v0.7.0

0.7.0 (2026-05-07)

Features

• Polars support (f8ba5ef)

  dataframe-expectations now supports Polars DataFrames alongside Pandas and PySpark. Polars is an optional dependency and must be explicitly requested:

pip install dataframe-expectations[polars] # pandas + polars
pip install dataframe-expectations[pyspark,polars] # pandas + pyspark + polars

All existing expectations that have a Polars implementation will automatically dispatch to the
correct validation path when a Polars DataFrame is passed to a suite.

Polars imports are fully deferred behind @lru_cache helpers (mirroring the PySpark pattern),
so importing dataframe_expectations has zero overhead when Polars is not installed.

• Backward-compatible API for expectation authors (52dda74)

The fn_violations_polars parameter on ColumnExpectation is optional (None by default),
so existing custom expectations that only implement Pandas and/or PySpark continue to work
without changes. @abstractmethod has been removed from validate_pandas, validate_pyspark,
and validate_polars — each raises NotImplementedError by default, allowing expectation
authors to implement only the backends they need.

• CI updated to cover Polars install scenarios (b636a15)

Job	Extras installed	Tests run
tests-without-optional	None	-m "not pyspark and not polars"
tests-with-pyspark	[pyspark]	-m "not polars"
tests-with-polars	[polars]	-m "not pyspark"
tests-with-all	[pyspark,polars]	All

Bug Fixes

• Tightened DataFrame type checks (15d1df7)

is_polars_data_frame() and is_pyspark_data_frame() no longer use __module__ name
fallbacks. Only actual DataFrame instances (via isinstance) are accepted, preventing
Series, Expr, LazyFrame, Column, Row, etc. from being mis-routed into validation paths.

Documentation

• Getting Started guide updated with tabbed examples (ee53581)

Installation and usage examples now use sphinx-design tab sets showing Pandas, PySpark, and
Polars side-by-side. sphinx-design>=0.6.0 added as a docs dependency.

Test Improvements

• Refactored test fixtures to use PyArrow (f9692bd)

Test data is now defined once via PyArrow tables and converted to Pandas, PySpark, or Polars
DataFrames through a dataframe_factory fixture, eliminating duplicated test data across
backends. All expectation tests are parametrized over [pandas, pyspark, polars].

• Exhaustive match/case blocks — all match df_lib blocks in tests now include
  case _: pytest.fail(...) to catch unhandled DataFrame types immediately.