Objects, attributes, maps, oh my

Explains how dynamic-language objects modeled as per-object name→value hash tables pay heavy costs in hashing and memory indirection, and shows how runtimes replace that with maps (aka shapes/hidden classes): immutable, persistent-tree descriptors that map attribute names to small integer slots and identify an object's in-memory layout. As attributes are added, maps produce new map nodes (next_map) and objects carry a map pointer plus a storage array indexed by slots. This removes string hashing from the fast path, enables inline attribute storage in low-level implementations, and gives a stable, cheap-to-compare layout identifier so callers can detect layout changes quickly.

Gives compact Python sketches for Base/Class/Instance, Map.get_index/next_map, and an InlineCache that caches a (map, index) pair and uses pointer equality to fast-path loads by index; on a miss it falls back to name lookup, updates the cache, and returns the value. Notes Python lacks true in-object inline storage, but the pattern still reduces loads and instructions in an interpreter; production runtimes and JITs use the same concepts and more aggressive optimizations to eliminate interpreted overhead and accelerate attribute access.