Hi Ron, The AutoTransform management always hits up against the issue of it's size being relative to each cull traversal and the cull traversal's culling is based on subgraphs bounding volume (computed from size and position of children) so it's a catch 22. The bounding volume initial traversal issue is partially addressed by leaving the bounding volume in it's invalid/undefined state, which is essential the max value < min value, the cull traversal knows to not do a cull test on nodes which such bounding volumes. Unfortunately this only works for the AutoTransform itself, the transforms parents don't know the position of a node with a invalid bounding volume and needs to compute it's own size so ignores the invalid ones but in the process can lead to an inappropriate size.
I have to head off line for a short while so can't provide a suggestion for a fix right away. Robert. On 18 July 2012 21:49, Ron Mayer <[email protected]> wrote: > UPDATE: > > Let me start by saying Paul was right 8) . It is the AutoTransform! The 'x' > did not show when I turned off AutoTransform because it may have been > obscured by another object, so I falsely thought it was not the AutoTransform. > > I spent the day debugging the bound speheres for culling and I found the > problem. Here is what happens: > > 1) I add the new 'x' to my scene with an AutoTransform and > AutoScaleToScreen=true > 2) frame() is called > 3) OSG prepares for culling and calls scene->getSceneData()->getBound() to > calculate the bounding sphere of the scene (ViewerBase.cpp line #760) > 4) The graph gets traversed in a depth-first manner, recursevly computing the > spheres > 5) At some point our AutoTransform is reached, and computeBound() is called > (Node line #366) > 6) But allas, AutoTransform::computeBound() returns an invalid bound since > the _firstTimeToInitEyePoint flag is true! (AutoTransform.cpp line #389) > 7) The _firstTimeToInitEyePoint flag gets a false value only after the first > CullVisitor has visited the AutoTransform, which happens only AFTER the > computation of the bound returns (AutoTransform.cpp lines 199 - 229). > 8 ) But at that time it is too late since the value of > _boundingSphereComputed is set to true (Node line #370) so the invalid > AutoTransform sphere does not get recalculated until later. > > What I did is set a breakpoint in AutoTransform::computeBound(), and set > _firstTimeToInitEyePoint to false inside the debugger, and then the 'x' > showed. > > So to summarize the problem - the implementation of > AutoTransform::computeBound()will return a valid spehere only if the > AutoTransform node was visited by the CullVisitor at least once, but since it > is called before the culling, it caches an invalid transform. > I am not sure if this is a bug or because I may not be seeing the whole > picture. > > Our plan to solve this is writing our own BoundCallback and set it on the > AutoTransform of our 'x'. > The new BoundCallback will return a small sphere (let's say radius 2, center > 000) becuse our 'x' only represents a point in the space, and we want it to > behave like a HUD element. This will also make the calculation of the > bounding spheres of parent nodes of the 'x''s more accurate since it will > take into account only the point in space with the small sphere around it. > > If it works for us, it might be worth thinking about adding this solution to > the osg code. But I need more guidance on that, so thank you again! :). > > > Cheers, > Ron > > ------------------ > Read this topic online here: > http://forum.openscenegraph.org/viewtopic.php?p=48922#48922 > > > > > > _______________________________________________ > osg-users mailing list > [email protected] > http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org _______________________________________________ osg-users mailing list [email protected] http://lists.openscenegraph.org/listinfo.cgi/osg-users-openscenegraph.org
