Think about covariant alternatives for list types, given that list is invariant

[skinkie]

When mypy'ing my code the problem arrises when list[union[TypeA, TypeB]] is specified and a type like list[TypeA] is assigned. I wonder if the library should offer something like an AppendableSequence that is Covariant but does allow anding multiple items.

[vvmruder]

👍 from me on that feature!

I try to handle this for some time now. Currently I do have the problem with this construction:

from dataclasses import dataclass, field
from typing import List, Optional

@dataclass
class LocalisedTextA:
    Language: Optional[str] = None
    Text: Optional[str] = None

    def test(self) -> bool:
        return True

@dataclass
class LocalisedTextB(LocalisedTextA):
    def test(self) -> bool:
        return self.Language == "en"

@dataclass
class LocalisedTextTYPEA:
    LocalisedText: List[LocalisedTextA] = field(default_factory=list)

@dataclass
class LocalisedTextTYPEB(LocalisedTextTYPEA):
    LocalisedText: List[LocalisedTextB] = field(default_factory=list)

mypy complains with:
Incompatible types in assignment (expression has type "list[LocalisedTextB]", base class "LocalisedTextTYPEA" defined the type as "list[LocalisedTextA]")

Which is correct. And the suggestion is to change from the list definition to Sequence. Would be nice if that is supported.

Currently I change the code to:

from dataclasses import dataclass, field
from typing import List, Optional

@dataclass
class LocalisedTextA:
    Language: Optional[str] = None
    Text: Optional[str] = None

    def test(self) -> bool:
        return True

@dataclass
class LocalisedTextB(LocalisedTextA):
    def test(self) -> bool:
        return self.Language == "en"

@dataclass
class LocalisedTextTYPEA:
    LocalisedText: List[LocalisedTextA] = field(default_factory=list)

@dataclass
class LocalisedTextTYPEB(LocalisedTextTYPEA):
    LocalisedText: List[LocalisedTextB|LocalisedTextA] = field(default_factory=list)

mypy is happy, but it feels a bit wrong. Even when I know why it should be like this. And I still need to test if it is working with xsdata.