retain objects for several frames before deleting them. Also added RenderStageCache
into CullVistor.cpp that is used for handling RTT osg::Camera's that are being
used in double buffered SceneView usage.
related support into osgViewer::Viewer and osgViewer::StatsHandler.
Added lazy updating of text in StatsHandler HUD to minimize the impact of
slow text updating on observed frame rates.
Added setting of osg_SimulationTime and osg_DeltaSimulationTime to the uniforms set by SceneView
Added frame(double simulationTime) and advance(double simulationTime) parameters to
osgViewer::SimpleViewer, Vewer and CompositeViewer.
Updated various examples and Nodes to use SimulationTime where appropriate.
Added s/getStats() to osg::View and osg::Camera.
Added population of View::getStats() with frame stats in osgViewer/Viewer.
Added Basic StatsHandler to osgviewer example.
as osg::GraphicsOperation. Unpdated parts of OSG depending upon these.
Added a virtaul bool valid() method to osg::GraphicsContext to allow apps to
test whether a valid graphis context has been created or not.
added frame stamp updating and update traversal to osgViewer::Scene/Viewer.
Updated osgcamera example to use new Viewer API calls instead of using local
rendering calls.
It turned out that nv.getDatabaseRequestHandler() returned NULL; using this pointer to call one of DatabaseRequestHandler's methods is of course illegal.
I patched it by adding: if (nv.getDatabaseRequestHandler())"
From Robert Osfield, reordered the addition of the nv.getDatabaseRequestHandler() check to allow the
default traverse() to be called when no database handling is attached to the visitor.
Performance tests on big models did not indicate any performance penalty in using doubles over floats,
so the move to doubles should mainly impact precision improvements for whole earth databases.
Also made improvements to osgUtil::PlaneIntersector and osgSim::ElevationSlice classes
method may not take the node into consideration when computing its bound.
In this case of:
switch->insertChild(0, child, false);
switch->insertChild(1, child, true);
child will not be used in computeBound, but will be drawn.
Solution:
Changed compute bound to loop over children using an index instead of an
iterator. This behaviour matches that of the traverse method."
-----------------------------------------------
incorrectly checks for a valid OpenGL version. Instead of checking if
the current version is exactly 1.2, it will check if the version is
greater than 1.2."
handle scenes with multiple views with elements that need coordinating on a per view basis.
Added beginings of new osgText::FadeText class (not functionality yet).
"I've made some changes to osg which I think make it easier to control
the render order of CameraNode's. Instead of using the built-in orders
(PRE_RENDER, POST_RENDER, NESTED_RENDER), you can specify an integer
order. Values less than zero are pre rendered in order. Values greater
than zero are post rendered in order. And a value of 0 is equivalent
to NESTED_RENDER.
The changes should be fully backward compatible. Also, I changed the
RenderStageList type from a vector to a list because I needed to be
able to insert values anywhere in the list.
The reason I made these changes was because I wanted to be able to set
the render order of a CameraNode at runtime without having to reorder
it in the scenegraph."
and later in the final submission message (relating to what has been finally been merged) :
"I've rethought my implementation and came up with something a little
better. The setRenderOrder will continue to take an enum, but will
have an optional orderNum parameter which can be both positive and
negative. I think this method is more intuitive and flexible."
dlopen/dlsym instead of NS*Symbol* functions for 10.3 or later. The
latter is now deprecated.
I'm not completely sure how to test the changes. But I compiled these
new changes in and ran the osgvertexprogram and osgshaders examples
and they seemed to display the same results as before.
With respect to the use of dlopen and dlsym, I was uncertain if the
usage should look like Sun or FreeBSD (or something else). Apple's man
pages discourage the use of RTLD_DEFAULT so I avoided the FreeBSD
implementation and copied the Sun implementation. I'm not sure if this
is correct/ideal or not."
"I was experiencing hard crashes of my application when using PBO's on
machines that don't support PBO's. I think osg incorrectly checks if
PBO's are supported.
I added a new method to the BufferObject::Extensions class which
returns if the "GL_ARB_pixel_buffer_object" string is supported. This
fixes the problem on my end. Machines without PBO support will
continue to work and machines with PBO support will still be able to
use it."
in incorrect texture assignment. Solution was to a compareTextureObjects() test to the Texture*::compare(..) method that
the osgUtil::Optimizer::StateSetVisitor uses to determine uniqueness.
Note that firstly it always returns the positive real quaternion (positive w)
Note also that it will sometimes slightly differ from the results of the other methods because it assumes that the input matrix really is a rotation matrix - if it isn't, e.g. because of rounding error, then the output quaternion will be very slightly different. For example, the test matrix
0 1 0 0
1 0 0 0
0 0 0.999999 0
0 0 0 1
will return 0.707107 0.707107 0.0005033 0.0005033
whereas the previous methods return 0.707107 0.707107 0.0 0.0
However, since quaternions are rotations, the meaning of how to convert a matrix that isn't a rotation is a little unclear..."
setFactorAndUnitMultipliersUsingBestGuessForDriver() static methods to
osg::PolygonOffset to help implement workarounds for lack of portablity of
settings between different hardware.
"If a texture is used that is not a multiple of four, and compression
was requested through the texture's internal format, the texture's
internal format reverts to a non-compressed type and a NOTICE is given.
At present, compressed textures must have a multiple of four in each
dimension."
"Attached is some fixes for Image::readImageFromCurrentTexture. It was
failing when used with a new Image object (some of the fields were
not set before they were used, and some were set to the wrong values).
A new optional parameter was added to give the desired data type, which
defaults to GL_UNSIGNED_BYTE, so programs should only need a recompile
to work properly."
textures. Near the top of the function that implements texture
subloading, osg::Texture::applyTexImage2D_subload(), the local variable
compressed_image examines the _image's_ pixel format to see if it is
compressed. However, further on, in calls to getCompressedSize() the
_texture's_ pixel format is used. In my application's Texture2D class,
I use osg::Texture::USE_ARB_COMPRESSION to
osg::Texture2D::setInternalFormatMode(), which causes the internal format
to become one of the generic ARB_COMPRESSED types, which do not have a
specific size. Thus the recent warning message added to
osg::Texture::getCompressedSize() is triggered. The correct behavior is
to use the format mode from the Image class instead of the Texture class
within the subload implementation, and then the size is calculated
correctly."
based on the internal format. There is similar code in the Texture class
but it does not account for the ARB types. I modified the Texture class
implementation to show a warning when an incomplete internal format is
used to calculate the image size."
when the binding mode changes, thereby avoiding uncessary calls to dirtyDisplayList.
Note, from Robert Osfield, moved bodies of set*Binding() into Geometry.cpp to avoid
clutter in the header.
that removeChild(Node*), removeChild(uint) and equivilant Geode methods are
now inline methods, not designed to be overriden, and seperated out the
multiple remove method to be called removeChildren(uint, uint) which is
now the only virtual method. There removeChildren is now the method to
override in subclasses.
This reorganisation requires some call code to be rename removeChild usage
to removeChildren.
osg::DisplaySettings class. OsgCameraGroup will now read the setting
from the DisplaySettings instead of hardcoding the value. I added the
following commandline option to be able to set the multisample value:
--samples <num>
One thing to note, OsgCameraGroup would previously check if the
computer is an SGI and set multisample to 4. I retained this check in
DisplaySettings to be backwards compatible."
an invalid bounding sphere if it hasn't seen a cull traversal yet. It
depends on _firstTimeToSetEyePoint, which is initially true, then false
after a cull. There might be a better way? If so, let me know.
This change does resolve the issue I had encountered with auto scale to
screen and incorrect culling."
FrameBufferObject.cpp there is also another fix: when initializing a FBO
attachment from a CameraNode attachment, the renderbuffer's format must be
set to the attachment's internal format, not to the image's pixel format.
Another problem is that attaching a renderbuffer to the FBO through
CameraNode is not simple (if not impossible) if you don't intend to specify
an Image object. Probably CameraNode could be enriched with an
"attach(buffer, width, height, format)" method. For example if you attach a
color buffer as a texture whose size is different than that of the
CameraNode's viewport you also need to attach a depth buffer of the same
size, because the depth buffer that is automatically attached by RenderStage
has the viewport's size. FBOs require that all attachment have the same
dimensions, so said setup will fail if you can't specify a custom depth
renderbuffer"
Added better handling in computeIntersections(..) of nodes that are internal
to the scene graph, correctly accounting for the accumulated transforms.
Changed the EventVisitor so that it only traveses active children rather than
all children.
Updated wrappers.
improve the backwards compatibility of peformance on systems that have OpenGL2.0
drivers but without hardware that can't handle non power of two textures.
