> J?r?me Grimbert wrote:
>  
>> I wonder if there is not a bug in the formula when r is k.a1 or k.a2 (but
>> maybe this is already tested, and the correction is obviously to replace
>> the null vector with the cross product of r and the other new axis)
>> 
>> Alas, I do not have the code at hand right now...

> Well, if anyone has interest about that patch, I see two bugs and more
> in the code of megapov 0.7 :

>  - first, minor impact: the length of the second vector is lost
>    the oriented area light is a square (or circle if "circular" is used)
>    (Looks like a copy-paste of code was made too quickly)   

>  - second, crash expected: the result of the cross-product is not checked
>    against the null-vector and there is a macro that will normalize 
>    the length of the new vector; if the new vector is null, there
>    will be a division by 0... With float values, the chance are low
>    that this will happen, but they still exist.

>  - more: I do not understand why the code wastes some time to
>    first invert (scale -1) the ray direction, do its calculation and
>    then invert again the ray direction.

Thanks for the info.  This is quite helpful to me, as I've often wondered
about this ever since I encountered it several months back.  Hopefully some
of the 3.5 team will fix these issues (are the area_light changes in 3.5?  I
assume so, since they've been around awhile, but I don't know it)

Geoff

Post a reply to this message