[ODE] Trimesh collision: not all triangles are colliding

[Frederic Marmond]

In fact, I don't know.
We tracked the bug with others (one who ported opcode to linux (W.P. van 
Paassen) and Pierre Terdiman, the Opcode creator) and we 'only' found a 
little hack to avoid the problem: "-O1" for OPC_TreeCollider.
But I stoped playing with that few weeks after, so, i don't know its 
current state:
- has opcode evolved?
- has the linux portage changed? (I made my own, with the same problem)
- is this bug removed with a new version of gcc ?

I haven't time right now, but i'll check with a sample prog I made to 
spot this problem, to see if it still the case.
See the post I made called "gcc 3.2 compiles fine OPCODE but errors at 
runtime. The solution here..." in this list the last jully 7th.

I'm back in 2 days, and i would have much time to re-investigate if it's 
needed.

Fred

Adam D. Moss wrote:

> Frederic Marmond wrote:
>
>> Linux?
>> If yes, I found a bug in the OPCODE library, that forced me to 
>> compile the collision part without optimisation for a file (-O0), on 
>> gcc-3.x compilator. Worked fine (-O3 for all files) with a gcc-2.95 
>> one. The result was as you said, some triangles collides, others 
>> don't as they would have been colliding.
>
>
> Was it really a bug in the OPCODE library?  If so, it would
> be very helpful to find and fix the bug.  But as you describe
> it, it sounds equally likely to be a GCC bug (though again,
> if it can be isolated we could probably make a workaround).
>
> --Adam