add support for unknown dependencies that provide a default value

[iverberk]

This PR adds support for dependencies of an unknown type that provide a default value, allowing them to be instantiated even if the type is not known in the registry.

Fixes #94

[maldoinc]

Hi @iverberk, thanks for this! Just a few minor comments but functionality-wise this is ready to go.

[maldoinc]

Would be good to have some tests that capture the same behavior for configuration injection via Inject(param=)

[iverberk]

I'll add them!

[iverberk]

Are the tests satisfactory?

[maldoinc]

Looks good, thanks! Just need to sort out the linting checks. You can run make lint, make test locally as well.

[iverberk]

linting errors are resolved, at least locally.

[iverberk]

The build breaks on Python < 3.10 due to the pydantic_settings library requirements. What would you propose in this case? I think I needed to add the Python >= 3.10 requirement, otherwise Poetry would fail to resolve the dependencies.

[maldoinc]

Let's skip the pydantic test case for versions older than py3.10.

[iverberk]

Not sure what is going on with the 3.9 tests? It works locally with 3.9.25.

[maldoinc]

The error comes from the stringify_type function when called with an optional type. You can get the raw type directly like this: stringify_type(analyze_type(parameter.klass).raw_type)

[iverberk]

Do you want me to update all the stringify_type(target) instances to the new form stringify_type(analyze_type(parameter.klass).raw_type)? I'm unable to reproduce this locally with Python 3.9.25, which makes it a bit hard to verify.

[maldoinc]

Hmm, strange. I tested it with the same version locally and I can reporoduce it. The failing test cases is test_unknown_service_with_default_value

[iverberk]

As expected, a configuration error on my system. I installed a a second venv for 3.9 but unbeknownst to me, it was still defaulting to the 3.13 one.

[iverberk]

Should we skip the Pydantic check for 3.14 also? Pydantic seems to have a compatibility problem with 3.14 at the moment.

[maldoinc]

Let's skip that for now.

[maldoinc]

Some edge case we have not addressed here: *args, **kwargs` which we can ignore as they have implicit default values. Same for untyped parameters as long as they have a default value.

@service
class Thing:
    def __init__(self, 
        foo: Foo | None, 
        x = 1,
        some_config: Annotated[str | None, Inject(param="test_config")] = None, 
        *args: Any, 
        **kwargs: Any
    ) -> None:
        self.foo = foo
        self.bar = bar

[iverberk]

I'm not sure I follow what I need to do here. Is your example something that wouldn't work with the new changes? I put it in a testcase and it was happy to provide the dependency. Can you make it a little more specific what is currently missing from the code? Thanks!

[maldoinc]

Yeah, the above will fail since some parameters have no type at all, and stuff like args and kwargs that also have "default" values are validated. We can skip these kind of parameters outright during dependency registration.

[maldoinc]

Looks good, thanks. Just py3.8 lint failing.

[iverberk]

Always something :-) Pushed a new commit to fix.

[maldoinc]

Thank you @iverberk!