A generic 'frame' type

[brentonmallen1]

I'm not sure how all this works, so if my idea doesn't make sense or isn't feasible, no worries.

My thought is to have a generic type that could encompass objects like the pandas or polars dataframe without having to import those libraries. Maybe it's a collection? The motivation is that these libraries can be large and so their use could be made optional, but in order to type a function that could take one as an argument, the library or a stubs dependency would have to be added. There is the pandas-stubs, which is what make this DOA, but one doesn't exist for polars (which might make it out of scope here). Not sure if there are other libaries/objects that would benefit from something like this. Anyway, just had the thought and I think mostly answered it to myself as I typed but maybe someone else has more thoughts/suggestions.

example:

# something object could either be a string (json) or a dataframe
def process_records(records: Union[str, typing.frame]) -> Union[dict, typing.frame]:
    ...

[AasthathecoderX]

I would like to work on this issue,as part of GSoC 2025.

[dangotbanned]

My thought is to have a generic type that could encompass objects like the pandas or polars dataframe without having to import those libraries.

@brentonmallen1
It sounds like the use case you have might be best served by https://github.com/narwhals-dev/narwhals

✅ Zero dependencies, Narwhals only uses what the user passes in so your library can stay lightweight

---

If you did still want to roll your own library dispatch handling, then something like the more minimal protocols here could be a starting point (https://github.com/narwhals-dev/narwhals/blob/3aefc4971aebb29f1400c3f7748b430b59ecf7b1/narwhals/typing.py)