Use splats in API about anndata HOT 2 CLOSED

I agree that splats are nicer... and they have been used in Scanpy since the very beginning, e.g. here.

However, both pandas and numpy don't use it in their concat function
(np, pd) and I took these as the model for AnnDatas concat function.

Hence, I'm not sure whether we should deviate from this - I think we should be as close as possible to numpy and pandas whenever it's not completely outrageous.

PS: The Scanpy users are often not very python experienced, and they are used to passing lists without a splat star (*alist) also in many other places, like plotting an AnnData with several colors

sc.pl.scatter(adata, color=['anno1', 'anno2', 'anno3'])

from anndata.

i see. that’s unfortunate, since numpy’s API is confusing for newbies as well. the linked question shows why that API violates several parts of the zen of python.

so we’re stuck between a rock and a hard place: be confusing for python beginners or be confusing for numpy-experienced users.

i think that a good error message goes a long way – using splats means the user can’t pass further arguments by position but only by keyword (which is a good thing), and we could provide either a custom error message, or special handling of sequences being passed as only positional argument, like:

def concatenate(*adatas, batch_key='batch', batch_categories=None):
    if len(adatas) == 1 and not isinstance(adatas[0], AnnData):
        try:  # convert iterable into tuple
            adatas = tuple(adatas[0])
        except Exception:
            pass

from anndata.