#22212 refactored the cudf_polars test suite to eventually parametrize over pytest fixtures of the engines we want to support (SPMD, Dask, Ray). We mostly pass this engine fixture to our assertion function in python/cudf_polars/cudf_polars/testing/asserts.py, but
assert_gpu_result_equal and assert_sink_result_equal provides the "default" if not specified
|
engine = engine or get_default_engine(executor) |
|
engine = engine or get_default_engine(executor) |
assert_ir_translation_raises, assert_collect_raises, and assert_sink_ir_translation_raises don't provide a way to override a hardcoded engine
|
translator = Translator(q._ldf.visit(), GPUEngine()) |
|
engine = GPUEngine(raise_on_fail=True) |
therefore we are potentially not testing the pytest engine fixture in our test.
We should change all these assertion function to require an engine argument to avoid this.
(This was discovered in #22281 when changing the default to rapidsmpf change the behavior of some of these assertions)
cc @madsbk when you're adding Ray testing
#22212 refactored the cudf_polars test suite to eventually parametrize over pytest fixtures of the engines we want to support (SPMD, Dask, Ray). We mostly pass this engine fixture to our assertion function in
python/cudf_polars/cudf_polars/testing/asserts.py, butassert_gpu_result_equalandassert_sink_result_equalprovides the "default" if not specifiedcudf/python/cudf_polars/cudf_polars/testing/asserts.py
Line 358 in d70bda4
cudf/python/cudf_polars/cudf_polars/testing/asserts.py
Line 100 in d70bda4
assert_ir_translation_raises,assert_collect_raises, andassert_sink_ir_translation_raisesdon't provide a way to override a hardcoded enginecudf/python/cudf_polars/cudf_polars/testing/asserts.py
Line 164 in d70bda4
cudf/python/cudf_polars/cudf_polars/testing/asserts.py
Line 293 in d70bda4
cudf/python/cudf_polars/cudf_polars/testing/asserts.py
Line 422 in d70bda4
therefore we are potentially not testing the pytest engine fixture in our test.
We should change all these assertion function to require an engine argument to avoid this.
(This was discovered in #22281 when changing the default to rapidsmpf change the behavior of some of these assertions)
cc @madsbk when you're adding Ray testing