I'm building the animation system of my game engine (the skeletal and skinned animation stuff), and I came to a point where I added so much functionality in the frame and node structures that it seems to be excessive for some applications, and it ends up adding too much complexity when working with simpler objects. Objects like boxes or simple meshes that are meant to be static objects would end up with complex properties such as the ability of having a node hierarchy and transformation frames, animation type checks, and also require a lot of unnecessary creation parameters that make really hard to work with when coding some functions that use them.

I was thinking of making ie. SimpleMesh and HierarchyMesh objects, which will also require that the renderer can deal with different types of objects in the same scene. I was also thinking about making a MeshNode class and then make a Mesh object that contains them, but then I have some conflict on where to store some data, as some meshes will have per-node shaders and other will have a single shader for the whole mesh, then storing the shader references in the node will be a redundance for complex meshes that share the same shader in every node.

Other option I was thinking was making some helper functions to deal with the simpler cases, which would set some default parameters and would ask only the most basic ones, for example

CreateSingleFrameMesh( pTextureName, pBuffers, nBuffers, Material, &pCreatedMesh ); 

but I'm not sure if this would be too time consuming and require too much memory for storing something that will be default values in most of the scene objects (I believe the scene won't be built from skinned hierarchy meshes but mostly static and even in some cases the animation is purely a shader thing which renders useless most of the animation property in the c++ side )

Also files are required to store the minimum data required to rebuild the mesh, so in smaller meshes or meshes without animation I would end up storing a lot of data that I don't really need, even if I'm indexing node and frame data when saving and then store the hierarchy with the indices to the actual data.

I don't know if my speculations are ok, as I don't have much experience with 3d animations yet. I want to make a well decision as any option I choose would require a lot of work to get it to render and I don't want to find out in the end that I have to rewrite everything again, as a lot of other objects will be working with these data.

Sorry if it's a too subjective matter, but I need some insight. I'm full of questions about this..

Thanks in advance.

EDIT: I'm gonna add some code about it

Yes probably I'm mixing up things, but it's complicated because there are a lot of data:

    INT32           nIndexBufferID;
    INT32           nVertexBufferID;
    UINT32          nIndexOffset;
    UINT32          nVertexOffset;
    UINT32          nIndexCount;
    UINT32          nVertexCount;
    UINT32          nTextureID;
    INT32           nImageID;       // reference the hardware image for this node (a video device texture)
    IMAGE_USAGE     nImageUsage;    // the way each image is applied in the node (color, bumpmap, etc)
    Vector2         vImageOffset;   // the offset for each image applied in the node (in pixels? o.O normalized coords?)
    SPRITE_DATA*    pSpriteData;    // Set to NULL or point to sprite-like animation data.
// This structure holds data about properties of an animation frame 
// Note that all these data may vary between frames, 
    UINT32                  nFrameID;       // Frame identifier
    float                   fTime;          // this frame starts at the given fTime (or lasts fTime, i'll see that later)
    Vector3                 vScale;         // this is interpolated and then applied for each node before the global transform
    Quaternion              qOrientation;   // this is interpolated and then applied for each node before the global transform
    Vector3                 vPosition;      // this is interpolated and then applied for each node before the global transform
    MATERIAL                Material;       // the base color of the object (set to 100% white to just use the texture color)
    float                   fReflectionAlpha;// The alpha value of the reflection (a transition of this value can be used to achieve some effects)
    bool                    bHidden;        // Set this variable to true to skip rendering this frame
    bool                    bAlpha;         // Set this variable to true in order to enable use of the fAlpha variable (otherwise it's gonna be discarded)
    float                   fAlpha;         // Set this variable in order to change the global alpha value for this frame
    LIGHT_TYPE              nApplyLights;   // Set the light types that will be processed for this node. If it's set to GLIGHTTYPE_NONE, the node will rendered with raw vertex and/or texture colors
    UINT8                   nBufferCount;   // The amount of buffers used by this node
    GEOMETRYBUFFER_DATA**   ppBufferData;   // these are the geometry buffers
    UINT32                  nLightCount;    // Store here the amount of lights 
    LIGHT_DATA**            ppLightData;    // Store here the pointers to the light data
    UINT32                  nTextureCount;
    TEXTURE_DATA*           pTextureData[GMAX_TEXREF];  

But then any change to the frame DATA will require that the other values get duplicated. So now I was thinking about making smaller structures with the fTime variable each so I can have attribute keyframes, light keyframes and buffer keyframes, then make the node structure look like this:

struct NODE3D_DATA { UINT   nNodeID;
    UINT                    nTransformFrames; 
    FRAMETRANSFORM_DATA**   ppTransformFrames;
    UINT                    nAttributeFrames; 
    FRAMEATTRIBUTE_DATA**   ppAttributeFrames;
    UINT                    nLightFrames;
    FRAMELIGHT_DATA**       ppLightFrames;
    UINT32                  nGeometryFrames;    // Store here the amount of buffer keyframes
    GEOMETRYFRAME_DATA**    ppGeometryFrames;   // Store here the pointers to the buffer data?

then probably:

struct GEOMETRYFRAME_DATA { float fTime; // or uint or whatever
    UINT nBufferCount; //
    INDEXEDGEOMETRY_DATA** ppGeometryBuffers;

Other ways I have tried or thought about seem to be even more redundant, and in the cases that they don't, the performance loss is untolerable for my needs. Ah and I forgot to mention that nodes can have hierarchy, but this is not a problem if the frame data is reduced. The structure instances don't get duplicated unless any value changes. When constructing a node object it's done referencing a pool of structure instances to prevent allocation for 2 equal frames. Also the lights are something that I'm not sure on how to handle as I need them to be attached to the node in most cases, even if I allow setting global lights to the scene.

@Nicol: Yes that's what I'm trying to figure out. You can see the code doesn't rely on any hardware in particular or contains any game attributes, it's just keyframed animation data and some IDs for them and to reference API objects. I use objects such as CFrame to hold pointers to frame data and to store runtime values. The thing is I don't want to create layers unless it's really worth it, to prevent excessive distinction for small values, such as a making a frame object to store only the Diffuse color variation or making it store any value that can change but making it so random that I woulnd't know how to relate the data to the other frames, or need a complex system for it. That's why at this level I prefer to have well-defined data so I can store and work with fast during runtime.

It's this complicated because it must support a lot of rendering and animation techniques, must support being accessed by game entities and other objects, and must allow (somewhat)easy optimization for them if needed.

Anything that helps me make a decision is a good answer for my question, thanks for the reply.


1 Answer 1


I'd say that your main problem is that you're giving meshes non-mesh properties, thus leading to a fat interface.

A mesh is a set of per-vertex data and the information about what those vertex attributes means. That's all it should be. It may have some information, like where the "center" of it is and a bounding box perhaps, but that's about it. A mesh should have no understand of animations. It should not have a world-space position. Meshes should not have shaders. And so forth.

All of that information should live elsewhere.

As to the specific division of information, that depends on your needs. I would design it based on layered functionality. Figure out what things need to talk to each other, then build objects that facilitate this communication. For example, a mesh holds vertex data. But that alone isn't enough to render it; you need a material, which contains the shader and associated parameters necessary for rendering. So a renderable object would contain a reference to a material and a mesh.

  • \$\begingroup\$ Thanks, I figured out a way of doing exactly what you say and it turned out to be the best option. I moved the animation thing to the "Entity" class (or actually something like "KeyframedEntity") which works as controller for the animation and binds the mesh to the physics and message system. Layered functionality is really powerful as long as there are well defined layers and there are few enough to be able to traverse from top to bottom with ease, specially on these systems which from build low level to high level processes. \$\endgroup\$ Commented Mar 26, 2012 at 22:33

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .