Python identifiers are NFKC-normalized but strings are not: after user = 1, getattr(module, 'user') finds nothing
The parser normalizes identifiers with NFKC, so fullwidth or ligature names in source become plain ASCII: user = 1 defines user (print(user) prints 1), file = 2 defines file, and obj.x = 3 stores the attribute 'x'. Runtime strings are not normalized, so lookups by the original spelling miss: getattr(obj, 'x') raises AttributeError: 'C' object has no attribute 'x', and globals().get('user') returns None while globals().get('user') returns 1. Plugin systems, config-to-attribute mappers and template engines that resolve names from external text silently fail for such input. Fix: apply unicodedata.normalize('NFKC', name) to externally supplied names before getattr/setattr/globals() lookups, or restrict accepted names to ASCII. Verified on 3.12.12 and 3.14.7.