fix: Add fallback for incompatible string concenation in pandas

[MarcoGorelli]

closes #3546

Description

What type of PR is this? (check all applicable)

• 💾 Refactor
• ✨ Feature
• 🐛 Bug Fix
• 🔧 Optimization
• 📝 Documentation
• ✅ Test
• 🐳 Other

Related issues

• Related issue #<issue number>
• Closes #<issue number>

Checklist

• Code follows style guide (ruff)
• Tests added
• Documented the changes

[dangotbanned]

Note

Edited to increase the range

Have you thought about handling the dtypes somewhere around here instead?

narwhals/narwhals/_pandas_like/namespace.py

Lines 320 to 328 in 2eaaa03

	def concat_str(
	self, *exprs: PandasLikeExpr, separator: str, ignore_nulls: bool
	) -> PandasLikeExpr:
	string = self._version.dtypes.String()
	def func(df: PandasLikeDataFrame) -> list[PandasLikeSeries]:
	expr_results = [s for _expr in exprs for s in _expr(df)]
	series = [s.cast(string) for s in expr_results]

E.g. you could do 2 passes where:

1. Collect all the native dtypes
2. Decide on what's needed to make them compatible & then do the casts at the native-level

This is reminding me a bit of #3398 (comment) 😄

[MarcoGorelli]

not sure about doing it for all ops tbh, but for string concatenation at least it seems to be necessary so i'd suggest starting with that, we can make it more generally available later if necessary

[dangotbanned]

not sure about doing it for all ops tbh, but for string concatenation at least it seems to be necessary so i'd suggest starting with that, we can make it more generally available later if necessary

Sorry if I wasn't clear, I was talking about concat_str - have edited (#3548 (comment))

[MarcoGorelli]

we'd then need to do the same casts for __add__, so i think it's cleaner to do just it in one place?

[dangotbanned]

we'd then need to do the same casts for __add__

I was a bit confused by this, I had no idea polars did it too 😂

import polars as pl

>>> pl.Series(["a", "b", "c"]) + pl.Series(["d", "e", "f"])
shape: (3,)
Series: '' [str]
[
	"ad"
	"be"
	"cf"
]