Description

I noticed str.slice(offset, length=0) gives different results depending on the backend. On pandas and dask it returns the rest of the string, but on Polars, PyArrow and the SQL backends it correctly returns an empty string:

import narwhals as nw, pandas as pd, polars as pl
data = {"s": ["abcdef", "xyz"]}
nw.from_native(pd.DataFrame(data)).select(nw.col("s").str.slice(1, 0)).to_native()["s"].tolist()
# ['bcdef', 'yz']   <- wrong
nw.from_native(pl.DataFrame(data)).select(nw.col("s").str.slice(1, 0)).to_native()["s"].to_list()
# ['', '']          <- correct

The cause is in _pandas_like/series_str.py and _dask/expr_str.py, which compute the stop as offset + length if length else None. A length of 0 is falsy, so it's treated the same as None ("slice to the end"). The Arrow and SQL backends already use length is not None, so this just brings pandas and dask in line with them.

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

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

Related issues

Found while reading the string namespace, no existing issue.

AI assistance

• No AI tools were used for this PR.
• AI tools were used.

I used Claude Code to help find the divergence and draft the test. I read the code, reproduced it across all backends, and understand and stand behind every line.

Checklist

• Code follows style guide (ruff)
• Tests added
• Documented the changes (behaviour fix, no doc change needed)
• If this is your first PR to narwhals, attach a screenshot of pytest passing locally (not CI):

This is my first PR here. I can't attach an image in the PR body, so here is the terminal output of the slice tests passing across the backends I have installed locally (pandas, pandas[pyarrow], pyarrow, polars eager+lazy, dask):

$ pytest tests/expr_and_series/str/slice_test.py \
    --constructors=pandas,pandas[pyarrow],pyarrow,polars[eager],polars[lazy],dask

tests/expr_and_series/str/slice_test.py ..............................   [100%]
============================== 30 passed in 0.61s ==============================

The new (1, 0, {"a": ["", ""]}) case fails on main for pandas/dask and passes with this change; happy to attach a screenshot too if you'd prefer.