SimpleNamespace and Protocol

[seasonsteven]

Consider this sample:

from typing import NamedTuple, Protocol
from types import SimpleNamespace


class ISwitch(Protocol):
    @property  # Intention: read required, write optional
    def off(self) -> str: ...

    @property
    def on(self) -> str: ...


class SwitchNT(NamedTuple):
    off: str
    on: str


class Switch:
    def __init__(self, off: str, on: str) -> None:
        self.off = off
        self.on = on


switch1: ISwitch = SwitchNT(off="A", on="B")
switch2: ISwitch = Switch(off="A", on="B")
switch3: ISwitch = SimpleNamespace(off="A", on="B")

switches1: list[ISwitch] = [SwitchNT(off="A", on="B")]
switches2: list[ISwitch] = [Switch(off="A", on="B")]
switches3: list[ISwitch] = [SimpleNamespace(off="A", on="B")]

The above passes mypy but pyright does not accept the SimpleNamespace instance.

Type "SimpleNamespace" is not assignable to declared type "ISwitch"
  "SimpleNamespace" is incompatible with protocol "ISwitch"
    "off" is not present
    "on" is not present  (reportAssignmentType)
Type "list[SimpleNamespace]" is not assignable to declared type "list[ISwitch]"
  "SimpleNamespace" is incompatible with protocol "ISwitch"
    "off" is not present
    "on" is not present  (reportAssignmentType)

Is it reasonable to expect pyright to consider them compatible?

[erictraut]

Is it reasonable to expect pyright to consider them compatible?

No, a SimpleNamespace class doesn't have any declared any attributes or types, so it is not compatible with a protocol that requires certain attributes with certain types.

If you want to write type-safe code, I recommend avoiding SimpleNamespace. Use a NamedTuple or dataclass instead.

[brianschubert]

For context, the reason mypy accepts this is because SimpleNamespace has a __getattribute__ definition in typeshed, which causes mypy to consider all possible attributes to exist and have type Any in subtype checks:

https://github.com/python/typeshed/blob/d70fad09dd7012bd43304e1c67b7599558b90c39/stdlib/types.pyi#L325