(🐞) TypeGuard behaves inconsistently on instance methods.
まだ誰も着手していません。
- 主要言語
- Python
- スター
- 20.6k
- フォーク
- 3.3k
- PR マージ指標
- PR 指標を取得中
説明
Bug Report
When type guards are implemented as methods, Mypy's type narrowing behaviour is inconsistent:
- It depends on whether the method only takes the implicit
selfargument or instead takes additional arguments. - It further depends on whether the second argument has a default value, regardless of whether an explicit value is passed to it in the call expression.
This does not appear to be in compliance with appears to fall through the cracks of PEP 647, whose stated behaviour does not depend on any arguments further than the first explicit argument. Some relevant quotes from the PEP follow:
Type checkers should assume that type narrowing should be applied to the expression that is passed as the first positional argument to a user-defined type guard. If the type guard function accepts more than one argument, no type narrowing is applied to those additional argument expressions.
If the type guard function is implemented as an instance or class method, an implicit self or cls argument will also be passed to the function.
If a type guard function is implemented as an instance method or class method, the first positional argument maps to the second parameter (after
selforcls).
A concern was raised that there may be cases where it is desired to apply the narrowing logic on
selfandcls. [...] It was [...] decided that no special provision would be made for it. If narrowing ofselforclsis required, the value can be passed as an explicit argument to a type guard function.
Expected Behavior
In Issue #16146, it was clarified that the intent of PEP 647 is for positionality to depend on the call expression: the first positional argument is c in the call expression a.f(c), but it is a in the call expression A.f(a, c).
Now, consider the following pattern where a type guard is implemented by an instance method, with arbitrary arguments:
class A:
def f(self, ...) -> TypeGuard[B]:
# ^^^ your choice of arguments here
return isinstance(self, B)
class B(A): ...
The My expectation here is that call expressions in the form A.f(a, ...)—where ... here stands for arbitary values passed to other arguments in such a way as to make the call legal—will always result in type narrowing of a from a: A to a: B. This behaviour should be independent of the number of additional arguments and/or whether additional arguments are given default values (mentioned here because default values increase the number of legal call expressions).
Actual Behavior
In the three code snippets below, Mypy fails to apply this logic to the case where methods only take the implicit argument self, so that A.f(a) does not result in type narrowing for a while A.f(a, c) does. Furthermore, A.f(a, c) results in type narrowing behaviour for a when c is not given a default value, but fails to do so when a default value for c is specified, regardless of whether the call expression is A.f(a, c) (explicit value passed for c) or A.f(a) (default value used for c).
If the method f is implemented as a classmethod, A.f(a) results in type narrowing for a, as expected.
Pylance behaves as expected I expect in all cases.
To Reproduce
❌ Case 1: only self argument.
from __future__ import annotations
from typing import Any, TypeGuard, reveal_type
class A:
def f(self) -> TypeGuard[B]:
return isinstance(self, B)
class B(A): ...
a = A()
assert A.f(a)
reveal_type(a)
# Pylance: Type of "a" is "B"
# Mypy: Revealed type is "A"
b: B = a
# Mypy: Incompatible types in assignment (expression has type "A", variable has type "B")
# Pylance: OK
✅ Case 2: self argument plus one explicit argument.
from __future__ import annotations
from typing import TypeGuard, reveal_type
class A:
def f(self, c: str) -> TypeGuard[B]:
print(c)
return isinstance(self, B)
class B(A): ...
a = A()
assert A.f(a, "Hi!")
reveal_type(a)
# Mypy: Revealed type is "B"
# Pylance: Type of "a" is "B"
b: B = a # No error
❌ Case 3: self argument plus one explicit argument with default value. Second argument passed explicitly in call expression.
from __future__ import annotations
from typing import Any, TypeGuard, reveal_type
class A:
def f(self, c: str = "") -> TypeGuard[B]:
print(c)
return isinstance(self, B)
class B(A): ...
a = A()
assert A.f(a, "Hi!")
reveal_type(a)
# Mypy: Revealed type is "A"
# Pylance: Type of "a" is "B"
b: B = a
# Mypy: Incompatible types in assignment (expression has type "A", variable has type "B")
# Pylance: OK
❌ Case 4: self argument plus one explicit argument with default value. Default value for second argument used implicitly in call expression.
from __future__ import annotations
from typing import Any, TypeGuard, reveal_type
class A:
def f(self, c: str = "") -> TypeGuard[B]:
print(c)
return isinstance(self, B)
class B(A): ...
a = A()
assert A.f(a)
reveal_type(a)
# Mypy: Revealed type is "A"
# Pylance: Type of "a" is "B"
b: B = a
# Mypy: Incompatible types in assignment (expression has type "A", variable has type "B")
# Pylance: OK
✅ Case 5: self argument plus two explicit arguments, only one of which has a default value. (Behaviour does not depend on whether the third argument is passed explicitly or not.)
from __future__ import annotations
from typing import Any, TypeGuard, reveal_type
class A:
def f(self, c: str, d: int = 0) -> TypeGuard[B]:
print(c)
return isinstance(self, B)
class B(A): ...
a = A()
assert A.f(a, "Hi")
reveal_type(a)
# Mypy: Revealed type is "B"
# Pylance: Type of "a" is "B"
b: B = a # No error
✅ Case 6: like Case 1, but using a class method instead of an instance method.
from __future__ import annotations
from typing import Any, Self, TypeGuard, reveal_type
class A:
@classmethod
def f(cls, a: Self) -> TypeGuard[B]:
return isinstance(a, B)
class B(A): ...
a = A()
assert A.f(a)
reveal_type(a)
# Mypy: Revealed type is "B"
# Pylance: Type of "a" is "B"
b: B = a # No error
Your Environment
- Mypy version used:
mypy 1.7.1 (compiled: yes) - Mypy command-line flags:
--strict - Mypy configuration options from
mypy.ini(and other config files): None - Python version used:
Python 3.12.0
コントリビューションガイド
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
調査の方向性
まず、issue にある 6 つの再現ケースを mypy 1.7.1 で実行し、報告された narrowing を期待される結果と比較します。暗黙の self とデフォルト引数を含む、インスタンスメソッド呼び出しに対する TypeGuard の処理を追跡します。完了の条件は、インスタンスメソッドのケースが、記載された PEP 647 の動作に一貫して従って narrowing され、既に通過しているケースをリグレッションさせないことです。
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- python
- 領域
- compilers, devtools
- issue の種類
- バグ
- 難易度
- 4/5
- 見積もり時間
- 3〜5日
- 活発さ
- 停滞
- 明瞭さ
- おおむね明確
- 初心者へのやさしさ
- 38/100