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
created local to RenderStage and passed on to rendering code which populated
the RenderInfo UserData, but without the restoring the new UserData to the
main RenderInfo. The local RenderInfo UserData is now passed back to the main
RenderInfo.
Under this mode, each 4x4 texel block can selectively use an alpha component or none. When alpha-enabled blocks are present, this is not reported in the DDPF_ALPHAPIXELS bit in the pixel format flags causing the reader-writer to report the file as a 3-components file (GL_COMPRESSED_RGB_S3TC_DXT1_EXT). The fix requires looking at each 4x4 texel block to detect the presence of 1-bit alpha encoding. When such a block is found, the internal & pixel formats are reported as GL_COMPRESSED_RGBA_S3TC_DXT1_EXT instead.
See http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/Opaque_and_1_Bit_Alpha_Textures.asp for more information."
following improvements:
- When all unsupported entities (SOLIDS, TEXT, ...) have been filtered out
from a block, dxfInsert::drawScene() was crashing because it assumed that
the block was not empty. It now returns silently so the other blocks can
still be imported.
- The DXF reader assumed all vertices were read first, then the face
indices. But, from the DXF "documentation" (www.autodesk.com/dxf) :
"Polyface meshes created with the PFACE command are always generated with
all the vertex coordinate entities first, followed by the face definition
entities. The code within AutoCAD that processes polyface meshes requires
this ordering. Programs that generate polyface meshes in DXF should generate
all the vertices then all the faces. However, programs that read polyface
meshes from DXF should be tolerant of odd vertex and face ordering."
So now the importer ignores the posted number of vertices and face indices,
and uses the size of the lists instead."
- does proper edge detection with the crease angle given in the ac file
- the reader is much more verbose in case of broken files
- no more static variables
- std::stream based
- much faster, up to a factor of 50 for a polygonal test model
- if the file contains polygonal surfaces the loaded geometry is still
polygonal as long as the polygons are convex. Concave polygons are still
tesselated. Rationale: If the user needs triangular models he can throw away
that extra information with the osgUtil::Optimizer himself. But the read step
preserves as much extra information that is contained in the file as
possible. Also the polygonal models look a bit smoother than the triangular
ones.
- a huge cleanup
- the basic parsing logic from the original ac3d developer is untouched
- the writer part is untouched
The tarball contains a replacement for the src/osgPlugins/ac3d directory that
is based on the current CVS (2006-11-13).
I have tested that loader with a osgconv foo.ac /tmp/xx.ive on everyfoo. ac
file I could find (~200 from flightgear and the object database). I also run
my flightgear development with that loader -> here I see the models.
I run valgrind testcases on some selected models."
