fix: return null for all-null rows in mean_horizontal (#4056)

[ioseba]

Description

For rows where every input column is null, mean_horizontal previously divided 0 / 0, returning NaN instead of null on PyArrow, DuckDB, and pandas[pyarrow]. Because NaN is not considered null on PyArrow and DuckDB, is_null() unexpectedly evaluated to False.

This PR:

• In _arrow/namespace.py: uses pc.if_else to return scalar null when non-null count is 0.
• In _duckdb/namespace.py: uses when(count > 0, total / count).otherwise(lit(None)).
• In _pandas_like/namespace.py: uses .zip_with(count > 0, None) to set None/<NA> when count is 0, resolving the issue for pandas[pyarrow] as well.
• Adds regression unit tests in tests/expr_and_series/mean_horizontal_test.py covering all constructors and duckdb.

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

• 🐛 Bug Fix
• ✅ Test

Related issues

• Closes [Bug]: mean_horizontal returns NaN instead of null for all-null rows on PyArrow and DuckDB #4056

AI assistance

• AI tools were used.

Checklist

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

Summary by CodeRabbit

• Bug Fixes
  • Horizontal means now return null when all input values in a row are null, consistently across supported backends.
  • Rows with a mix of null and non-null values continue to average only the non-null values.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

• Configuration used: defaults
• Review profile: CHILL
• Plan: Advanced
• Run ID: c202684a-4393-4388-af49-db4634ff69c7

---

📥 Commits

Reviewing files that changed from the base of the PR and between 4d6f7b0 and 31e5072.

---

📒 Files selected for processing (4)

• src/narwhals/_arrow/namespace.py
• src/narwhals/_duckdb/namespace.py
• src/narwhals/_pandas_like/namespace.py
• tests/expr_and_series/mean_horizontal_test.py

---

Included review availability: This review used your included allowance. Your plan provides up to 4 included reviews per hour; 3 remain after this review.

---

---

📝 Walkthrough

Walkthrough

Horizontal mean implementations in Arrow, DuckDB, and pandas-like backends now return null for rows where all inputs are null. Tests cover these rows and rows with non-null values.

Changes

Horizontal Mean Null Handling

Layer / File(s)	Summary
Guard horizontal means against zero counts src/narwhals/_arrow/namespace.py, src/narwhals/_duckdb/namespace.py, src/narwhals/_pandas_like/namespace.py	Each backend now returns null when a row has no non-null inputs.
Test horizontal mean results tests/expr_and_series/mean_horizontal_test.py	Tests cover all-null rows and rows with values for a constructor-backed dataframe and a DuckDB lazy dataframe. The DuckDB test skips when DuckDB is unavailable.

Priority: ⬇️ Low

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix · Severity of issue fixed: Low

Suggested reviewers: fbruzzesi

---

Merge Risk | ⚪ Minimal · up to 31e50

Merge Risk: ⚪ Minimal · up to 31e50

The change is mergeable after normal checks; no actionable risk has been identified.

Security Architecture Review

Security architecture risk: ⚪ Minimal · up to 31e50

The change is limited to numerical result handling and regression tests. It preserves existing inputs and execution paths, with no identified expansion of access, privileges, or security-sensitive behavior.

Retained concerns
No architecture-level concerns identified.

Security review details

Security Blast Radius

• inferred — The supported exposure is a result-semantics change for callers already able to evaluate horizontal means on these backends. The inspected changes do not expand the set of evaluated input expressions or introduce a new privileged operation.

---

Pre-merge checks | 4 | 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage		Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%. Docstring coverage is scoped to functions touched by this diff. Analyzed 9 functions across 4 files.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check		The title clearly and concisely identifies the main bug fix: returning null for all-null rows in mean_horizontal.
Description check		The description explains the bug, identifies the affected backends, summarizes the implementation, references the related issue, and completes the relevant checklist items.
Linked Issues check		Issue #4056 requires a null result and is_null() == True when every input is null across pandas, Polars, PyArrow, and DuckDB. The PR masks zero-count rows in ArrowNamespace.mean_horizontal, `DuckD…
Out of Scope Changes check		The changes stay within issue #4056. The namespace changes implement zero-count null handling, and the added tests verify the reported regression and preserve valid means. No unrelated production or t…

---

• Fix all pre-merge checks with AI

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create a new PR

---

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}

[FBruzzesi]

Hi @ioseba thanks for the contribution and welcome to Narwhals. Sadly this is a duplicate of #4059 .
I hope to see you keep contributing in the future, focusing on issues that are still open and no one is working on them

[ioseba]

Thanks for the heads up and the warm welcome @FBruzzesi!

My apologies, I hadn't seen the other PR when submitting. Totally understand, no worries at all! I will keep an eye on open issues that need a hand to contribute to soon. Keep up the great work with Narwhals! 🚀