The return type of get_subcommand is the same as the first argument

[ausaki]

from pydantic import BaseModel
from pydantic_settings import BaseSettings, CliSubCommand, get_subcommand
from typing_extensions import reveal_type


class Foo(BaseModel):
    bar: int = 1


class CliSettings(BaseSettings):
    foo: CliSubCommand[Foo] = None


def main():
    subcommand = get_subcommand(CliSettings())
    reveal_type(subcommand) # Type of "subcommand" is "CliSettings | None"


if __name__ == '__main__':
    main()

In the above code, subcommand's type is CliSettings, but it's supposed to be BaseModel. It's because get_subcommand's first parameter model is a generic type PydanticModel.

[hramezani]

Thanks @ausaki for reporting this.

@kschwab do you have time to check it?

[kschwab]

Hi @ausaki, there are a few issues with the above example. Currently the example fails since get_subcommand is required. i.e., running the example results in:

Error: CLI subcommand is required {foo}

After updating the example to use a populated Foo object, it will return the correct type:

def main():
    subcommand = get_subcommand(CliSettings(foo=Foo()))
    reveal_type(subcommand)
    #> Runtime type is 'Foo'

Also, please note that CliSubCommand annotated fields are required (i.e., no default values). For example, to accomplish the original example, it would be:

class CliSettings(BaseSettings):
    foo: CliSubCommand[Foo]

def main():
    subcommand = get_subcommand(CliSettings(foo=None), is_required=False)
    reveal_type(subcommand)
    #> Runtime type is 'NoneType'

[ausaki]

@kschwab Thanks for point out the issues for the example code, I update the code as following:

from pydantic import BaseModel
from pydantic_settings import BaseSettings, CliSubCommand, SettingsConfigDict, get_subcommand
from typing_extensions import reveal_type


class Foo(BaseModel):
    bar: int = 1


class CliSettings(BaseSettings):
    model_config = SettingsConfigDict(
        cli_parse_args=True, cli_enforce_required=True, cli_implicit_flags=True, cli_kebab_case=True
    )

    foo: CliSubCommand[Foo]


def main():
    subcommand = get_subcommand(CliSettings())
    reveal_type(subcommand) # Type of "subcommand" is "CliSettings | None"


if __name__ == '__main__':
    main()

run python test.py foo get

> python test.py foo
Runtime type is 'Foo'

Runtime type is correct, but the static type checker like Pyright infers subcommand's type as CliSettings | None, you can see that in the screenshot, besides that, Pyright also complains Argument missing for parameter "foo" for CliSettings.

[kschwab]

@ausaki thanks for the clarification. The get_subcommand return type cannot be inferred statically, so its return type is Optional[PydanticModel] (defined here), which is either Any or a Union[PydanticDataclass, BaseModel].

However, I am not a static typing expert. @hramezani do you know what should be expected in the above case? Or, is more of a pyright specific issue?

[ausaki]

mypy get the same return type:

→ uvx --with mypy --with pydantic-settings mypy test.py
test.py:18: error: Function is missing a return type annotation  [no-untyped-def]
test.py:18: note: Use "-> None" if function does not return a value
test.py:19: error: Missing named argument "foo" for "CliSettings"  [call-arg]
test.py:20: note: Revealed type is "Union[test.CliSettings, None]"
test.py:24: error: Call to untyped function "main" in typed context  [no-untyped-call]
Found 3 errors in 1 file (checked 1 source file)

PydanticModel is a generic type variable which bounds to Union[PydanticDataclass, BaseModel], so get_subcommand's return type will be the same as the first parameter.

Maybe change the return type and the first parameter's type of get_subcommand to Union[PydanticDataclass, BaseModel]

[kschwab]

Thanks @ausaki, I understand the issue now. Please let me know if PR #659 resolves your issue.

[ausaki]

@kschwab The PR resolved this issue