and one which we don't really
      know yet how to fix:

      The source of the problem is that VRML nodes use SoSFNode and
      SoMFNode *fields* to keep track of their child nodes, while the
      "traditional" Inventor nodes use the SoChildList class.

      For certain functionality of the library, we do however need a
      SoChildList -- like for e.g. SoPath-instances being able to
      track their way down through the scene graph.

      So to work around this problem, there is a hack in place at
      various spots in the VRML nodes with MFNode "children" field(s):
      the node internally keeps an SoChildList, which is attempted
      kept synchronized with the MFNode field. This SoChildList
      instance is then available through the virtual
      SoNode::getChildren() method, just as for the "normal" Inventor
      nodes.

      This hack does however have certain ugly weaknesses:

        * it has not been set up everywhere it is probably needed
          (none or few of the nodes with SoSFNode fields have an
          SoChildList "on the side", for instance)

        * it causes various problems with the SoPath class, which has
          no knowledge of SoSFNode / SoMFNode, and is completely and
          utterly geared towards going through SoChildList

        * there is copy'n'paste-jobs of code for
          SoChildList<->SoMFNode synchronization at various places --
          code which is not straightforward to bug-fix and maintain

        * the SoNode::getChildren() method returns a mutable
          SoChildList (!)

      So, this is one of the really major design flaws of the library,
      and it looks like we need a stroke of genius to resolve this
      knot.

      One venue which may be interesting to investigate would be to
      see if letting SoChildList take care of the synchronization vs
      SoSFNode & SoMFNode fields would be sensible. Or perhaps let the
      SoMFNode and SoSFNode classes have a built-in SoChildList which
      they could auto-sync?

      Another option which may be a possibility would be to let SoPath
      know about SoMFNode and SoSFNode. My gut feeling about that one
      is that it would quickly make the SoPath API a lot uglier, tho'.

      I've been asked for an estimate on this task, but given that we
      don't yet know /how/ to fix it, makes this impossible as of
      now. I'm quite certain however that we are talking about > 1
      week of efficient coding and testing.

=====================================================================

203 openal.dll instances may have different calling convention.

    Different than the custom under Windows, and different than from
    what we assume it is. This causes a crash on those platforms when
    attempts are made at using sound. (The first attempt at calling an
    openal.dll function will crash the application, because of
    unexpected stack layout on return.)

    My suggestion for a "best fix" would be to include code in Coin to
    detect the calling convention used for a DLL. To do this, one
    needs to do some trickery with checking the stack pointer after
    calling a function of the DLL -- something which probably requires
    some x86 assembly programming.

    This is still not a perfect fix, as functions in a DLL can have
    different calling conventions. Perhaps it would be possible to
    query the DLL directly about this? Request for assistance at a
    Usenet Windows programming group did not result in any useful
    hints for how to do this.

    20050701 mortene.

=====================================================================

205 Hardcopy export does not work correctly with viewport region
    changes.

    As reported on coin-discuss by user Ole Jacob Hagen:

        > Generating a postscript file when my scenegraph consists of
        > one instance of Smallchange's [ViewportRegion] works, but
        > when I have arranged my scenegraphs into four viewports
        > (2x2-array) it doesn't work.

    Is this a bug or a limitation? Waiting for pederb's input on this.

    20050714 mortene.

=====================================================================

208 Too high tessellation by default for 3D text?

    The 06.3.Complex3DText example updates at a very slow framerate,
    so it seems like our default settings for 3D text complexity is
    too high.

    Investigate, compare with SGI Inventor.

    Not particularly high priority item.

    20051021 mortene.

=====================================================================

209 Bug with SoTranslate1Dragger.

    The following iv-file should work the same in Coin as in SGI
    Inventor, but it doesn't:

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

Switch { DEF DRAG Translate1Dragger { } }

RotationXYZ {
   angle  = DecomposeVec3f { vector = USE DRAG.translation }.x
   axis Z
}

USE DRAG

Cube { width 0.5 height 0.5 depth 0.5 }
Separator {
   Translation { translation 0 -5 0 }
   Cylinder { radius 0.2 height 10 }
}
Translation { translation 0 -10 0 }
Sphere { radius 0.4 }
----8<-------------- [snip] -------8<-------------- [snip] -------------

    In Coin, the dragger geometry is translated, for some odd
    reason. In SGI Inventor, it only rotates along with the rest of
    the geometry, as expected.

    20051208 mortene.

    UPDATE 20060130 mortene:

      As for an estimate on this bug, I expect it to be on the order
      of just a few hours to fix. Need to investigate and figure out
      why it happens, and the fix is likely to just be a little
      modification of the internal default geometry for the dragger,
      i.e. to the Coin/data/draggerDefaults/translate1Dragger.iv file.

=====================================================================

210 SoText2 font size policy in Coin incompatible with SGI / TGS
    Inventor?

    According to Frank Hibbeln:

> The doc's for SoFont indicate the size for Text2 nodes is
> interpreted as pixels whereas the TGS Inventor doc's indicate it's
> "printer points" (ie i point = 1/72 inch ).

    Should investigate this with SGI Inventor, and see if we need to
    fix anything for Coin.

    20051212 mortene.

=====================================================================

211 SoTextureCoordinateSphere et al does not work with some
    primitives.

    SoSphere, SoCube, SoCone and SoCylinder has not been modified to
    work properly with SoTextureCoordinateSphere,
    SoTextureCoordinateCylinder etc, but instead just ignoring this
    texture function nodes.

    20060213 mortene.

=====================================================================

213 SbTesselator fails on polygon which it is supposed to handle.

    The following Inventor file demonstrates a problem with
    SbTesselator (according to kintel, it is supposed to be able to
    handle this case):

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

Separator {

   ShapeHints {
      faceType UNKNOWN_FACE_TYPE

   }

   FaceSet {
      vertexProperty
      VertexProperty {
         vertex [5616.34650  -94900.63137  0,
          19459.61213 -83978.42044  0,
          18951.60431 -69500.14309  0,
          34699.90509 -70643.16262  0,
          35080.91681 -85883.45559  0,
          21491.65119 -91598.56887  0,
          19459.61213 -83978.42044  0,
          5616.34650  -94900.63137  0,
          9172.41681  -108997.90091 0,
          52607.25275 -107600.87747 0,
          55909.31525 -90328.54153  0,
          46257.12775 -61371.98684  0,
          20475.63556 -53243.82669  0,
          282.24494   -65690.06887  0,
          5616.34650  -94900.63137  0]
         orderedRGBA 0xac00ff
         materialBinding OVERALL

      }
      numVertices 15

   }
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    Example provided by Frank Hibbeln.

    To see how this polygon is supposed to be tessellated, use the GLU
    tessellator by setting the following envvar:

        COIN_PREFER_GLU_TESSELLATOR=1

    20060216 mortene.

=====================================================================

214 VMLScript JS support does not work at all with latest Spidermonkey?

    We've had an external report about the VRMLScript node no working
    at all with JS, not even the models/vrml97/script/simplest-possible.wrl
    test case.

    The errors output from the code is

        Coin info in SoJavaScriptEngine::field2jsval(): no handler found for MFString
        Coin info in SoJavaScriptEngine::field2jsval(): no handler found for SFBool
        [and more of the same]

    There may be bugs in the Script.cpp, JS_VRMLClasses.cpp or
    SoJavaScriptEngine.cpp code which manifests itself with the latest
    Spidermonkey release, and not with previous versions (I'm not able
    to reproduce the errors on my system with the earlier version
    1.5rc6a).

    20060217 mortene.

=====================================================================

215 SoText2 behaviour under the influence of an SoClipPlane is
    peculiar.

    The following scene graph demonstrates how SoText2 nodes doesn't
    work well with SoClipPlane:

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

ClipPlane {
   plane 1 0 0 -1
   on TRUE
}

Text2 {
   string "hepp"
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    Spin the camera around, and see how the SoText2 string slips in
    and out of view, sometimes partially, sometimes fully.

    Should perhaps try to implement a more consistent behaviour, for
    instance always showing or not showing the string depending on how
    the its 3D origin point is placed vs the clip plane(s).

    We need to make a policy decision on this, and document that, at
    least.

    20060307 mortene.

    This happens since we do culling against the view volume and the
    clipping plane based on the bounding box of the node and each
    character. I'm not sure if we should do culling against clipping
    planes for 2D texts or not. Should investigate how SGI Inventor
    handles this.

    20060327 pederb

=====================================================================

217 SoDrawStyle::POINTS not handled for linesets in SoToVRML2Action

    If a scene contains a line set where the draw style is set to
    POINTS, the lines will still be converted to lines and not points.

    Needs a bit of code to fix so I'll probably delay this until the
    next Coin bugfixing session.

    20060327 pederb

=====================================================================

219 View All doesn't work properly with SoText2 nodes

Pressing "View All" in an examinerviewer with the attached scenegraph
moves the "ZZZ" string outside the frustum and clips the "YYY" string
in half.
The initial view when setting the scenegraph looks correct though.

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

SoText2 {
  string "XXX"
}
Translation {
  translation 0 100 0
}
SoText2 {
  string "YYY"
}
Translation {
  translation 0 100 0
}
SoText2 {
  string "ZZZ"
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    20061129 kintel

=====================================================================

221 Global field values are not evaluated when reading from file

When creating and connecting from a global field from an .iv file, the
value of the field is not evaluated until the field is touched. If
nobody touches the file from code, the initial value will never be
set.

The offending code is probably related to the "do not notify" flag
being set when making the field connection (SoField.cpp:2283).

----8<-------------- [snip] -------8<-------------- [snip] -------------
#Inventor V2.1 ascii

SoText2 {
  string "XXX" = GlobalField {
    type SFString
    vfctime "Test"
  }.vfctime
}
----8<-------------- [snip] -------8<-------------- [snip] -------------

    20061129 kintel

=====================================================================
