examples.md

Flyweight — where it lives outside this repo

Cited, real implementations to study (or point an agent at) when designing or reviewing sharing/interning code.

Python standard library

• sys.intern. Explicit string interning: one shared copy, pointer-fast equality. The docs call out dictionary keys as the winning case. docs.python.org/3/library/sys.html#sys.intern
• CPython small-int interning. Integers −5..256 are pre-built singletons; the interpreter runs the pattern under you, which is why careless is checks on small ints "work" and then betray you at 257. docs.python.org/3/c-api/long.html
• functools.lru_cache. A memoizing decorator that, applied to a factory, is the flyweight pool — the guide chapter's own recommendation. docs.python.org/3/library/functools.html#functools.lru_cache

Major ecosystems

• spaCy StringStore. Interns every vocabulary string to a 64-bit hash so tokens across a corpus share one copy — flyweight at NLP scale. spacy.io/api/stringstore
• Apache Arrow dictionary arrays (pandas Categorical). Column-scale value sharing: each distinct value stored once, rows hold small indices. arrow.apache.org/docs/python/data.html#dictionary-arrays

What to notice across all of them

Every production flyweight shares only immutable values, and none of them expose the pooled object for mutation. And each one earned its place with a measurement — interning pays at corpus/column scale, not at 52 cards.