From Wang Rui, "I've found a very covert bug in current VDSM implementation. It makes the computation of positional texgen incorrect if the shadow scene is under one or more child cameras with view/proj offsets. Using current stage instead will fix it."

src/osgShadow/ViewDependentShadowMap.cpp

@@ -2303,9 +2303,8 @@ bool ViewDependentShadowMap::assignTexGenSettings(osgUtil::CullVisitor* cv, osg:
     osg::ref_ptr<osg::RefMatrix> refMatrix =
         new osg::RefMatrix( camera->getInverseViewMatrix() * (*(cv->getModelViewMatrix())) );
 
-    cv->getRenderStage()->getPositionalStateContainer()->addPositionedTextureAttribute( textureUnit, refMatrix.get(), texgen );
-
-
+    osgUtil::RenderStage* currentStage = cv->getCurrentRenderBin()->getStage();
+    currentStage->getPositionalStateContainer()->addPositionedTextureAttribute( textureUnit, refMatrix.get(), texgen );
     return true;
 }