Repository navigation
Open
Description
Bug or feature request: feature request (I think)
I tried to use TypedDict for my codebase and ran into some issues. Consider the following code snippet:
from typing import Any, Union, Optional
from typing_extensions import Literal
from mypy_extensions import TypedDict
class Test(TypedDict):
a: str
b: int
c: int
TestKeys = Literal["a", "b", "c"] # QUESTION: is there a Test.KEYS or something equivalent?
TestValue = Union[str, int] # QUESTION: is there a Test.VALUE_TYPES or something equivalent?
class Foo:
def __init__(self):
self._foo: Optional[Test] = None
def _set(self) -> None:
self._foo = {"a": "a", "b": 0, "c": 1}
def __getitem__(self, key: TestKeys) -> Any:
self._set()
assert self._foo is not None
return self._foo[key]
def a(self) -> str:
return self["a"]
def b(self) -> int:
return self["b"]
class Bar:
def __init__(self):
self._bar: Optional[Test] = None
def _set(self) -> None:
self._bar = {"a": "a", "b": 0, "c": 1}
def __getitem__(self, key: TestKeys) -> TestValue:
self._set()
assert self._bar is not None
return self._bar[key]
def a(self) -> str:
return self["a"] # WRONG: Incompatible return value type (got "Union[str, int]", expected "str")
def b(self) -> int:
return self["b"] # WRONG: Incompatible return value type (got "Union[str, int]", expected "int")
a = Foo().a()
a.find("a")
b = Foo().b()
b.find("a") # CORRECT: "int" has no attribute "find"
c = Foo()["a"]
c.find("a")
d = Foo()["b"]
d.find("a") # MISSING: no error reported here (if self._foo is not lazily initialized the error gets correctly reported)
e = Bar().a()
e.find("a")
f = Bar().b()
f.find("a") # CORRECT: "int" has no attribute "find"
g = Bar()["a"]
g.find("a") # WRONG: Item "int" of "Union[str, int]" has no attribute "find"
h = Bar()["b"]
h.find("a") # WRONG: Item "int" of "Union[str, int]" has no attribute "find" (should be "int" has no attribute "find")(Python 3.8.0, mypy==0.740, mypy-extensions==0.4.3)
I'm not sure what the correct return type signature for a wrapping __getitem__ would be. Both Any and Union seem to erase the actual type coming from the TypedDict and lead to incorrect results. Is there currently a way to achieve the desired behavior here? Also, is there a way to specify Literals for the keys without manually copying them?
Something like
KT = TypeVar("KT", bound=TestKeys) # or better Test.KEYS
def __getitem__(self, key: KT) -> Test[KT]:
...would be cool if it worked.
Reacted by yetone, Shaun Cutts, Dominic , Théo Matricon, Sophon Aniketos, herr kaste and ivicak-optimo
Activity
JelleZijlstra commentedon Nov 2, 2019
I think the only current way is to use an overload for every key, which is obviously not great.
[-]Non-trivial TypedDict wrapper does not work as intended[/-][+]Precise typing for TypedDict wrappers[/+]ilevkivskyi commentedon Nov 4, 2019
A feature that would allow this (key types) was discussed a while ago at typing meetup (the primary goal for it was typing for pandas dataframes). We will likely support something like this at some point, but not in near future.
JosuaKrause commentedon Nov 4, 2019
Workaround that currently works, suggested by @JelleZijlstra , however, it will be cumbersome with more keys.
Should I close this issue for now?
ilevkivskyi commentedon Nov 4, 2019
No, I think it is useful to have this use case in mind while working on key types.
A-F-V commentedon Dec 8, 2022
Ideally you would want a way for
wrapper[key]to be perfectly substituable forunderlying_dict[key]in the eyes of the type system, but that does not seem possible at the moment (unless someone has any solutions?)A-F-V commentedon Dec 8, 2022
What I have found is that the type given is usually the union across all possible keys even if it is a literal known before running. I wonder if a change could be made to refine this type if it is known. Like C++
constexpr