pd.concat accepts iterables that may contain Series, DataFrame and None, where None are simply ignored.
pd.concat([None, series, None, series, None]) # works, same as pd.concat([series, series])
pd.concat([None]) # raises
It would be great to deprecate/remove accepting Nones. This will help to better annotate concat https://github.com/pandas-dev/pandas-stubs/issues/888
Pandas-stubs currently has to choose between either false-positives (accept concat([None]) ) or false-negatives (reject some concat(Iterable[Any/Unknown]) that could succeed).
Comment From: phofl
I am -1 on simply deprecating this because of typing reasons. Let's think about use cases before we make a decision. We definitely can't do this without a deprecation
Comment From: twoertwein
I agree, it would be good first to understand which use case supporting None enables!
@Dr-Irv found a way to get it to type check without having to decide between false negatives/positives :)
Comment From: Multihuntr
If the function should accept None's in objs, then it seems most natural to me for it to return an empty DataFrame if objs contains only None's.
This is backwards compatible and would also resolve the typing issue.
A hypothetical use case for Iterable[DataFrame | None]: you have a list of files to load and parse and you want to put them into a single DataFrame, but not all of the filenames you've been given exist, or maybe parsing is allowed to fail. I could see someone implementing this as a function that returns DataFrame | None, and putting that into a list comprehension. So a situation where you just want to put everything you have into one table, and if everything you have is None, then your table is empty.
Comment From: jbrockmendel
I was surprised to find this works. Looks like Nones are filtered out in _clean_keys_and_objs, which has a comment pointing to #1649 back in 2012. I don't see much conversation there about whether this should be allowed.
+1 on deprecating.