Bundle OpenGL 3.3 core rendering on top of SDL2. Builds on sdl2.obl for the window and context; this bundle is the GL side. Compile with -lib sdl2. Targets the **3.3 core profile**, forward-compatible. That is the highest common denominator across the platforms Objeck ships: Windows, Linux and macOS desktop all reach it, and macOS caps at 4.1 core so nothing above 4.1 is portable there. GLES-only hardware (Raspberry Pi and similar) is out of scope by construction. Layers, from thinnest to friendliest -- all public, so nothing is capped: * GLWindow -- a window and a 3.3 core context, opened and paced for you. * GL -- static state operations. Thin, but never per-vertex. * Shader -- a linked program; Flat/Textured/TexturedFog/Normals built in, so the common cases need no GLSL at all. * Mesh -- geometry uploaded once into buffer objects, drawn many times, with Cube/Plane/Quad/Sphere built in, OBJ loading, and instancing. * Texture2D -- a texture, from a file or generated; Solid/Checker built in. * Light -- one directional light plus ambient, written into a shader. * RenderTarget -- draw into a texture instead of the window, then sample it. * Material -- a surface: its texture, tint and shininess. * Frustum -- the six planes of a view, for skipping what is off screen. * ShadowMap -- shadows from a directional or spot light, depth pass and all. * PointShadow -- shadows from a point light, in every direction, via a cube map. * Overlay -- text and rectangles over the scene, positioned in pixels. * Transform -- position, rotation and scale, with a cached matrix. * Box, Scene -- a world of boxes: draws itself, and answers collision. ## Why the API is shaped this way The native call boundary is expensive: the VM resolves each native symbol by string on EVERY call (GetProcAddress/dlsym) and boxes every argument into a fresh holder. A 1:1 mapping of OpenGL onto that would be thousands of lookups and allocations per frame. So each call here does real work -- 'compile a program from two sources' is one call, not the five GL calls it decomposes into -- and bulk data crosses as whole arrays, never element by element. This happens to be exactly what GL 3.3 core wants anyway: upload geometry once, then draw with few calls. ## Adding a call One `void fn(VMContext&)` in the OpenGL section of core/lib/sdl/sdl.cpp, and one method here that names it. See that file's header for the two rules that matter (positional slot indices, and keeping each call coarse).