inserted a static transform to do the work of rotating the scene, then
applied a standard osgUtil::Optimizer::FlattenStaticTransformsVisitor
traversal to flatten the transfrom down on to the geometry nodes.
Moved CullVisitor/RenderStage/RenderStageLighting across to managing lights
entirely within RenderStageLighting, and changed the management of modelview
matrices so that RenderLeaf now stores the absolute modelview matrix, rather
than a model matrix as done previously. The later allows RenderLeaf's to do
a glLoadMatrix rather than a glPushMatrix/glMultMatrix/glPopMatrix.
level which appears to cause the code to hang, -O works fine for some strange
reason, I havn't found any direct problems in the code. This may be compiler
bug or a code bug, the bug may go away in later versions of GNUC++, the
problem currently occurs under GCC2.96.
plugin names hanging around despite being deprecated. These definiations
confused the make install/make instclean so it tried to link a non existant
library. Have removed old definitions and everything intalls properly once
more:-)
Change the osg::NodeVisitor::getLocalToWorldMatrix() /
getWorldToLocalMatrix() methods so it do not initilize the matrix being
passed to them, and assume the the users has already initiliazed them
appropriately, such as to take account of camera positions.
coords of the intersection point with the scene rather than the world
coord value for that point. The code now uses
osgUtil::Hit::getWorldIntersectionPoint() method for getting the world coords.
Added support for getWorldIntersectionPoint to Viewer.cpp.
Put and #ifdef around the setting of the default display list visitor so
it nolonger sets in under IRIX. This is a world around to the IR graphics
imbending lighting info into display lists, if the display lists are
created before state is set up it produces lighting artifacts such as
flickering. Remove the the default display list init visitor removes
these problems, display lists are then built on the fly and drawables
a drawn for the first time.
does not flatten drawables which have double references to them, but only
one reference with a static transform on it. This case can't be flattened
but this wasn't picked up. The code now detects the case correctly.
distribution clean from GUI examples, these instead should live in the bazaar.
This can better accomodate the many different types of viewer that users are
developming.
the conclusion that the osg::Matrix::inverse was broken, have lifted a new
implementation from sgl and it seems to work fine. Will need further testing
but looks good.
objects that are deleted whilest still having a positive _refCount, such
as when a object has been deleted on the stack, yet other references still
exist for it. Have put the desctructor implementation in Object.cpp to
avoid adding yet another file with only a couple of lines of code in.
by adding a ComputeTransformCallback to osg::Transform, and have now removed
the recently added AutoTransform since it is nolonger required. Have also
updated CullVisitor to account for the new ways for tracking transformation
matrices in the scene.
on the stack thus ensure that they arn't created locally and have their
memory deleted incorrectly.
Also updated the IntersectVisitor so that it no used osg::ref_ptr<> internally
for storing all data. Have also move the IntersectState helper class to inside the
IntersectVisitor which simplifies the external interface to the class.
class now combines Cloner and DeepCopy into one class. Cloner and DeepCopy
will be removed in next commit.
Also have added osgcopy app to Demos which shows how the CopyOp have be
subclassed to create users own specific handling of copying.
Have fixed copy constructor problems in GeoSet which fix the deep copy
problem experienced yesterday.
copy constructor which takes an optional Cloner object, and the old
osg::Object::clone() has changed so that it now requires a Cloner as paramter.
This is passed on to the copy constructor to help control the shallow vs
deep copying. The old functionality of clone() which was clone of type has
been renamed to cloneType().
Updated all of the OSG to work with these new conventions, implemention all
the required copy constructors etc. A couple of areas will do shallow
copies by design, a couple of other still need to be updated to do either
shallow or deep.
Neither of the shallow or deep copy operations have been tested yet, only
the old functionality of the OSG has been checked so far, such running the
viewer on various demo datasets.
Also fixed a problem in osg::Optimize::RemoveRendundentNodesVisitor which
was not checking that Group didn't have have any attached StateSet's, Callbacks
or UserData. These checks have now been added, which fixes a bug which was
revealled by the new osgscribe demo, this related to removal of group acting
as state decorator.
method
by the _unitScale to compenstate for different scales in the flt model. This
_unitScale had already been applied to the geometry data.
Integrated new defaults for the texturing wrap mode, for when now .attr file
is supplied with a flt file.
and getWorldToLocalMatrix(), computeLocalToWorld() and computeWorldToLocal()
methods.
Have updated the CullVisitor, IntersectVisitor and Optimizer to use the
new osg::Transform::getLocalToWorldMatrix() which has the same functionality
as the old getMatrix() but is now supports subclasses of osg::Transform
transparently.
Have added osg::PositionAttitudeTransform as subclass of osg::Transform
which manages the transform as position and attitude via a Vec3 and Quat
respectively.
correct redundent groups but for groups which had a single child it was
was not using replaceChild to rettach the child to its parents. It was using
parent->removeChild(group)...parent->addChild(child); which was allow the child
to be attached in a different place than the original group, this broke LOD code.
Have used removeChild instead which solves the LOD problem.
modes first then attributes since the it was cause a display bug on some
datesets. It seems that the modes needs enabling before glMaterial's
take affect, at least on the NVidia drivers under Windows and Linux.
The OpenGL reference guide doesn't mention any dependancy so I'm not
sure what the official line is. Some other OpenGl attribute and modes
need to be applied in that order according to the blue book, however,
drivers, at least the NVidia drivers seem require the opposite. This may raise
the spectra of before and after mode applies, but this will require extra
support in osg::State and osg::StateAttribute, and would have to be handled
on a per attribute basis, and possibly different of each platform. Yuck.
of objects which were transformed by multiple matrices at one time - this
cannot be handled in the flattening process (since we only have one piece
of geometry to transform). This visitor now handles this case by disabling
flattening of any objects and transforms associated in this way.
change in osg::Billboard.
Removed in instance of an object being passed () brackets, have removed the
brackets to avoid an amiguity with the compiler intepreting it as a function.
Alas changed the indenting to use 4 spaces instead of tabs, this irons out
differences when working under Windows and Unix, keeping the identing
consistent.
attributes are now set first, then modes. This is keep consistent with
the setting of glColorMaterial and glEnable(GL_COLOR_MATERIAL) as specfied
in OpenGL documentation.
defined the value to be 2048, this should avoid problems with path lengths
which have been overflowing the 1024 limit (Sasa report this bug). The
real solution is to use std::string throught FileUtils, but this will have
to wait for another day.
from passing the vector contents to the GeoSet::setPrimLengths(..) etc.
methods. This has been done to fix the compile under the MS .NET compiler
which has changed the definiation of the std::vector<..>::iterator to
a class rather than a pointer.
