fix(python,rust): str.concat on empty list

[Julian-J-S]

aggregation methods should always return a single value, otherwise it can lead to unexpected behavior and bugs (e.g. #12053)

current behavior:

• empty list: return Series of shape (0,) >>>>> this is the bug
• otherwise: return Series of shape (1,) with the concatenated/joined string

PR:

• always return Series of shape (1,) with the concatenated/joined string
• if empty list, return empty string (as discussed here:str.concat should skip nulls and be null for empty groups #10814 (comment))

pl.Series([], dtype=pl.Utf8).str.concat()

# OLD
shape: (0,)
Series: '' [str]
[
]

# NEW
shape: (1,)
Series: '' [str]
[
	""
]

Note: this does not address the issue of skipping nulls (#10814) but only fixes the bug for empty lists

[reswqa]

I personally +1 for this change and it keep the behavior before #11488.

I will take a look at the remaining issue of skipping nulls.

[Julian-J-S]

I will take a look at the remaining issue of skipping nulls.

If you work on this you should probably look into here #6919
I wrote a small summary of the problem and ritchie also said:

I would accept a PR with "ignore" and "raise" as null_behavior arguments.

So I think the goal is to either "ignore" or "raise" if nulls occur (current problem is that they appear as null in text form :D)

[reswqa]

So I think the goal is to either "ignore" or "raise" if nulls occur (current problem is that they appear as null in text form :D)

I'm a little hesitant to add the raise option, for the same reasons mentioned in #10814 (comment). And I also support ignore and propogate null(which means ["foo", null, "bar"].str.concat() == null). Of course, I'm open to discussion :)

[orlp]

@reswqa I would greenlight implementation with ignore and propagate, with ignore as the default. We can always add a raise option later if we do change our mind.

[reswqa]

@orlp Ah, make sense. I will draft this recently.