Changelog

Added

• Data Quality Engine: Added profile(), DataQualityReport, and ColumnProfile for dataset diagnostics.
• Smart Cleaning: Added suggest_cleaning() and auto_clean() with safe and strict modes.
• Schema Validation: Added Schema, Field, validate(), validation result objects, and field builders for Int64, Float64, String, Bool, Email, and URL.

Fixed

• Hardened CSV parsing for quoted multiline records, unterminated quotes, duplicate headers, empty headers, and non-UTF-8 encodings.
• Improved Python-level errors by wrapping native read failures as CsvReadError.
• Improved pandas conversion for mixed numeric/object columns and rejected unsafe nested objects.

Documentation

• Redesigned the README with clearer product positioning, architecture notes, benchmarks, and contribution paths.
• Added language identifiers to fenced code blocks for cleaner Markdown linting.

Added

• Automated Release Pipeline: Integrated Google Release Please for automated versioning and changelog generation.
• Architecture Documentation: New internal documentation explaining the Python-to-C++ boundary and memory management.
• CI/CD Resilience: Fixed GITHUB_TOKEN trigger limitations in the release workflow using reusable workflows.

Fixed

• Fixed hardcoded relative paths in test fixtures that caused cibuildwheel failures on CI.
• Removed deprecated inputs in the Release Please workflow for v4 compatibility.

Added

• Cross-Platform Wheels: Full cibuildwheel automation delivering pre-compiled native wheels for Windows, Linux, and macOS (Intel & Apple Silicon).
• Google Colab Compatibility: Linux wheels are now fully manylinux compliant, allowing pip install arnio to work out-of-the-box on Colab and Ubuntu.
• Production-Grade Packaging: Resolved ModuleNotFoundError by removing double-nesting issues in scikit-build-core config.
• CI/CD Excellence: Fully automated PyPI publishing pipeline via Trusted Publishing with integrated source distributions (sdist).
• Stable API: Officially marked arnio as stable for production workloads with "Development Status :: 5 - Production/Stable".

Fixed

• Migrated from FetchContent to find_package(pybind11) for faster, offline, and more robust cross-platform builds.
• Refactored cibuildwheel configuration entirely into pyproject.toml for standard and declarative packaging.

Fixed

• normalize_case() now accepts case_type kwarg as documented in README (previously accepted case=, causing TypeError for all README users).
• to_pandas() completely rewritten using zero-copy NumPy buffer interface — eliminates O(rows × cols) pybind11 boundary crossings, restoring actual performance advantage over pandas.
• from_pandas() implemented with correct null handling and round-trip fidelity.

Added

• ar.register_step(name, fn) — register pure-Python pipeline steps without C++.
• arnio.exceptions module with ArnioError, UnknownStepError, CsvReadError, TypeCastError — replaces opaque C++ errors with actionable messages.
• arnio.__version__ now available programmatically.
• benchmarks/generate_data.py — deterministic 1M row test dataset generator.
• benchmarks/benchmark_vs_pandas.py — reproducible end-to-end benchmark.

Fixed (Internal)

• CI now verifies compilation on Ubuntu and Windows across Python 3.9–3.12.

Fixed

• Stability improvements and initial PyPI release.