Attribute lookup, step by step

Everything is a namespace lookup

You already know obj.x reads an attribute. Advanced Python means knowing the exact order Python consults when resolving obj.x, and using that order as a design tool.

For obj.x on an instance, Python (conceptually) tries, in order:

1. Data descriptor on the type (a descriptor defining __set__ or __delete__) — e.g. property, most validators.
2. Instance dictionary — obj.__dict__.
3. Non-data descriptor on the type (defines only __get__) — e.g. plain functions (that is how bound methods are created!), staticmethod, classmethod, functools.cached_property before first call.
4. __getattr__ fallback (if defined) — called only when normal lookup fails. Great for delegation, dangerous for hiding typos.

Two consequences worth internalizing:

• A data descriptor beats an instance __dict__ entry. That is why property setters can validate even though you could assign obj.x = 5.
• Functions are non-data descriptors; the "bound method" you call is produced at lookup time by __get__. Nothing magic is stored on the instance.

class Loud:
    def __getattr__(self, name):      # only called on MISS
        return f"<fallback for {name}>"

loud = Loud()
loud.real = 1
print(loud.real)                     # 1  (found normally)
print(loud.anything)                 # <fallback for anything>

Use vars(obj) / obj.__dict__ to inspect the instance side, and type(obj).__mro__ to see the class side. When behavior surprises you, ask: "which step of the lookup produced this?"