(C) 2012 Michel Pollet buserror@gmail.com

WARNING This API is not your nanny. It is made to be lean, mean, efficient with no frill, no asserts, no bounds checking, no sugar coating.

On the other hand it's fast, reasonably clean and is a micro-fraction of the other giganormous 'scene graphs' or 'game engine' libraries around.

It's vaguely inspired by THREE.js funnily enough, because it allows you to hack around and quickly get stuff on screen with the minimal amount of effort.

The general idea is that the library keeps track of geometry and stuff, but doesn't do any opengl or related calls. Instead, it uses callbacks into code that will take care of the rendering related tasks.

So for example a c3pixels represents a texture, but a callback into the rendering layer will be responsible to push the pixels to OpenGL, and store the object back into the c3pixels for reference.

The API is generally functional, but it's brand new. I try not to add bits that I aren't needed, and I also don't add stuff that isn't tested.

There is an ASCII STL file loader that works, and a few other bit of geometry related helpers.

It's currently used in one 'serious' project and also in my 3D printer simulator. libc3 also builds a companion library called libc3-gl that implements a default OpenGL renderer 'driver' as a set of callbacks. libc3-gl does the 'dirty' work for loading shaders, creating and maintaining frame buffer objects, vertex buffer objects etc.

There is a PDF Flowchart of how things are mostly organized as far as data structure goes, but the following is a breakdown of the major components.

The API has various bits:

• c3algebra: C derivative of an old C++ piece of code I had lying around and that has been present in my toolset for a long time. It gives you vectors (c3vec2, c3vec3, c3vec4) and matrices (c3mat3, c3mat4) with various tools to manipulate them.
• c3quaternion: Quaternion implementation using c3algebra
• c3camera/c3arcball: camera manipulation bits

The basic data structure is as follow:

• c3context: Hosts a "root" object, and a list of 'viewpoints' (ie either cameras, or lights). it can reproject the objects & geometries, and call the callbacks to draw them.

  The context also keeps a list of c3pixels and c3program that are referenced by the geometries.

• c3object:

  • Has a list of (sub) c3objects
  • Has a list of c3transforms (ie matrices)
  • Has a list of c3geometry (ie real vertices and stuff) The object is a container for other objects, and for geometry itself. Objects don't necessary have geometry and/or sub objects, and don't even need transforms if their vertices are already projected.

• c3geometry:

  • Has a 'type' (raw for simple vertices, texture, triangles etc)
  • Has a 'subtype' (mostly can be used to draw GL types)
  • Has a 'material' (ie color, texture, a GPU program... to be completed)
  • Has a list of vertices
  • Has a list of texture coordinates (optional)
  • Has a list of vertices colors (optional)
  • Had a list of vertices indexes (optional)

• c3transform: Is just a sugar coated matrix, with an optional name.

Also there are:

• c3pixels: Is just a wrapper/holder for some pixels, either allocated, or inherited, it's mostly used for c3texture
• c3texture: Associates a c3geometry with a c3pixels and has a standard Quad for vertices. The OpenGL drawing is not done there, it's done by the application using the generic c3context driver.
• c3cairo: Placeholder for now, inherits from c3texture and will contain a cairo surface mapped to a GL texture.
• c3light: Source of light, as a hidden geometry do it get transformed like a normal one. Right now the implementation uses glLight unless there is a shader installed.

Various object uses static tables of callbacks to implement their behaviours it's kinda cheap c++ inheritance, without the usual bloat.

There just a couple macros to call the driver chain for a particular function call. The relevant bits are in c3driver*.h.

Mostly the code looks for a matching callback in a static table, and call it if found. If that callback wants, it can also call the inherited object callback too.

There is a notion of 'dirtyness' in the tree, when you touch a c3transform, and/remove objects and geometry, a dirty bit is propagated up the tree of object. This tells the rendering it needs to reproject the dirty bits and repopulate the projected vertice cache.

The 'dirty' bit moves both ways, when setting a dirty bit to true, it propagates upward, when you set it to false, it propagates downward in the tree.