Can I derive a TypedDict from a class?

[vadimcn]

Is it possible to express the concept of "a TypedDict containing the same fields as <some other class>" to Pyright?
For example:

@dataclass
class Params:
    a: int = 42
    ...
def fucntion(**kwargs: AsTypedDict[Params]):
    ...

(I can't use a straight TypedDict, because they don't support defaults. Besides, it's be nice to be able to use .field syntax for such objects, i.e. params.a = 43, rather than params['a'] = 43).

[erictraut]

No, there's nothing in the Python type standards that enable this. A TypedDict has very specific semantics as specified in PEP 589. If you simply "cast" a dataclass to a TypedDict, those semantic rules will be violated. For example, TypedDict doesn't allow the use of member access expressions (e.g. params.a) whereas dataclass does.

If you want to create a TypedDict from a class, you'd need to programmatically (at runtime) transform the dataclass object into a new TypedDict object.

@dataclass
class Params:
    a: int = 42

class ParamsDict(TypedDict):
    a: int

def convert_params_to_dict(val: Params) -> ParamsDict:
    return {"a": val.a}

There's also currently no way to specify that a **kwargs parameter accepts values within a TypedDict because the notation **kwargs: T means that every value in the kwargs dictionary is of type T.

[vadimcn]

There's also currently no way to specify that a **kwargs parameter accepts values within a TypedDict because the notation **kwargs: T means that every value in the kwargs dictionary is of type T.

😢

Still, I'm curious: Is it possible to implement AsTypedDict in user code, or would this need to be a "magic type" in Pyright?

[erictraut]

I'm not sure what you mean by this. As I said above, you cannot simply "cast" a dataclass to a TypedDict. They have different semantics. So you would need to write a function that allocates a new dictionary object and copies the data out of the dataclass into the dictionary. If you want a type checker to understand the "shape" of that dictionary object (i.e. its allowed keys and corresponding value types), you would need to define that statically in the form of a TypedDict subclass as I showed above.