"
Found in the join file the fix for the bug found by Rafa.
Problem :
osgIntrospection::Value grp(new osg::Group);
osgIntrospection::ValueList vlcall;
vlcall.push_back(osgIntrospection::Value("toto"));
const osgIntrospection::MethodInfo *m =
grp->getType.getCompatibleMethod("setName", vlcall, true);
if (m)
{
m->invoke(grp, vlcall); // ** SEGFAULT here
}
Algorithm explanation :
The "invoke" method try to convert "grp", which reflect an
"osg::Group*", in a
"osgIntrospection::Value", which reflect a "osg::Node*".
This because
the "setName(const char *)" method found by
"grp->getType.getCompatibleMethod"
is an "osg::Object" type method.
When osgIntrospection do this conversion it try :
- to found a "osgIntrospection::Converter" to convert
from "osg::Group*" to "osg::Node*"
- to found a chain of "osgIntrospection::Converter" to convert
from "osg::Group*" to "one or many type" to "osg::Node*"
- to converte an Enum to int or unsigned int
- to convert the value in its "value string representation",
then converte this string in the destination value
Else it throw a "TypeConversionException".
Bug :
1)
When osgIntrospection try to found a chain of
"osgIntrospection::Converter"
It could do any downcast or (Type to SuperType) or upcast
(SuperType to Type).
This mean the the chain could be :
osg::Group to osg::Transform to osg::Camera to
osg::CullSettings to osg::CullStack to
osg::CollectOccludersVisitor to
osg::NodeVisitor to osg::Referenced to osg::Object
During the convertion with this chain, A METTRE failed and
the pointer in
"grp" is set NULL. But the "grp" is always a valid
"osgIntrospection::Value"
and so, osgIntrospection accept the conversion. Then it try
to use this pointer
to call the "setName" function. And Bing SEGFAULT.
2)
In "bool Reflection::accum_conv_path( ... )"
the convection path isn't accumulate in the recursive loop.
this cause multi request of a conversion path, and a
slowdown in the
conversion algorithm.
3)
Use of the last conversion way in a conversion from
pointer to pointer
this mean you can do this :
"osg::Node*" to " value string representation" to "osg::Material*"
What a bad thing !!!
Solution :
1)
Introduce the concept of dynamic_cast and static_cast.
now, to do a conversion, osgIntrospection does this :
- to found a "osgIntrospection::Converter" to convert
from "osg::Group*" to "osg::Node*"
- to found a chain of "osgIntrospection::Converter" to convert
from "osg::Group*" to "one or many type" to "osg::Node*"
only with static_cast, downcast (Type to SuperType)
- to found, if the source and the destination are two pointer,
a chain of "osgIntrospection::Converter" to convert
from "osg::Group*" to "one or many type" to "osg::Node*"
only with dynamic_cast, upcast (SuperType to Type)
- to convert an Enum to int or to unsigned int
- to convert the value in its "value string representation",
then convert this string in the destination value
Else it throw a "TypeConversionException".
Add the "enum CastType" to distinguish the static_cast or
dynamic_cast converter.
Add file OpenSceneGraph/include/osgIntrospection/CastType
2)
add a line to accumulate converter in converter Path.
3)
add a line to check if source and destination are pointer.
"
is not the usual OpenGL BOTTOM_LEFT orientation, but with the origin TOP_LEFT. This
allows geometry setup code to flip the t tex coord to render the movie the correct way up.
Stephan: "attached you'll find some modifications to the GraphicsWindow-class and
their platform-dependant implementations.
The problem:
setWindowRectangle and setWindowDecoration do not update the
traits-object, so, if you call setWindowRectangle on a
not-realized-window it will open with another size when realized later.
getWindowRectangle reports possible wrong sizes if setWindowRectangle
called before.
My solution:
split the implementation in two parts:
GraphicsWindow::setWindowRectangle will update its traits-object and
call afterwards the virtual method setWindowRectangleImplementation
(which is implemented by the derived platformspecific classess). For
setWindowDecoration I am useing a similar mechanism.
I hope you'll find the submission useful, the Win32 and X11 changes are
not tested but should work."
Changes to this made by Robert are call of resized in setWindowRectangle
instead of setting of Traits, and use of a bool return type.
changes I made:
+ put a warning in the console if a nonexistant screen is requested
+ add getters for the aglcontext and pixelformat -- I need access to
them in my own code.
"
the declaring file for a given type via the new member function
osgIntrospection::Type::getDeclaringFile. This information is useful
in order to know what header to include when auto-generating wrappers
for a given type.
During the C# wrapper generator development I've been keeping the
declaring file configuration state up-to-date manually with changes
to OSG, and it's proven to require substantial effort. So it would be
extremely valuable to get this change in before 2.0 to reduce maintenance
during the lifetime of the release. It'll also be equally useful to
others looking to create wrapper generators using osgIntrospection.
This is a fairly simple change and was tested with a fresh rebuild of the
entire suite of osgWrapper libraries, so it should be relatively low risk
(fingers crossed)."
that the terrain is always the Earth planet. I changed the constructor
method to accept the Equator radius and the Polar radius like
parameters. By default, it assumes the Earth radius for the
EllipsoidLocator. I added a setEllipsoidModel method, too.
Now, we are developing some libraries for a GIS applicacion, and our
libraries can visualize terrains of planets like Mars. I think that is
a interesting change."
I added _preDrawCallback member and neccessary access methods plus modified osgUtil RenderStage.cpp to invoke it before all drawInner calls are made. I tried to maintain symmetry with postDrawCallback but you know better where is a proper place for this call ;-)
"
window.
The win32 implementation is still in its original shape since I have no win32
implementation available.
I have chosen the enum approach for the first cut. That is benefitial since
the user does not need to track creation of mouse cursors for different
windows and displays in presence of multiple viewer windows.
The default set of available mouse shapes is the same set that was available
with glut. That set served many OpenGL applications well, so the hope is that
this is enough.
Even though, that implementation is still extensible:
I have digged out the way SDL defines new mouse cursors and added a still
documented out function prototype in the GraphicsWindow that can be used to
extend the current implemtation for arbitrary mouse shapes. That is not
implemented yet.
I hope that somebody with a win32 test system can catch up that implementation
on win32."
The major modification concern the LineProjector class in Projector.cpp. The intersection was previously done in window space, I've modified it to compute it in object space."
"Since we desperately needed a means for picking Lines
and Points I implemented (hopefully!) proper geometrical tests
for the PolytopeIntersector.
First of all I implemented a new "GenericPrimiteFunctor"
which is basically an extended copy TriangleFunctor which also
handles Points, Lines and Quads through suitable overloads of
operator(). I would have liked to call it "PrimitiveFunctor"
but that name was already used...
I used a template method to remove redundancy in the
drawElements method overloads. If you know of platforms where
this will not work I can change it to the style used
in TriangleFunctor.
In PolytopeIntersector.cpp I implemented a
"PolytopePrimitiveIntersector" which provides the needed
overloads for Points, Lines, Triangles and Quads to
the GenericPrimitiveFunctor. This is then used in the
intersect method of PolytopeIntersector.
Implementation summary:
- Points: Check distance to all planes
- Lines: Check distance of both ends against each plane.
If both are outside -> line is out
If both are in -> continue checking
One is in, one is out -> compute intersection point (candidate)
Then check all candidates against all other polytope
planes. The remaining candidates are the proper
intersection points of the line with the polytope.
- Triangles: Perform Line-Checks for all edges of the
triangle as above. If there is an proper intersection
-> done.
In the case where there are more than 2 polytope
plane to check against we have to check for the case
where the triangle encloses the polytope.
In that case the intersection lines of the polytope
planes are computed and checked against the triangle.
- Quads: handled as two triangles.
This is implementation is certainly not the fastest.
There are certainly ways and strategies to improve it.
I also enabled the code for PolytopeIntersector
in osgkeyboardmouse and added keybindings to
switch the type of intersector ('p') and the picking
coordinate system ('c') on the fly. Since the
PolytopeIntersector does not have a canonical
ordering for its intersections (as opposed to
the LineSegementIntersector) I chaged the
implementation to toggle all hit geometries.
I tested the functionality with osgkeyboardmouse
and several models and it seems to work for
polygonal models. Special nodes such as billboards
do not work.
The next thing on my todo-list is to implement
a an improved Intersection-Structure for the
PolytopeIntersector. We need to know
which primitives where hit (and where).
"
implementation of GraphicsWindow:
- usage of WindowData, you can specify an existing window to use via
osg::Traits
- implementation of setScreenResolution and setScreenRefreshRate
- implementation of setWindowDecoration when window is already created.
There seems to be a bug regarding multiple threads and closing windows,
see my other mail on osg-users.
"
selectively set the pixel format for windows that are inherited, following
some discussions on the mailing list last week.
This is implemented through a new traits flag
(setInheritedWindowPixelFormat) with a default state of false (to avoid
breaking existing applications). When set to true, the pixel format of the
inherited window will be set according to the traits specifications.
"
setActiveTextureUnit methods of osg::State so they return false if the
texture unit is outside the range of allowable units for the driver.
Currently, the functions would return true even if the units are
invalid. This would cause the osg::State to become out of sync with
the actual driver state, which can cause some bugs in certain cases.
The change I made would verify that the unit passed to
setClientActiveTextureUnit is below GL_MAX_TEXTURE_COORDS, and the
unit passed to setActiveTextureUnit is below
max(GL_MAX_TEXTURE_COORDS,GL_MAX_COMBINED_TEXTURE_IMAGE_UNITS). I
modeled this behavior from the OpenGL docs for these commands which
can be found here:
http://www.opengl.org/sdk/docs/man/xhtml/glClientActiveTexture.xmlhttp://www.opengl.org/sdk/docs/man/xhtml/glActiveTexture.xml
"
Most of the code is from the osgviewer application, I have took the fullscreen handler and the threading one, and I have just added a fonctionality to be able to change the screen resolution in windowed mode."
"Attached are updates of src/osg/Sequence.spp and include/osg/Sequence.
I've taken _sbegin/_send/_ubegin/_uend and _step our of the include file
and made them local variables in whatever method might need them.
I got rid of the _recalculate method as it was only getting used in
one place.
I also found a cut/paste bug in setMode's START case."
Note from Robert Osfield, Also includes some guards against crashes that was occuring in this new
code when handling empty Sequences.
"By repurpose, I'm creating a new plugin that uses much of the .osg fileformat,
but with some changes. Specifically, I'm creating a ".osgfs" plugin, which
represents the scenegraph hierarchy as a filesystem of nested subdirectories and
individual files for each node, rather than nested braces with everything in a
single monolithic file. I intend to incorporate file alteration monitor events
to watch the filesystem for modifications and automatically reload.
The problem I'm running into is osgDB is too tightly coupled to the .osg format.
osgDB::Output::writeObject() contains literal .osg format-specific strings like
"{" to represent the Object hierarchy at too low a semantic level. I propose
using virtual methods; my plugin can then derive from osgDB::Output and
represent Object hiearchy differently.
"
~~~~~~~~~~~~~~~~~~~~~~~~~
This is a simple change to permit databases other than those named
"*.osga" to be used. It is hardcoded in read() at present.
It is non-critical and does not affect existing program functionality.
Registry and Registry.cpp
~~~~~~~~~~~~~~~~~~~~~~~~~
Added a new typedef: typedef std::vector< std::string>
ArchiveExtensionList;
a list of extensions: ArchiveExtensionList _archiveExtList;
and an "add" method: addArchiveExtension(const std::string ext)
This is initialised by adding "osga" in Registry() and used in
Registry::read() where the list is searched for the extension used.
Archive.cpp
~~~~~~~~~~~
This submission is a little more tentative. openArchive() is modified to
automatically add the filename extension to the Registry extension list.
"
return the length of the stream.
Implemented the virtual methods in QuicktimeImageStream, (getLength,
getReferenceTime, setTimeMultiplier), to return valid value for each.
"
returned from Viewer::getContexts/getWindows will be the left most window on the lowest screen number.
Added ability for StatsHandler and HelpHandler to support end users setting their
Camera's graphics context.
it to be assigned as a vertex attribute array to an osg::Geometry.
Removed the osgTerrain::ArrayLayer as its no longer required thanks to the above change
which makes the osgTerrain::HeightFieldLayer more flexible.
Updated wrappers
Lost the functionality to find the real type pointed by a pointer.
Ex: a osg::Node pointer point on a osg::Group, if I look for information
on the pointer type, the introspection say it is a "osg::Node*".
But if I want information on the pointed type,
the introspection must return the "osg::Group".
This bug come from the osgIntrospection::Value::Ptr_instance_box::ptype() function.
In the original version, this function use the member "Instance_base *inst_"
like this :
typeof(*static_cast<Instance<T> *>(inst_)->_data)
But in the new version, this function use the template argument "T":
typeof(typename remove_pointer<T>::type)
This is a good meta-programming use, but here we need a dynamic request.
Moreover the "typeof" macro define in "Reflection" header accept only a type in parameter with the new version.
fix:
Add the macro "typeofvalue" in "Reflection" header which accept a value or a type in parameter.
Restore original code in osgIntrospection::Value::Ptr_instance_box::ptype() function.
"
multithreaded-opengl-engine on os x or not. I set its default to false,
perhaps other os x users can test this setting with their data/apps, to
see if we can enable it by default.
I changed also the borderless-window-type, so expos?works correctly."
class to encapsulate the pixel coords, SceneView and picking operations in prep for
making the code more general purpose, and less reliant on classes like osgUtil::SceneView and osgUtil::IntersectVisitor.
Vivek's email to osg-submissions:
"I'm happy to release the osgdragger nodekit to the OSG community. I
implemented the nodekit for my company, Fugro-Jason Inc., and they
have kindly agreed to open source it.
The nodekit contains a few draggers but it should be easy to build new
draggers on top of it. The design of the nodekit is based on a
SIGGRAPH 2002 course - "Design and Implementation of Direct
Manipulation in 3D". You can find the course notes at
http://www.pauliface.com/Sigg02/index.html. Reading pages 20 - 29 of
the course notes should give you a fair understanding of how the
nodekit works.
The source code also contains an example of how to use the draggers."
To set up extension aliases using a config file, an app calls:
osgDB::Registry::instance()->readPluginAliasConfigurationFile(), passing in the file name as the parameter. (Of course this should be done before loading any files whose names depend on the mapping.) osgDB will search for the file using OSG_FILE_PATH.
The file should contain a line for each mapping, with the "map" extension first, followed by a space or tab, then the plugin identifier. For example, a file containing this line:
flt OpenFlight
would map the ".flt" extension to the OpenFlight plugin."
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.
on the screen, it looks more aesthetically pleasing to have a larger
gap between lines than is given by default. I added a new parameter,
lineSpacing, in the Text class to allow the line spacing to be adjustable
by the application. The default value is 0 meaning there is no extra
spacing given. The value should be given as a percentage of the character
height. A good value for longer paragraphs is 0.25 (25%) or more."
copy constructors that take a const osg::CopyOp rather than a
osg::CopyOp&, forcing an unnecessary copy. The attached header fixes
this, based off OSG 1.2.
Also fixed duplicate "or" openArchive()'s comment."
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.
