Skip to content

Changelog

All notable changes to this project are documented here.

The format follows Keep a Changelog. This project uses semantic versioning.

Unreleased

Added

  • Initial project structure (src/ layout, pyproject.toml, mkdocs.yml)
  • Package skeleton: FrozenFrame, field, FrozenSeries, FrozenFrameError, SchemaValidationError stubs
  • roadmap.md and motivations.md project documentation
  • MkDocs Material documentation site with full guide, API reference, and contributing pages
  • ruff (linting + formatting) and ty (type checking) configured in pyproject.toml
  • column.py — Phase 1 implementation complete:
  • _PYTHON_TO_ARROW default type map covering bool, int, float, str, bytes, datetime.datetime, datetime.date, datetime.timedelta, decimal.Decimal
  • register_type() for global third-party type registration
  • field() for optional per-column Arrow type overrides and documentation
  • resolve_arrow_type() internal resolver handling bare types, T | None unions, Optional[T], and field(arrow_type=...) overrides
  • 28 tests for column.py covering all type mappings, nullable unions, overrides, error cases, field(), and register_type()
  • schema.py — Phase 1 implementation complete:
  • build_schema(annotations, defaults) builds a pa.Schema from a class's __annotations__ and any field() defaults, respecting arrow_type overrides and T | None nullability
  • validate(schema, batch) checks a pa.RecordBatch for extra columns, missing columns, type mismatches, and nullability violations — each with a descriptive error message naming the offending column
  • 21 tests for schema.py covering basic types, nullability, field() overrides, column ordering, error cases, and passing validation
  • frame.py — Phase 1 implementation complete:
  • FrozenFrameMeta metaclass with @dataclass_transform() — compiles __annotations__ into cls.__schema__ at class creation time, resolves forward references via get_type_hints, collects field() defaults
  • FrozenFrame base class with from_dict, from_arrow, and _from_batch (internal fast-path) constructors
  • Mutation guards: __setattr__, __delattr__, __setitem__, __delitem__ all raise FrozenFrameError
  • Column access via df["col"] and df.col (returns pa.Array; upgrades to FrozenSeries in next step)
  • __hash__ — SHA-256 of Arrow IPC serialisation, cached after first computation
  • __eq__, __len__, __iter__ (row-wise dicts), __repr__
  • 45 tests for frame.py covering metaclass schema compilation, construction, mutation guards, column access, iteration, equality, hashing, and repr
  • __init__.py — public API wired up: FrozenFrame, field, register_type, FrozenFrameError, SchemaValidationError all importable directly from freezeframe
  • series.py — Phase 1 implementation complete:
  • FrozenSeries[T] wraps a pa.Array with a fully immutable interface
  • __len__, __iter__ (yields Python values; nulls become None), __getitem__ (int → scalar, slice → new FrozenSeries)
  • Comparison operators (==, !=, <, <=, >, >=) via pyarrow.compute — return pa.BooleanArray masks for use with filter() in Phase 2
  • equals() for value equality between two series (returns bool, null-aware)
  • to_pylist(), to_arrow(), type, null_count, __repr__
  • Explicitly __hash__ = None — unhashable by design since __eq__ is element-wise
  • FrozenFrame.__getitem__ and __getattr__ updated to return FrozenSeries instead of pa.Array
  • FrozenSeries added to public API exports in __init__.py
  • 45 tests for series.py covering construction, immutability, sequence protocol, all six comparison operators (scalar and series operands), equals(), conversion helpers, and repr