[ODE] Triangle-box collider in current CVS version

[Adam D. Moss]

Shaul Kedem wrote:
>  Starting the test from scratch, with no boxes, the
> first couple of boxes I drop go into infinity. so
> basically there are no boxes to stack and so this
> doesn't happen. 
> 
>  On the other hand, I might be doing something wrong,
> so if anyone has a linux machine , gcc 3.3.2 and the
> cvs version I'd be happy if someone else will check
> this issue as well,

Confirmed, things work well on gcc 2.95.x but go to crap
on gcc 3.1.1.

However, I don't think that it's a triangle-box specific
problem, because trimesh-trimesh also fails miserably
under gcc 3.1.1, so I'm guessing it's an OPCODE issue --
and now that I think about it, I seem to recall a
GCC optimization bug being discussed on this list w.r.t.
Opcode some time ago.  The solution IIRC was to hand-compile
one of the OPCODE modules without -O2 optimization.

*investigations*

http://q12.org/pipermail/ode/2003-July/009265.html

For me, that fixes trimesh-trimesh, but leaves tri-box
broken.  However, compiling /all/ of OPCODE with -O1 fixes
trimesh-box too, under gcc 3.1.1.  Someone keener might
care to isolate the other specific module(s) which impair
tri-vs-box's effectiveness (and someone super-keen might
even track down what it is that OPCODE's doing to upset
GCC 3.[123] and whether we can work around it).

--Adam
-- 
Adam D. Moss   . ,,^^   adam at gimp.org   http://www.foxbox.org/   co:3