feat: concat_str

[FBruzzesi]

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

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

Related issues

• Closes [Enh]: Add nw.concat_str to horizontally concatenate string columns #1116

Checklist

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

If you have comments or can explain your changes, please do so below.

This was a bit harder/more complex than expected to get right. The idea being that polars can switch between:

• ignore_nulls=False (default), which means that a single null in the row will be propagated to the final result.
• ignore_nulls=True, when this happens, it is not as simple as replacing the null with an empty string, since otherwise the separator will get repeated (I am using a reduce). Also the separator "attached" to the same column and row needs to be nullified to empty string.

All this comes out of the box with polars>=0.20.6 and pyarrow (not sure about min version, CI will tell). On the other hand for pandas, dask and polars<0.20.6, a fair bit of manipulation had to be done, especially for ignore_nulls=True.

The gist of it is using a null_mask, a list of series corresponding to the null values of the original series, and use it to map both the null values and the separators to an empty string.

It has also been a long day, so maybe tomorrow I can take another look with some fresh eyes

[FBruzzesi]

@MarcoGorelli I can't reproduce the CI failure even with the same pinned versions (screenshot attached)

Can reproduce by lowering pandas version 🤔

Fixed via the following change:

+ series = (s.astype(str) for _expr in parsed_exprs for s in _expr._call(df))
- series = (s for _expr in parsed_exprs for s in _expr.cast(self._dtypes.String()._call(df))

[MarcoGorelli]

nice one @FBruzzesi, great use of pl.fold!