False negative in variance of dict value

[Viicos]

(Sorry for the multiple questions these past few days).

Code sample in pyright playground

The following does not seem to complain about the invariance of dict values, while mypy does.

from typing import TypeAlias, Mapping

IncEx: TypeAlias = 'dict[int, str] | dict[str, IncEx | bool]'


def func(arg: IncEx): pass


func({"elements": {"__all__": True}})
func({"__all__": True})

We are going to switch to Mapping on Pydantic but just wanted to make sure pyright wasn't missing something here. Thanks in advance.

[erictraut]

Pyright's behavior is correct here. The expression {"elements": {"__all__": True}}) constructs an object that is compatible with the type dict[str, IncEx | bool]. This type can be evaluated through bidirectional type inference (what mypy calls "context"), so no error should be generated here.

This appears to be a bug in mypy. It looks like it is failing to apply bidirectional type inference in this case and is falling back on regular type inference rules. Pyright does this as well if you separate the code into two lines.

v1: IncEx = {"elements": {"__all__": True}} # OK
reveal_type(v1) # IncEx

v2 = {"elements": {"__all__": True}}
reveal_type(v2) # dict[str, dict[str, bool]]
v3: IncEx = v2 # Type error due to invariance

Interestingly, you can avoid the mypy bug if you simplify the type alias slightly.

from typing import TypeAlias

IncEx: TypeAlias = "dict[str, IncEx | bool]"

def func(arg: IncEx):
    pass

func({"elements": {"__all__": True}})
func({"__all__": True})

[Viicos]

Thanks for the answer. Indeed having dict[int, str] in the alias is what seems to be causing the mypy issue. We'll be using Mapping in the meanwhile.