The data model

Roughly, for an instance:

1. Python looks up type(obj).__mro__ for a data descriptor named x (defines __set__/__delete__, e.g. property). If found, its __get__ wins.
2. Otherwise it checks obj.__dict__ for x.
3. Otherwise it checks the class MRO again for a non-data descriptor (e.g. a plain function, which becomes a bound method) or a plain class attribute.
4. Otherwise it calls type(obj).__getattr__(obj, "x") if defined, else raises AttributeError.

__getattribute__ is what actually runs this whole algorithm; __getattr__ is only the fallback for the miss case.

• __repr__ is for developers: unambiguous, ideally eval-able, shown in the REPL, in containers ([obj]), and by the debugger. Always define it.
• __str__ is for end users / readable output; used by str() and print(). Falls back to __repr__ if not defined.

A good default __repr__ is f"{type(self).__name__}(name={self.name!r})".