instancing

FreeBodyEngine.graphics.instancing

Call batching for Renderer.flush_opaque() (see graphics/renderer.py).

This is state-minimizing batching - grouping consecutive draws that share a material/mesh so flush() issues them together, cutting redundant glUseProgram/uniform/texture-bind churn - not real GPU instancing. Real instancing (drawing N copies of a mesh in one glDrawElementsInstanced call via a per-instance transform buffer) needs every mesh-drawing shader to support a second, attribute-driven model-matrix path alongside the uniform one it has today, which is a real ripple across default_shader/tilemap/text/ UI shaders - out of scope for this rebuild and tracked separately as the engine roadmap's own "Instancing" item. Renderer.draw_mesh_instanced() is real, correct infrastructure for that when it's built, but flush() below does not call it - grouping here only ever changes call order, never call count.

group_by_state(calls)

Groups calls (preserving each group's relative order of first appearance) by (material, mesh) identity, so Renderer.flush() can draw every call sharing a material/mesh back-to-back. Materials/meshes are grouped by id() (object identity), not equality, since neither class defines __eq__.