Drongle McMahon

In the past I have had problems with particular textures that were not resolved by clearing (or deleting) the cache, but were resolved by deleting the user-specific settings.xml file. Some of the settings do affect the cache - so I assumed that was the problem. Of course you do loseALL your settings if you do this. It gets replaced from the settings.xml in Program Files. Don't delete that! On W7 the one to delete is in Users/username/AppData/Roaming/SeondLife/user_settings. For safety, just rename it instead of deleting it. Then you can get it back if you need to.

Ah yes. I'm always getting that the wrong way round. (It gets much worse when you use materials!). Black is transparent and white is opaque. S in this case the parts where the shiny has to show through are black. Here is the alpha channel of the texture I used. The colour channels were all black. It's a from a fractal, by the way.

Here's the sort of thing you can get with the (experimental) materials viewer. I'm a novice using it, so expect others to do much btter.

I haven't tried, but it should be easier to set up in a mesh than with prims. You would just need two very slightly separated faces with thier normals pointing the same direction. Getting at the textures to edit them in SL might be awkward unless you make sure a small piece of the underneath one is accessible somewhere where it won't be visible. As for doing it in Blender, I don't think I would want to. You can set up something with an image using alpha influence to see that things are arranged properly, but it won't look much like it will with SL shiny. Blender is, of course, capable of making the desired effect in one material without needing the overlayed surfaces. Yo be quite honest, I would wait until the SL materials stuff is released (real soon now?), with proper specular and environmental lighting maps, and not bother with this hack.

You may if you want to. It may well be that the type of thing will not work in the conditions where it is needed. It It's not very good anyway. Everything will be better with materials (we hope!)

Here is a linkset of two cylinders at the same location. The inside one has dimensions 0.005 less than the outer one, and is a solid black colour on the blank texture with shininess set to medium. The outer one has an alpha texture with appropriate repeats. Alpha textures can't be shiny. This is with advanced lighting, but it works without it too.

Material #1: Faces = 4,408; Material #2: Faces = 960; Material #3: Faces = 4,104; Material #4: Faces = 96; Uploader traingle count on High = 150676 (before adjusting/dropping it down at all) That seems very, very high. It would appar that those are the counts before subsurf is applied. Each level of subsurf will multiply the triangle count by three or four fold. So you very quickly get excessive counts. Two levels could take this over 150,000. Try to rely on smooth shading rather than subsurf to get smoothness except at visible edges. From the 150,000 in the upload, I think this is certainly the problem. When you have one material, it gets to 174,752 before it start culling triangles, but each extra material will have a list of triangles that has unused capacity, so that the ninth pseudo-material get started earlier. That space could be could up to 20k per material. So the number you see is entirely consistent with having too may triangles. Incidentally, flat shading will also cause multiplication of the vertices, as they have to be stored with each different normal they are used with. With smooth shading they always have the same normal in all the triangles they appear in. Highly fFragmented UV maps have similar effects. These effects can easily treble the triangle count.

"... for example how far away should the boxes be?" The boxes are when you are going to use the "Analyze" button which turns your mesh into a set of convex hulls. The reason it helps to keep them simple convex shapes and separated by small spaces is that it helps to get the best possible hulls. When the uploader has to cut up convex shapes, or overlapping shapes, it often does it in an inefficient or obstructive way. If you have to use the simplify function, because the shape mesh is too complicated, that is even more likely to produce poor results. The gaps can be very small. "...Is it more important to keep it simple with boxes than to simply have the lowest tri's like in LOD versioning?" See above for "Analyzed" shapes. The physics weight depends on the number of hulls and the number of vertices they have. This can be very high if you have many convex details. So you still need to keep it simple. It's better for the engine too. "... Havoc loves boxes so I am guessing that's the rationale behind this? " It loves boxes when it knows that's what they are, but the uploader doesn't recognise boxes and tell havok that they are there (collada has that facility, but it isn't used). So uploaded boxes are treated like any non-box convex hulls. More efficient than the same shape in triangles, but less efficient than when Havok knows to use it's own primitives. When you use undistorted linked prim boxes as the physics shape, on the other hand, these are recognised and the primitives are used. "... what about materials and UV's, ... does this apply to the physics only model? " No. (It used to, but that was a bug fixed long ago). "... can I simply construct a fresh model without any UVs or relation to the main model or the LOD's as long as the bounding box fits and it works as it should as the physics model? " You can use anything you like, as long as it doesn't have degenerate triangles (too small in some direction to still have an area after rounding to 16-bit coordinates). Whatever its bounding box, it will be stretched so that it becomes the same as the bounding box of the high LOD mesh. The reason we say use the same bounding box is that this is generally neeeded to make sure the parts of the physics mesh will fit correctly to the visual mesh.

It may be too many triangles. Can you do the following on a copy - in Blender, select all and triangulate faces, then select each material and record the face count (which is now triangles) - then record the triangle count you see in the uploader. Tell us these numbers and we will see if that could be the explanation. The uploader constructs lists of triangles for each material and starts a new secret material when it gets to 21844 triangles. So it can then see more materials than were in the collada file. If it gets to more than 8 materials, it simply discards the rest without telling you (this is a reported bug).

The normal effect of uploading a collada file with more than 8 materials in one mesh is that the triangles for the 9th and subsequent materials are simply omitted from the upload. From that point, it behaves as if it had 8 textures; so you can use a medium LOD with just 8 textures, but not 7.

That's what I was thinking. But how does the Frenchified scanf read the english decimals? In the French system, the point is a thousand separator, n'est ce pas? Maybe it has some clever way of recognising the difference that doesn't quite work all the time?

But which form of numeric decimal point is used in your collada files?

"how can I animate a normal map??" At least this morning it just did the same thing as the texture animated on the same face(s). That was... llSetTextureAnim(ANIM_ON|SMOOTH|LOOP,ALL_SIDES,0,0,0,0,-0.015 ); No guarantee it will stay that way, of course. It would be more versatile if it could be distinct from the texture animation. As it is, the waves and the diffuse map have to move the same speed and direction. I'll leave my sea out on Mesh Sandbox 3 for a bit, so you can look at it. The texture isn't seamless, so it's not reall acceptable quality!

In the materials viewer you can get a nice seething wavy pattern using an animated normal map on a mesh with displaced vertices. Unfortuantely. you still can't make it transparent without losing the specular reflection. I'm not sure if that's going to be available eventually. The mesh I tried was based on a 16x16 square mesh and was14LI for a 64x64. So that's 224 for a whole region ... not too expensive. Hopefully the wait for materials might no be too long.

:matte-motes-confused: That's even more wierd than it was before.

Unfortunately, nobody but Lindens, and a very small number of others, is allowed to see your jiras. Good luck with it though.

Not really enough information here to know what the problem is. Can you give a more detailed description of how you "applied physicsto the mesh" and what you maen when you say the physics is s separate prim? Also, if you can, make a screenshot of the physics shape after Develop->Show Metadata->Physics shapes.

Assuming you mean joining the two objects in Blender, compared with uploading them as two Blender objects, then yes, that is true. The reason is that the LOD switch distance is determined by the size of the whole mesh prim (object in Blender). So the small detailed mesh part will be downloaded and displayed for cameras in a much larger area. The download weight calculator knows this and takes account of it in the calculation. using separate objects is a way of dealing with the mixture of different scales of detail like this. The other thing you can do is make much more drastic reduction of small details than of bigger details in the medium LOD etc. That will be rewarded with a reduced weight too.

"These textures can not be reused in Second Life because you have no access to them from your Inventory." I do, at least in the standard viewer - they appear in a subfolder of the Textures folder with the name of the object. So for an object called "thingy" it's in Inventory/Textures/thingy/thingy - Texture 1; .../thingy - Texture 2; etc. If you upload the model again, you get another subfolder with the same name. Another problem with including textures in uploads is that if you always do it, you get to pay the L$10 per texture every time you upload a new version of the mesh, even if the textures haven't changed. Also, you pay again and litter you inventory if you use the same texture on multiple models.

Do you by any chance have another copy of the door object with the same image in the UV editor but located far away from any occluding objects? If so, that could be overwriting the baked occlusion. It usually gives more obviously garbled results though.

This is very confusing. As far as I was aware, the server should have no effect on the preview because the server doesn't get any information about the mesh until you click the Calculate button! How can it be influencing a process it doesn't even know is happening?

You can apply solidify to just a selected part of an object by using Mesh->Faces->Solidify (Ctrl+F,S).

"...do the p shapes meshes need to be selected in the same order the visible object meshes were selected?" I'll explain this in more detail... The thing that matters is the order the c<geometry> sections appear in the collada files. They must be the same, so that if it's A, B, C in the visual mesh, they it must be A, B, C in the physics mesh. Otherwise the uploader will attach the wrong physics mesh to each visual mesh. The selection order in Blender does not control the order in the exported collada. So it doesn't work. Instead, you need to name your Blender objects so that they are in the right order alphabetically, then check the "Sort by Object Name" option in the exporter (it will automatically be checked if you select the SL Presets). The same applies to the LOD meshes if you are supplying those. This is most easily done by adding a postfix to the same names for the visual and physics objects.So, for example, the names might be "thingy", "rubbish" etc..., for the high LODs (no postfix here because this becomes the object name in SL); then the LOD meshes are "thingy_med", "thingy_low", "rubbish_med", "rubbish_low", etc., and the physics meshes are "thingy_phys" and "rubbish_phys". Now when each set of objects is exprted with the "Sort by Object Name" option, they will automatically appear in the right order in the collada files.

The LI is the highest of the three weights, download, physics and server. So if you download weight is the highest, then reducing the physics weight will have no effect on the LI. It;s still a good thing though, as it means you are asking for less work from the physics engine.

Did you set the physics shape type to "Prim" on the Features tab of the edit dialog? If not, it will use the default convex hull which is always present whether you provide a physics shape or not. That wil block access to the the concave surface of the wall.