No errors on typed function internals if function definition itself is untyped; I'd argue that's probably incorrect behavior, but at the very least it's *super* confusing and *very* hard to diagnose when it happens #10895
Labels
bug
mypy got something wrong
Comments
|
With |
posita
changed the title
posita
changed the title
posita
changed the title
posita
changed the title
posita
changed the title
posita
changed the title
|
I have now verified this happens with Python versions 3.6.14, 3.7.11, 3.8.2, 3.8.11, and 3.9.6. |
|
I think this is because of https://mypy.readthedocs.io/en/stable/common_issues.html#no-errors-reported-for-obviously-wrong-code. Does running mypy with |
% rm -fr .mypy_cache ; mypy --config=/dev/null --check-untyped-defs test_case3.py
/dev/null: No [mypy] section in config file
test_case3.py:5: error: No overload variant of "__getitem__" of "list" matches argument type "str"
test_case3.py:5: note: Possible overload variants:
test_case3.py:5: note: def __getitem__(self, SupportsIndex) -> int
test_case3.py:5: note: def __getitem__(self, slice) -> List[int]
test_case3.py:8: error: No overload variant of "__getitem__" of "list" matches argument type "str"
test_case3.py:8: note: Possible overload variants:
test_case3.py:8: note: def __getitem__(self, SupportsIndex) -> int
test_case3.py:8: note: def __getitem__(self, slice) -> List[int]
Found 2 errors in 1 file (checked 1 source file)I still think this is a bug, for two reasons. First, the fact that it works at the module level and not the function level is just weird. Second, I swear this used to work with classes that defined their own # test_case4.py
from typing import List
def func():
a: List[int] = [0, 1, 2]
key: str = "asdf"
a[key] # <- This doesn't generate an error
func()
b: List[int] = [0, 1, 2]
key: str = "asdf"
b[key] # <- But this does?!Runtime: |
|
Ugh, talk about a super sharp edge. This works as expected (since I typed the enclosing function): # test_case5.py
from typing import List
def func() -> None:
a: List[int] = [0, 1, 2]
key: str = "asdf"
a[key] # <- *Now* this generates an error
func()
b: List[int] = [0, 1, 2]
key: str = "asdf"
b[key] # <- So does this |
|
As does this: # test_case6.py
from typing import List, Union
class GetItemTest:
def __getitem__(self, key: Union[int, slice]) -> Union[int, List[int]]:
return list(range(10))[key]
def test_a() -> None:
a = GetItemTest()
a["asdf"] # <- errors |
posita
changed the title
|
This is yet another example that supports my thinking that |
posita
changed the title
Maybe as a work-around, but functions seem like pretty arbitrary boundaries to try and walk this particular line, especially where bodies contain explicit type annotations. I can't see how ignoring those can be considered correct under any setting. In other words, yeah, totally make the sane thing the default thing, but I'm pretty sure this is still a bug. |
|
It's documented that mypy doesn't type check unannotated functions (https://mypy.readthedocs.io/en/stable/common_issues.html#no-errors-reported-for-obviously-wrong-code). I don't disagree with you that this is unintuitive, but the other issue is a sufficient place to discuss that. |
Yes, this appears to be a dupe of #3948. But I don't think the documentation is even accurate. Consider: # test_case9.py
from typing import Callable
a: Callable[[], int]
def _untyped_init_a():
global a
a = lambda: [0, 1, 2]["key"] # errors
def _untyped_init_b():
global b
b = lambda: [0, 1, 2]["key"] # doesn't errorAccording to the docs, |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Issue description
This doesn't generate an error:
This does:
That maybe technically "documented", but that's really esoteric and completely unintuitive. I've spent hours chasing this down.
Archived original flailing to arrive at core problem
UPDATE 2: Okay, this just keeps getting weirder and weirder. Consider:
Runtime:
UPDATE 1: You don't need the
@overloads. It still doesn't work without them. Here's the simplified case:Original case (same effect):
Runtime:
I can't prove it, but I feel like this used to work. It seems pretty basic.
The text was updated successfully, but these errors were encountered: