What is the difference between __getattr__ and __getattribute__?

Question presented to candidate: "Python classes can define both __getattr__ and __getattribute__. What is the difference, when does each one run, and which would you use to build something like a proxy or a lazy-loading object?"

What a strong answer should cover:

• __getattribute__ is called for every attribute read on an instance (including methods and __dict__); object.__getattribute__ implements the normal lookup.
• __getattr__ is a fallback, called only when the normal lookup raises AttributeError; the normal case pays nothing for it.
• The recursion trap: any self.x inside __getattribute__ calls it again; use object.__getattribute__(self, name) or super().__getattribute__(name).
• Both must raise AttributeError for unknown names so hasattr and getattr(obj, name, default) keep working; an AttributeError escaping a property silently triggers __getattr__.
• Implicit special-method lookups (len(x), x + y) bypass both on the instance, so a proxy must define the dunders it needs; modules can define __getattr__ too (PEP 562, Python 3.7).
• Choice: __getattr__ for delegation, defaults and lazy attributes; __getattribute__ only for rare cases that must see every access (tracing, access control), because it is slower and easy to get wrong.

Code / implementation expected: Optional — a delegating proxy with __getattr__, or a tracing __getattribute__ that calls object.__getattribute__.