[3.12] `EnumMeta.__getattr__` removed without deprecation

[jacobtylerwalls]

Bug report

>>> from enum import Enum
>>> class Color(Enum):
... RED = "red"
... 

3.11

>>> Color.__getattr__("RED")
<Color.RED: 'red'>

3.12b4

>>> Color.__getattr__("RED")
Traceback (most recent call last):
 File "<stdin>", line 1, in <module>
AttributeError: type object 'Color' has no attribute '__getattr__'. Did you mean: '__getitem__'?

---

I see that __getattr__ is documented, so I would have expected either a deprecation notice or an entry in What's New for Python 3.12.

Linked PRs

• gh-106762: Add news for EnumMeta.__getattr__ removal #107466
• [3.12] gh-106762: Add news for EnumMeta.__getattr__ removal (GH-107466) #107509

[ethanfurman]

It's actually EnumMeta.__getattr__, but yes, it needs to be removed from the documentation as well as having a news entry.

[jacobtylerwalls]

Do you remember what the rationale was for removing __getattr__ in 4e704d7? Most of that commit was reverted in #103236 but not the removal of __getattr__.

The docs reference to __getattr__ was already removed in 6a980de.

[ethanfurman]

It was the original method of looking up enum members, but it was quite slow; a more performant method is being used now.

Do you have a situation where __getattr__ was being directly called?

[jacobtylerwalls]

Yes, I found it when trying to pass the pylint test suite against Python 3.12. Apparently the presence of a __getattr__ attribute is used as a heuristic for whether dot-access syntax is permitted, so pylint's linting of enums will have false negatives without a patch.

Do you have a situation where getattr was being directly called?

So, not called directly, but more like if getattr(EnumClass, "__getattr__"):

A deprecation period would give just a little more runway to deal with that, but the patch probably isn't particularly involved.

[jacobtylerwalls]

What about just keeping __getattr__ as a shim for __getitem__?

[ethanfurman]

Where is has_dynamic_getattr defined? Is it just looking for presence, or something more?

[jacobtylerwalls]

It's in an upstream dependency astroid. Here's has_dynamic_getattr: I'm fairly certain it just checks for presence. (When it calls self.getattr, it's checking for the presence of an entry in self.locals, where self is astroid's version of the ast node for the class, and locals is the set of names defined on the class, like this):

# python 3.11
(Pdb) self
<ClassDef.Color l.2 at 0x106fb0910>
(Pdb) self.metaclass()
<ClassDef.EnumType l.468 at 0x1072c9450>
(Pdb) '__getattr__' in self.metaclass().locals
True

[jacobtylerwalls]

cc/ @Yhg1s to put on your radar for rc1.