List comprehensions now have access to the enclosing class scope

[JelleZijlstra]

Code:

class X:
    y = 1
    print([x + y for x in range(3)])

On current main:

Python 3.12.0a7+ (heads/main:94f30c7557, May 10 2023, 17:27:39) [Clang 14.0.0 (clang-1400.0.29.202)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> class X:
...     y = 1
...     print([x + y for x in range(3)])
... 
[1, 2, 3]

But on 3.11:

>>> class X:
...     y = 1
...     print([x + y for x in range(3)])
... 
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "<stdin>", line 3, in X
  File "<stdin>", line 3, in <listcomp>
NameError: name 'y' is not defined

I actually like the new behavior, and I discovered this bug as I was preparing a post proposing that we make things work this way in 3.13. But it's a behavior change, and we should only make this change on purpose and not by accident.

This could plausibly break real code if the name is defined in both the module and class scope, and current code relies on getting the name from the module scope.

cc @carljm

Linked PRs

• gh-104374: don't inline comprehensions in non-function scopes #104519
• gh-104374: Remove access to class scopes for inlined comprehensions #104528

[asishm-wk]

on 3.10.8 - this prints [1,2,3]

[JelleZijlstra]

I don't think so. I just confirmed on 3.10.9 (don't have a 3.10.8 around), and it throws a NameError. Make sure you run the sample in a class scope.

[asishm-wk]

sorry, you're right! - I forgot to restart the interpreter when testing different variations - so there was already a y defined in the outer scope!

[carljm]

I started a Discourse thread about options for handling this: https://discuss.python.org/t/pep-709-one-behavior-change-that-was-missed-in-the-pep/26691

[carljm]

I filed a SC issue about this: python/steering-council#189