[ty] Add attribute assignment tests

[thejchap]

Summary

Related to: #18347

This PR adds tests for some of the current behavior of validate_attribute_assignment, to help with the refactor. This catches some issues that surfaced in the ecosystem check.

Opening as a separate PR so I can see these pass against main then rebase #18347 on top of this and fail them

Projects

• beartype
• datadog
• sphinx
  • https://sourceforge.net/p/docutils/code/HEAD/tree/trunk/docutils/docutils/frontend.py#l568 conditional initialization
  • https://github.com/sphinx-doc/sphinx/blob/master/sphinx/builders/__init__.py#L689
• streamlit

Test Plan

New mdtests

[thejchap]

mypy:

../test.py:12: error: Item "A" of "A | B" has no attribute "x" [union-attr]

[github-actions]

mypy_primer results

No ecosystem changes detected ✅

[thejchap]

this should address the beartype regression this doesn't catch the issue/fail on the other branch

[thejchap]

@sharkdp

making slow progress on these - limited time :)

i'm still unclear on why the warning in beartype (warning[possibly-unbound-attribute] beartype/_util/api/external/utilclick.py:108:5: Attribute callbackon typeBeartypeableT is possibly unbound) is no longer being emitted on #18347

however while investigating i did find what i think is an issue on main with setting attributes on generics where the upper bound is a union, after narrowing. added a test to illustrate

this results in a possibly-unbound-attribute from ty but no diagnostic from mypy

mypy

../test.py:16: note: Revealed type is "test.B"
Success: no issues found in 1 source file

ty

info[revealed-type]: Revealed type
  --> /Users/justinchapman/src/test.py:16:23
   |
14 |     if not isinstance(b, B):
15 |         raise TypeError("Expected instance of B")
16 |     print(reveal_type(b))
   |                       ^ `C & B`
17 |     b.x = 42
   |

code

from typing import Union, TypeVar
from typing_extensions import reveal_type

class A:
    pass

class B:
    def __init__(self) -> None:
        self.x: int = 0

C = TypeVar("C", bound=Union[A, B])

def _(b: C):
    if not isinstance(b, B):
        raise TypeError("Expected instance of B")
    print(reveal_type(b))
    b.x = 42

b = B()
_(b)

[sharkdp]

That looks like a bug, yes. A similar bug was mentioned on Discord the other day. Would you mind reporting that as a separate issue? A simplified example would be https://play.ty.dev/62d91bdb-cbf1-4758-8f2c-6a26722c9f47

[thejchap]

@sharkdp astral-sh/ty#653

[thejchap]

this should address the datadog regression doesn't seem to