[ODE] Problems with 0.039 and trimesh

[Flavien Brebion]

First of all, i've build my MSVC6 workspace and makefiles from
scratch, so i don't know if that is a problem but the behavior
"they just head straight through the ground and fall to infinity"
makes me believe you haven't enabled the flag "dTRIMESH_ENABLED".
By default in 0.039, tri-mesh collisions are disabled.

I have no idea about the polygonal garbage, but i'd first check
if it's a problem with drawstuff or incorrect data given to
ODE/Opcode.

My experiences so far with ODE and tri-meshes are good. There is
no performance problem with large meshes (i currently use a
tri-mesh of 130000 polys for my quadtree terrain), and although
i exclusively work with OBBs, i haven't seen any problem (except
my recent "tunneling" problem, which is still here. But that's
another story). However it's important to note that in 0.039,
the tri-mesh bounding boxes are not calculated (left to infinity)
which lead to performance problems with lots of tri-mesh objects.
I did implement a better AABB calculation, although it does not
handle rotations yet. You'll probably have to do the same if
performance becomes a problem. Now my collisions usually take
less than 1 ms with one terrain (130000 polys) and about 600
boxes in the scene.

F. Brebion


-----Original Message-----
From: ode-admin@q12.org [mailto:ode-admin@q12.org]On Behalf Of Thomas
Harte
Sent: Wednesday, July 30, 2003 12:23 AM
To: ode-list
Subject: [ODE] Problems with 0.039 and trimesh


I first built ODE 0.039 exactly as I had 0.035 and 0.03 before. No
problems - all binaries in ode\test appeared okay and things were rosy. Then
I decided to have a go with the new trimesh features, so I built opcode
using the provided MSVC workspace (although I had it put the lib files into
the same directory as the workspace rather than in "..\..\lib" which doesn't
even exist in the hierarchy on my computer), uncommented the
OPCODE_DIRECTORY line in user-settings and ran make again. The result is two
further executables in ode\test - test_trimesh.exe and
test_moving_trimesh.exe. Neither of these work.

test_trimesh displays a sort of skeletal satellite dish and is otherwise
like test_boxstack with one very important difference - whatever objects are
dropped, they just head straight through the ground and fall to infinity.

test_moving_trimesh displays a screen of polygonal garbage with the
occasional section of floor showing. I've uploaded a screenshot here:
http://www.btinternet.com/~t.harte/trimesh.png

My compiler is MSVC 6, and I get the same results under both Windows 98 and
2000 - both of which I have installed on my machine.

Am I doing something dense, or do my experiences reflect the maturity of
bundling trimesh to ODE? Assuming I can ever get the thing to work, from
where can I get a manual for trimesh? The only mention in the docs is as a
return result for dGeomGetClass.

My setup is that I have a quadtree and I wish to collide boxes and capped
cylinders against this - am I right in thinking that trimesh can help me
when combined with the contributed capped cylinder class?

-Thomas

Film & TV Extras urgently required in your area - See Yourself in major
Films & TV? Call 0907 1512440 to Register. calls cost 150pm