Is convex hull phantom?

[Pamela Galli]

I tried making my floors and exterior walls convex hull, but they are phantom. I thought convex hull meant not phantom, and holes filled.I thought "none" meant phantom.

 

Most of the time when I change physics type I see no change in the properties of the mesh or the LI.

 

 

[Drongle McMahon]

Physics shape type and Phantomness are independent. You can set an object to phantom whatever its physics shape type. If you can walk through a wall when it is physics shape Convex Hull and has phantom unchecked, then something very strange is going on. What do you see if you check Develop->Render Metadata->Physics Shapes? Physics shape None is not quite the same as Phantom. I think Phantom objects still collide with the terrain, while None objects have no physics at all. being completely ignored by the physics engine. Phantom is a property of the linkset*, while None is a property of a single prim.

*hacks apart.

[WADE1 Jya]

I haven't used mesh very much in SL, but I am using "convex hull" meshes in the design projects I am now busy with outside of SL.

Convex hull should in no way be equivelent to 'phantom' :catsurprised:

Outside of SL, convex hull is a physics simplification of the mesh to a closed 255 tris shape which can then interact physically with other mesh objects. It only works for shapes 255 tris or less.... SO that might be your problem.

It is a gross simplification too, gross (not as in disgusting :catvery-happy:) as in reduced bigtime.

So a wall or a floor might collapse into a funny geometry you wouldn't expect.

It will create an invisible geometry for physics which you can not see, and that shape may or may not align with the rendered shape you can see, which is the other possible problem.

To correct this issue, make a reduced version of your mesh for physics (in SL I know you can asign a different mesh for physics during upload), or try just reducing the tris of your mesh if that works easier for you. Definitely make sure you optimize your mesh to get rid of stray points & bad seams which could be heavily distorting your physics geometry.

For the environments I'm in, that is how it works anyway. Of course, Lindens might have twisted the original meaning of 'convex hull' to signify something else entirely in SL, or you might be encountering an SL-specific bug.

Hope that info helps, or if not that someone more actively working away in SL comes along to explain any further particulars

[Innula Zenovka]

As I understand it, you need to set the whole object to not-Phantom, set the root prim to PRIM_PHYSICS_SHAPE_CONVEX and then set  to PRIM_PHYSICS_SHAPE_NONE the parts you want to be able to walk through.

 

[Pamela Galli]

Thanks everyone! When SL is open again I will try what Innula says.  I did the opposite, turned everything but the root to None, then tried setting some walls to Convex Hull. That did not work.

[Innula Zenovka]

I've just tested it on Aditi (since I couldn't get into the Main Grid) and it seems to work on a simple prim build as I describe. 

Best of luck!

 

[Pamela Galli]

Nope did not work on my house.  I made it all ConHull, then went around and made walls with doors None.

I have two ConH walls that I can walk into but not through; the rest I can wall through.

All the floors are ConH and phantom.

I notice in the bedroom, with phantom walls and floors, there is a place in the middle of the room that is blocked, like with a ghosted prim.

 

 

[Pamela Galli]

Maybe once you change a mesh type, you can't change again?

[Drongle McMahon]

I am having some difficulty picturing what you describe. Did you upload several separate meshes in one upload with a separate physics shape file? There can be errors in associating the right visual meshes in that case. Otherwise, I think I would need to look at the house to make more sensible comments.

[Pamela Galli]

It is separate meshes uploaded together, generated physics shape.

 

I sent you a copy of one of my floor pieces, convex hull, phantom.

 

It is not a big deal if the whole house has to be phantom, I can just put prims in the walls. Just might save a few prims if I did not have to.

[Drongle McMahon]

I inspected Pamela's floor mesh and found that its default convex hull shape was a single triangle, the same triangle as the lowest LOD. This surprised me, as the last time I investigated this, way back in the beta, the default convex hull was made from the high LOD mesh. There were good reasons for this, among which was the very unexpected behaviour Pamela describes here when lower LOD meshes were used. In my view, reversion to using the low LOD is a mistake.

So I did some experiments to find out the current situation, using the current release viewer, Second Life 3.3.1 (254524) Apr 24 2012 06:28:07 (Second Life Release). These showed that (1) if you do not specify a physics shape mesh, the default convex hull (the shape used when you set physics shape type to convex hull) is the convex hull of the low LOD mesh. (2) If you do specify a physics shape mesh, the default convex hull is the convex hull of the supplied physics shape mesh.

So, Pamela was using the automatic LOD which reduced the low LOD to a single triangle and ended up with the default convex hull being that triangle. This was, of course, completely useless.The solution in this case is to use the high LOD as the physics shape mesh*.

I will eventually make a jira asking for the return of the old behaviour to avoid this sort of trap for new users. Before doing that, I wil start a thread here to see whether people agree with me that the current behaviour is unacceptable. The default convex hull should be one that gives the sort of behaviour people might reasonably expect. This is presumably motivated by a desire to avoid over-complex default hulls. In that case, it should make an intelligent decision based on the complexity of the hulls derived from each LOD mesh. In this case, it should obviously choose the high LOD hull, which has only eight points.

*although there is still an outstanding bug which makes the default convex hull smaller than the supplied mesh. So for an exact shape, it is presently necessary to use the Prim type as well as providing the mesh and decomposing it.

[Drongle McMahon]

Oh dear. Getting stuff together for the jira etc., I found another problem. The physics weights for convex hulls with anything but the simplest shapes are substantially higher than given by the calculation shown in the wiki. For convex hulls, ignoring the terms for perfect shapes and triangles, this should be 0.2*H + 0.04*( V - 4*H) where H is the number of hull(s) and V is the number of vertices in the hull(s). This is simply reduced to 0.04*( V + H ). For very simple hulls, this gives the weight exactly, as for cylinders with 5-10 verts around each end. As soon as you add another ring of vertices and expand it to male a barrel shape, the physics weight shown in the uploader (and on inspection after upload) is higher than given by the calculation. This is taking into account that even the single the hulls produced by the decomposer by using simplify sometimes have more vertices than the meshes they are made from (Ugh!!). This will have to be another jira. Too tired now.

[Pamela Galli]

Thanks very much for looking into this Drongle.  Both of these bugs seem pretty significant to me. Frankly I am not surprised -- there is just too much weird bugginess with the uploader. Keep LI down is too difficult for a number of reasons -- one of which has been scripts, but it looks like they are going to rectify this.

I have spent months turning out mesh, and it is disappointing to think that some of the compromises I was forced to make really might not have been necessary. Very disappointing.

 

One can only hope that they will step up quickly and take care of these bugs.