Add the __annotations__ attribute to the ClassDef object model.

[mbyrnepr2]

Type of Changes

	Type
✓	🐛 Bug fix
	✨ New feature
	🔨 Refactoring
	📜 Docs

Description

Add the __annotations__ attribute to the ClassDef object model.
Pylint now does not emit a no-member error when accessing __annotations__ in the following cases:

case 1.

class Test:
    print(__annotations__)

case 2.

from typing import TypedDict

OtherTypedDict = TypedDict('OtherTypedDict', {'a': int, 'b': str})
print(OtherTypedDict.__annotations__)

case 1 is similar to the behaviour of some of the other special attributes:

class Test:
    print(__module__)
    print(__qualname__)
    print(__annotations__)

It turns out that this fix also fixed case 2.
However, it's still unclear why case 2 was only a false positive for Python 3.9 and older 😬

Closes pylint-dev/pylint#7126

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 92.72%. Comparing base (2c38c02) to head (90f6f67).
Report is 185 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2431      +/-   ##
==========================================
- Coverage   92.78%   92.72%   -0.06%     
==========================================
  Files          94       94              
  Lines       11098    10996     -102     
==========================================
- Hits        10297    10196     -101     
+ Misses        801      800       -1     

Flag	Coverage Δ
linux	92.60% <100.00%> (+0.01%)	⬆️
pypy	92.72% <100.00%> (-0.06%)	⬇️
windows	92.70% <100.00%> (+0.02%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
astroid/interpreter/objectmodel.py	93.27% <100.00%> (+0.02%)	⬆️
astroid/nodes/scoped_nodes/scoped_nodes.py	92.77% <100.00%> (+0.25%)	⬆️

... and 70 files with indirect coverage changes

[DanielNoord]

I'm fine with the change, but should this behind a version flag?

See https://docs.python.org/3/howto/annotations.html. It seems it was only added to all objects in 3.10+?

[mbyrnepr2]

I'm fine with the change, but should this behind a version flag?

See https://docs.python.org/3/howto/annotations.html. It seems it was only added to all objects in 3.10+?

I see what you mean @DanielNoord great point. On Python 3.8, for example, we can demonstrate your point (this corresponds to case 2 in the description above):

>>> class Fruit:
...     pass
... 
>>> Fruit().__annotations__
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'Fruit' object has no attribute '__annotations__'

Now, let's look at case 1 on Python 3.8:

>>> class Fruit:
...     print(__annotations__)
... 
{}
>>> 

Without the fix, Pylint does emit the error here. At this moment I don't know if it's better to make a distinction or keep it as-is but I'm open to ideas.

[DanielNoord]

I think a distinction would be better as that better represents the actual versions right?

[mbyrnepr2]

I would need to investigate a bit further @DanielNoord but my thinking is that we need the attribute anyway to satisfy both cases; so is the distinction necessary? Possibly but I need to see how that could work

[DanielNoord]

Thinking about this some more, we shouldn't optimize for old versions. Most people will be running their linter with any of the newer versions anyway. Let's merge! (with one last nit :))

[jacobtylerwalls]

However, it's still unclear why case 2 was only a false positive for Python 3.9 and older 😬

At a glance, I expect that's because on 3.9 and below, the base class __annotations__ was returned.

[jacobtylerwalls]

Thanks for this!

[jacobtylerwalls]

This is not important at the present moment, but elsewhere I've been encouraged to avoid boilerplate like "test that" and "we", and I think it's a helpful edit ;)