Sync fork main with upstream

[nelson2005]

Brings fork main level with upstream/main, which was two commits ahead: c10525b (improved mapinarrow_factory) and 0f696a9 (updated authors, license).

Merged, not squashed, so upstream's commits keep their ancestry and do not reappear as unmerged at the next sync.

Conflict resolution

One conflict, in numbarrow_ci.yml, both hunks in the shared build job:

• python-version — upstream pins "3.12", the fork drives it from matrix.python-version. Kept the fork's matrix.
• dependency install — upstream bumped pandas 2.2.3 → 2.3.2 on a line the fork had also changed to add numba==${{ matrix.numba-version }}. Kept both: the fork's numba matrix line and upstream's pandas bump.

The fork-only build-macos job never conflicted, so it kept pandas==2.2.3 and would have tested a different pandas than Linux and Windows. Aligned to 2.3.2 in a follow-up commit on this branch.

requires-python raised to >=3.12

0f696a9 also raised requires-python from >=3.10 to >=3.12 (alongside numba<0.67.0, pyarrow<=24.0.0, and a setuptools test dep). That made every 3.10 and 3.11 matrix job fail at pip install -e .:

ERROR: Package 'numbarrow' requires a different Python: 3.11.15 not in '>=3.12'

The floor is metadata only — nothing in the code or the new dependency ceilings needs 3.12. Measured on this branch against pyarrow 24.0.0:

CPython	numba	result
3.10.20	0.60.0	34 passed
3.10.20	0.63.0	34 passed
3.11.15	0.60.0	34 passed
3.11.15	0.63.0	34 passed

python -m build --wheel also succeeds on 3.10.

So both jobs now install with --ignore-requires-python, keeping the existing three-interpreter coverage rather than dropping two. This deliberately tests below upstream's declared floor: a real user on 3.10 is refused by pip until upstream changes it, so these jobs are regression signal rather than a supported configuration.

Verified after the merge

• numbarrow/, LICENSE, pyproject.toml byte-identical to upstream/main
• README.md differs by the two pre-existing fork-only lines defining df_in / output_schema in the example, which the fork's own doc-codeblock check requires and upstream has no equivalent of
• fork-only files all present: .github/scripts/extract_codeblocks.py, doc-codeblock-flake8.yml, link-check.yml, CLAUDE.md
• no conflict markers; workflow YAML parses; matrix intact at 3 python × 2 numba × 3 arch, plus macOS

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[github-actions]

MiniMax Code Review

The user is asking me to review a pull request with multiple changes. Let me analyze each file change:

1. .github/workflows/numbarrow_ci.yml - Modified

   • Updated pandas from 2.2.3 to 2.3.2
   • Added --ignore-requires-python flag to pip install -e .
   • These changes appear in two places (matrix jobs and a specific job)

2. .github/workflows/numbarrow_release.yml - Modified

   • Changed Python version from 3.10 to 3.12

3. CLAUDE.md - Modified

   • Updated Python requirement info
   • Added info about CI testing below declared floor
   • Removed specific dependency versions

4. LICENSE - Modified

   • Changed copyright from 2025 Mikhail Goykhman to 2026 NumbArrow GitHub Repository Contributors

5. numbarrow/core/mapinarrow_factory.py - Modified

   • Changed how arrow_array_adapter return value is unpacked
   • Previously: col_bitmap, col_data = arrow_array_adapter(col_pa)
   • Now: vals = arrow_array_adapter(col_pa) then *_, col_bitmap, col_data = vals
   • This suggests the function now returns more values than before

6. pyproject.toml - Modified

   • numba: 0.60.0,<0.64.0 → 0.60.0,<0.67.0
   • pyarrow: <=15.0.0 → <=24.0.0
   • requires-python: >=3.10 → >=3.12
   • Author changed from Mikhail Goykhman to "NumbArrow GitHub Repository Contributors"
   • Added setuptools>=75.0.0 to test dependencies

Now let me review each change for issues:

pyproject.toml changes:

The shift to Python 3.12 as a minimum requirement represents a substantial compatibility leap. The dependency expansions for numba and pyarrow are reasonable, but the version bumps could introduce breaking changes. The --ignore-requires-python flag seems like a potential workaround that might mask underlying compatibility problems.

CLAUDE.md changes:
The documentation's current framing about CI testing below the declared floor creates confusion. The language around regression signals and supported configurations appears contradictory and potentially misleading for contributors.

mapinarrow_factory.py changes:
The unpacking modification reveals potential fragility. Using *_ to discard initial values risks silently ignoring critical return values, which could lead to unexpected runtime behavior or subtle bugs.

General concerns:
The workflow modifications suggest a gradual relaxation of version constraints, potentially compromising project stability. The explicit version tolerance might encourage less rigorous testing and dependency management.

Code Review: Pull Request

Summary

This PR makes several coordinated changes: updating Python version to 3.12, updating dependencies (numba, pyarrow, pandas), and modifying CI to test "below the floor" with --ignore-requires-python. There are some issues to address.

---

Issues Found

1. CLAUDE.md: Conflicting Documentation (Medium)

The CLAUDE.md says both:

• "CI deliberately tests below the declared floor"
• "Treat those jobs as regression signal, not as a supported configuration"

This is contradictory. If the package "still builds, imports and passes the full suite on 3.10 and 3.11", then 3.10/3.11 are effectively supported. The framing suggests this is a workaround to avoid updating requires-python, but the real fix is to set requires-python = ">=3.12" (as done in pyproject.toml) and have CI test exactly what users will actually install.

Recommendation: Remove the --ignore-requires-python workaround and have CI test exactly what requires-python declares. This is cleaner and avoids confusing users.

---

2. numbarrow/core/mapinarrow_factory.py: Fragile Unpacking (Medium)

vals = arrow_array_adapter(col_pa)
*_, col_bitmap, col_data = vals

This discards all but the last two return values using *_. This is fragile because:

• If arrow_array_adapter returns additional values in the future, they'll be silently ignored
• It hides what the actual return structure is

Recommendation: Either:

• Explicitly unpack the known return values if the function returns a fixed number of items
• Or add a comment documenting what values are being discarded and why

---

3. pyproject.toml: Unbounded Upper Version Limits (Low)

"pyarrow<=24.0.0"

This upper bound will continue to require updates as PyArrow releases new versions. Consider:

• Testing against the latest PyArrow in CI
• Using a more relaxed bound or no upper bound at all

---

4. LICENSE: Copyright Change (Info)

Changed from "Mikhail Goykhman" to "NumbArrow GitHub Repository Contributors". Ensure this change is intentional and that Mikhail Goykhman (the original author) has agreed to this re-licensing/copyright change.

---

5. CI Workflow: Duplicate Changes (Low)

The same two changes (pandas version, --ignore-requires-python) appear in two separate job sections in numbarrow_ci.yml. Consider using a workflow matrix or consolidated steps to avoid duplication.

---

Positive Changes

• ✅ Updated pandas to 2.3.2 (recent stable version)
• ✅ numba upper bound updated to 0.67.0 (accommodates newer numba)
• ✅ pyarrow upper bound significantly expanded (up to 24.0.0)
• ✅ Python 3.12 as minimum is reasonable for 2025/2026

---

Action Items

1. Fix the *_ unpacking in mapinarrow_factory.py to be more explicit
2. Reconsider the --ignore-requires-python strategy — CI should test what users will actually experience
3. Verify copyright change is agreed upon by original author

[Copilot]

Pull request overview

This PR syncs the fork’s main branch with upstream/main, bringing in upstream changes to packaging metadata, licensing/authorship, and PySpark mapInArrow integration, plus aligning CI dependency pins.

Changes:

• Updated project metadata and dependency constraints in pyproject.toml (Python/Numba/PyArrow bounds; test extras).
• Adjusted make_mapinarrow_func to accept the newer multi-value return shape from arrow_array_adapter.
• Updated GitHub Actions workflows to run/build with Python 3.12 and align pandas pinning across jobs; updated LICENSE header.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
pyproject.toml	Updates supported Python version and dependency bounds; adds a test extra requirement.
numbarrow/core/mapinarrow_factory.py	Updates adapter unpacking logic used to build data_dict / bitmap_dict for mapInArrow.
LICENSE	Updates copyright holder/year to reflect upstream changes.
.github/workflows/numbarrow_release.yml	Moves release build environment to Python 3.12.
.github/workflows/numbarrow_ci.yml	Aligns pandas pin across build jobs; retains python/numba matrices.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.