Am 17.01.2014 22:16, schrieb Cousin Ricky:
> Blurred reflection doesn't work on some surfaces parallel to the in the x-y
> plane.  (It's a problem for planes and boxes, but not for cylinders.)  The
> following scene file illustrates the problem.  The cube on the right has been
> rotated ever so slightly, and the problem does not show up.

That's seriously odd; for the blurred reflections algorithm it shouldn't 
make any difference which type of primitive is used - all it knows and 
cares about is the surface normal.

The only conclusion I can draw from this (after having confirmed your 
observations) is that there are two issues involved here:

(1) A flaw in the box code that makes it return a flipped normal for the 
x-y faces. (This isn't strictly a bug, as the code downstream is 
supposed to cope with flipped normals.)

(2) A bug in the blurred reflection code that makes it break down if the 
normal is either +z or -z and points the wrong way round. (Most likely 
the bug is not in the blurred reflections code itself, but auxiliary 
code that is supposed to flip the normal as needed.)


(BTW, experiments show that cylinders /are/ affected as well, but only 
at one end.)

Post a reply to this message