Sub-types of a union type need to be treated as covariant
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 20.6k
- Forks
- 3.3k
- PR merge metrics
- PR metrics pending
Description
Bug Report
Starting from microsoft/pylance-release#2172.
Currently, A or B seems not to be considered covariant with the union type A | B when A or B comes as a type variable. Because of this, Mypy tends to give false positive error about variance settings in protocols.
To Reproduce
Here are two sample files.
https://mypy-play.net/?mypy=latest&python=3.8&gist=2c061e7b00dbf6024dc4bf13cea7d896
from __future__ import annotations
from typing import Protocol, TypeVar
_T0 = TypeVar("_T0")
_T0_co = TypeVar("_T0_co", covariant=True)
_T1 = TypeVar("_T1")
_T1_co = TypeVar("_T1_co", covariant=True)
#%%
class FP0(Protocol[_T0, _T1]):
def __call__(self, arg0: _T0 | _T1, /) -> tuple[_T0, _T1]:
...
def return_fp0_float_float() -> FP0[float, float]:
def f(x: float) -> tuple[float, float]:
x += 1
return (0, 0)
return f
def return_fp0_object_object() -> FP0[object, object]:
# Dangerous cast, reported.
return return_fp0_float_float()
#%%
class FP0A(Protocol[_T0_co, _T1_co]):
def __call__(self, arg0: _T0_co | _T1_co, /) -> tuple[_T0_co, _T1_co]:
...
def return_fp0a_float_float() -> FP0A[float, float]:
def f(x: float) -> tuple[float, float]:
x += 1
return (0, 0)
return f
def return_fp0a_object_object() -> FP0A[object, object]:
# Dangerous cast, but not reported.
return return_fp0a_float_float()
fp0a_object_object: FP0A[object, object] = return_fp0a_object_object()
_ = fp0a_object_object(object())
https://mypy-play.net/?mypy=latest&python=3.8&gist=686538007d253a9241dfeaf4c7a3598d
from __future__ import annotations
from typing import Protocol, TypeVar
_T0 = TypeVar("_T0")
_T0_contra = TypeVar("_T0_contra", contravariant=True)
_T1 = TypeVar("_T1")
_T1_contra = TypeVar("_T1_contra", contravariant=True)
#%%
class FP1(Protocol[_T0, _T1]):
def __call__(self, arg0: _T0, arg1: _T1, /) -> _T0 | _T1:
...
def return_fp1_object_object() -> FP1[object, object]:
def f(a: object, b: object) -> object:
return object()
return f
def return_fp1_float_float() -> FP1[float, float]:
# Dangerous cast, reported.
return return_fp1_object_object()
#%%
class FP1A(Protocol[_T0_contra, _T1_contra]):
def __call__(self, arg0: _T0_contra, arg1: _T1_contra, /) -> _T0_contra | _T1_contra:
...
def return_fp1a_object_object() -> FP1A[object, object]:
def f(a: object, b: object) -> object:
return object()
return f
def return_fp1a_float_float() -> FP1A[float, float]:
# Dangerous cast, but not reported.
return return_fp1a_object_object()
fp1a_float_float: FP1A[float, float] = return_fp1a_float_float()
result: float = fp1a_float_float(1.0, 2.0)
result += 1
FP0 and FP1 are in my expected forms, but get reported with errors because of variance settings. FP0A and FP1A are the revised versions based on the errors Mypy gives for FP0 and FP1.
Running either file would lead to a crash on the last line respectively, because of the unreported dangerous casts which are marked with comments.
Expected Behavior
Variance settings in FP0 and FP1 are right, and those in FP0A and FP1A are wrong.
_T0 and _T1 are both covariant with the argument type _T0 | _T1 in FP0, so they should be invariant when they also appear in the return type.
_T0 and _T1 are both covariant with the return type _T0 | _T1 in FP1, so they should be invariant when they also appear in the argument types.
Actual Behavior
Variance settings in FP0 and FP1 are considered wrong, and those in FP0A and FP1A are considered right.
Your Environment
- Mypy version used: Latest(0.910) | Master
- Mypy command-line flags: None
- Mypy configuration options from
mypy.ini(and other config files): N/A - Python version used: 3.8
- Operating system and version: Mypy Playground
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by running the two reproduction snippets from the issue in the linked mypy playground examples and compare the variance diagnostics for FP0/FP1 with FP0A/FP1A. Trace the type-checker's variance handling for union types and type variables; done means the expected variance settings are accepted and the unsafe assignments are rejected.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- compilers
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 38/100