Typing hint for non-strict types?

[futurewasfree]

Wondering if we could also adjust generated typing to have concrete type in type-hint? While keeping strict_data_typing=False and runtime validation/conversion.

E.g. could we replace:

Np1DArray = Annotated[
    Union[np.ndarray[tuple[int], np.dtype[Any]], FilePath, MultiArrayNumpyFile],
    NpArrayPydanticAnnotation.factory(data_type=None, dimensions=1, strict_data_typing=False),
]

with

Np1DArray = Annotated[
    np.ndarray[tuple[int], np.dtype[Any]],
    NpArrayPydanticAnnotation.factory(data_type=None, dimensions=1, strict_data_typing=False),
]

?
(notice removal of FilePath and MultiArrayNumpyFile)

[caniko]

The annotated validator would still allow Union[np.ndarray[tuple[int], np.dtype[Any]], FilePath, MultiArrayNumpyFile], which is why I would advocate against this idea.

What is the motivation?

[futurewasfree]

Maybe there is some misunderstanding here, I suspect.

Depends on how this annotation is being used: the field itself (let's call it .arr) is np.ndarray only within pydantic base model. after the validation and conversion/reading it becomes just np.ndarray of given type (let's say npt.NDArray[np.float32]) so then all the usages of this .arr should be compared against npt.NDArray[np.float32] by static analysis.
but when we instantiate it we could also pass FilePath for example and then for instantiation you need to compare against Union (wider type) in static analysis.

Somewhere you need to cast. Either when you instantiate it or when you start using field of resulting pydantic base model.
I would end up narrowing Union to just npt.NDArray[np.float32] or using Strict types where possible.

[caniko]

I don't think there is a misunderstanding. Is this correct? 🤣

The annotated validator would still allow Union[np.ndarray[tuple[int], np.dtype[Any]], FilePath, MultiArrayNumpyFile], which is why I would advocate against this idea.

I was ignoring static analysis, and focusing on what happens during interpretation.

Another way to put it is from an information perspective. The validator and the type-hint would be out of sync because the validator expects the Union; while the static type only expects np.ndarray.