1 issue was caused by detecting 2d horizontal line intersections for lines that had points equal to the horizontal Y value - solved by detecting point on line cases.
Another was because the 2D bounding box for painting could have faces edges running along it - solved by adding a small margin to the bounding box.
* add support for building redcode on win32/msvc, but disabled for now, as there are linking problems
- I cleaned the redcode sconscript - the copying of headers within the source tree is not a clean solution
This needs to be fixed later on. For now, lets use redcode from extern/ until a better way is found.
Python dict error: when trying to access a Bone via a key, and the key
was not found, a wrong error message got printed. Fix provided by
reporter Gregor Riepl. Thanks!
Use stroke normal to align roll of the retargetted bones.
Unlike stroke conversion, the final roll is not forced to point exactly to the normal. Rather, It tries to align the X or Z axis of the bones (whichever is closest) to the stroke normal. Forcing the roll can invert knees while this rightly only makes one of the joints rotation axis perpendicular.
When entering a wrong expression (or garbish) in a Nkey panel button,
the cursor jumps to the place where the button was clicked. On failure
the button could return a B_NOP, not a 0
* When making a proxy, the lib linked IPO driver was also changed to
point to the proxy object, and after undo this local proxy object
was replaced so the pointer became invalid. In fact it is not needed
at all to change this because the IPO code maps the pointer to the
local proxy object already.
* Undoing the make proxy operation would crash because the proxy_from
pointer in the library linked object would still point to the removed
object. Now it clears all these pointers before undo, because on each
undo memory file read they will be set again anyway.
Bah... fix for envmaps just before 2.48 release gave good looking envmaps
only when there was no sky involved...
The alpha in environment maps should be reset to 255... something that was
never done before, but also didn't show errors until other fixes in image
rendering were done.
* The second opengl texture coordinate (gl_TexCoord[1]) are now filled
in as well, and will give canvas coordinates from 0.0 to 1.0. The
first texture coordinates still give the coordinates in the texture
that is being used, which may not match the canvas exactly, so both
coordinates are needed.
* Also optimization to allow using smaller texture sizes with multiple
smaller viewports.
* Print the detailed GLSL shader errors (once), for easier debugging.
