Chore: Add CI step to test without pyarrow

.github/workflows/testing.yml

@@ -20,7 +20,8 @@ jobs:
             ci-py309-min-optional-deps,
             ci-py311-transformers,
             ci-py313-latest-deps,
-            ci-py313-latest-optional-deps
+            ci-py313-latest-optional-deps,
+            ci-py313-polars-without-pyarrow
         ]
     runs-on: ${{ matrix.os }}
     steps:

pyproject.toml

@@ -168,6 +168,10 @@ pyarrow = "==15.0.2"
 polars = "==0.20.15"
 plotly = "*"
 
+[tool.pixi.feature.polars-without-pyarrow.dependencies]
+polars = "*"
+plotly = "*"
+
 [tool.pixi.feature.py309.dependencies]
 python = "~=3.9.0"
 
@@ -211,6 +215,7 @@ ci-py311-transformers = ["py311", "transformers", "test"]
 ci-py312-latest-deps = ["py312", "test"]
 ci-py312-latest-optional-deps = ["py312", "optional", "test"]
 ci-py313-latest-deps = ["py313", "test"]
+ci-py313-polars-without-pyarrow = ["py313", "test", "polars-without-pyarrow"]
 ci-py313-latest-optional-deps = ["py313", "optional", "test"]
 
 [tool.black]

skrub/_dataframe/tests/test_common.py

@@ -18,6 +18,7 @@
 import skrub
 from skrub import selectors as s
 from skrub._dataframe import _common as ns
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 def test_not_implemented():
@@ -103,6 +104,7 @@ def test_to_numpy(df_module, example_data_dict):
     assert_array_equal(array[2:], np.asarray(example_data_dict["str-col"])[2:])
 
 
+@polars_installed_without_pyarrow
 def test_to_pandas(df_module, pd_module):
     with pytest.raises(TypeError):
         ns.to_pandas(np.arange(3))
@@ -725,6 +727,7 @@ def test_mean(df_module):
     )
 
 
+@polars_installed_without_pyarrow
 def test_corr(df_module):
     df = df_module.example_dataframe
 

skrub/_reporting/tests/test_patch_display.py

@@ -3,10 +3,12 @@
 import pytest
 
 from skrub import patch_display, unpatch_display
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 @pytest.mark.parametrize("repeat_patch", [1, 2])
 @pytest.mark.parametrize("repeat_unpatch", [1, 2])
+@polars_installed_without_pyarrow
 def test_patch_display(df_module, repeat_patch, repeat_unpatch, capsys):
     df = df_module.make_dataframe(
         dict(

skrub/_reporting/tests/test_summarize.py

@@ -9,11 +9,13 @@
 from skrub import _dataframe as sbd
 from skrub._reporting import _sample_table
 from skrub._reporting._summarize import summarize_dataframe
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 @pytest.mark.parametrize("order_by", [None, "date.utc", "value"])
 @pytest.mark.parametrize("with_plots", [False, True])
 @pytest.mark.parametrize("with_associations", [False, True])
+@polars_installed_without_pyarrow
 def test_summarize(
     monkeypatch, df_module, air_quality, order_by, with_plots, with_associations
 ):
@@ -114,6 +116,7 @@ def test_high_cardinality_column(pd_module):
     assert "10 most frequent" in summary["columns"][0]["value_counts_plot"]
 
 
+@polars_installed_without_pyarrow
 def test_all_null(df_module):
     df = df_module.make_dataframe(
         {
@@ -142,6 +145,7 @@ def make_summary(df_size, **summarize_kwargs):
     return make_summary
 
 
+@polars_installed_without_pyarrow
 def test_small_df(small_df_summary):
     summary = small_df_summary(11)
     thead, first_slice, ellipsis, last_slice = summary["sample_table"]["parts"]
@@ -256,6 +260,7 @@ def test_duplicate_columns(pd_module):
     assert cols[1]["mean"] == 3.5
 
 
+@polars_installed_without_pyarrow
 def test_high_cardinality_columns(df_module):
     df = df_module.make_dataframe(
         {
@@ -269,6 +274,7 @@ def test_high_cardinality_columns(df_module):
     assert cols[1]["is_high_cardinality"]
 
 
+@polars_installed_without_pyarrow
 def test_bool_column_mean(df_module):
     df = df_module.make_dataframe({"a": [True, False, True, True, False, True]})
     summary = summarize_dataframe(df)

skrub/_reporting/tests/test_table_report.py

@@ -11,12 +11,14 @@
 from skrub import TableReport, ToDatetime, datasets
 from skrub import _dataframe as sbd
 from skrub._reporting._sample_table import make_table
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 def get_report_id(html):
     return re.search(r'<skrub-table-report.*?id="report_([a-z0-9]+)"', html).group(1)
 
 
+@polars_installed_without_pyarrow
 def test_report(air_quality):
     col_filt = {
         "first_2": {
@@ -64,11 +66,13 @@ def test_report(air_quality):
     assert len(all_report_ids) == len(set(all_report_ids))
 
 
+@polars_installed_without_pyarrow
 def test_few_columns(df_module, check_polars_numpy2):
     report = TableReport(df_module.example_dataframe)
     assert "First 10 columns" not in report.html()
 
 
+@polars_installed_without_pyarrow
 def test_few_rows(df_module, check_polars_numpy2):
     df = sbd.slice(df_module.example_dataframe, 2)
     TableReport(df).html()
@@ -84,13 +88,15 @@ def test_open(pd_module, browser_mock):
     assert b"the title" in browser_mock.content
 
 
+@polars_installed_without_pyarrow
 def test_non_hashable_values(df_module):
     # non-regression test for #1066
     df = df_module.make_dataframe(dict(a=[[1, 2, 3], None, [4]]))
     html = TableReport(df).html()
     assert "[1, 2, 3]" in html
 
 
+@polars_installed_without_pyarrow
 def test_nat(df_module):
     # non-regression for:
     # https://github.com/skrub-data/skrub/issues/1111
@@ -103,6 +109,7 @@ def test_nat(df_module):
     TableReport(df).html()
 
 
+@polars_installed_without_pyarrow
 def test_bool_column_mean(df_module):
     df = df_module.make_dataframe({"a": [True, False, True, True, False, True]})
     html = TableReport(df).html()
@@ -116,6 +123,7 @@ def test_duplicate_columns(pd_module):
     TableReport(df).html()
 
 
+@polars_installed_without_pyarrow
 def test_infinite_values(df_module):
     # Non-regression for https://github.com/skrub-data/skrub/issues/1134
     # (histogram plot failing with infinite values)
@@ -129,6 +137,7 @@ def test_infinite_values(df_module):
     TableReport(df).html()
 
 
+@polars_installed_without_pyarrow
 def test_duration(df_module):
     df = df_module.make_dataframe(
         {"a": [datetime.timedelta(days=2), datetime.timedelta(days=3)]}
@@ -187,6 +196,7 @@ def test_write_html_with_not_utf8_encoding(tmp_path, pd_module):
     assert "</html>" not in saved_content
 
 
+@polars_installed_without_pyarrow
 def test_verbosity_parameter(df_module, capsys):
     df = df_module.make_dataframe(
         dict(
@@ -209,6 +219,7 @@ def test_verbosity_parameter(df_module, capsys):
     assert capsys.readouterr().err != ""
 
 
+@polars_installed_without_pyarrow
 def test_write_to_stderr(df_module, capsys):
     df = df_module.make_dataframe(
         dict(
@@ -229,6 +240,7 @@ def test_write_to_stderr(df_module, capsys):
     assert re.search(pattern, captured.err)
 
 
+@polars_installed_without_pyarrow
 def test_max_plot_columns_parameter(df_module):
     df = df_module.make_dataframe(
         {f"col_{i}": [i + j for j in range(3)] for i in range(10)}
@@ -288,6 +300,7 @@ def test_error_input_type():
         TableReport(df)
 
 
+@polars_installed_without_pyarrow
 def test_single_column_report(df_module):
     # Check that single column report works
     single_col = df_module.example_column

skrub/conftest.py

@@ -95,6 +95,21 @@ def _example_data_dict():
 except ImportError:
     _POLARS_INSTALLED = False
 
+if _POLARS_INSTALLED:
+    try:
+        import pyarrow  # noqa: F401
+
+        _polars_installed_without_pyarrow = False
+    except ImportError:
+        _polars_installed_without_pyarrow = True
+else:
+    _polars_installed_without_pyarrow = False
+
+polars_installed_without_pyarrow = pytest.mark.skipif(
+    _polars_installed_without_pyarrow=False,
+    reason="When polars is installed, requires pyarrow to be installed too",
+)
+
 
 def _pl_from_dict(data):
     import polars as pl

skrub/tests/test_check_input.py

@@ -4,6 +4,7 @@
 
 from skrub import _dataframe as sbd
 from skrub._check_input import CheckInputDataFrame
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 def test_good_input(df_module):
@@ -38,6 +39,7 @@ def test_input_is_an_array():
         check.fit_transform(np.ones((2,)))
 
 
+@polars_installed_without_pyarrow
 def test_wrong_dataframe_library_in_transform():
     pl = pytest.importorskip("polars")
     df = pl.DataFrame({"a": [0, 1], "b": [10, 20]})

skrub/tests/test_column_associations.py

@@ -5,8 +5,10 @@
 
 from skrub import _dataframe as sbd
 from skrub import column_associations
+from skrub.conftest import polars_installed_without_pyarrow
 
 
+@polars_installed_without_pyarrow
 def test_column_associations(df_module):
     x = (np.ones((7, 3)) * np.arange(3)).ravel()
     y = 2 - 3 * x
@@ -26,6 +28,7 @@ def test_column_associations(df_module):
     )
 
 
+@polars_installed_without_pyarrow
 def test_infinite(df_module):
     # non-regression test for https://github.com/skrub-data/skrub/issues/1133
     # (column associations would raise an exception on low-cardinality float

skrub/tests/test_fuzzy_join.py

@@ -8,6 +8,7 @@
 from skrub import ToDatetime, _join_utils, fuzzy_join
 from skrub import selectors as s
 from skrub._dataframe import _common as ns
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 @pytest.mark.parametrize(
@@ -237,6 +238,7 @@ def test_numerical_column(df_module):
     assert ns.shape(fj_num3) == (2, n_cols)
 
 
+@polars_installed_without_pyarrow
 def test_datetime_column(df_module):
     """
     Testing that ``fuzzy_join`` works with datetime columns.
@@ -292,6 +294,7 @@ def to_dt(lst):
     assert ns.shape(fj_time3) == (2, n_cols)
 
 
+@polars_installed_without_pyarrow
 def test_mixed_joins(df_module):
     """
     Test fuzzy joining on mixed and multiple column types.

skrub/tests/test_join_utils.py

@@ -6,6 +6,7 @@
 
 from skrub import _dataframe as sbd
 from skrub import _join_utils
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 @pytest.mark.parametrize(
@@ -173,6 +174,7 @@ def test_left_join_wrong_right_type(df_module, left):
         )
 
 
+@polars_installed_without_pyarrow
 def test_left_join_types_not_equal(df_module, left):
     try:
         import polars as pl

skrub/tests/test_to_datetime.py

@@ -16,6 +16,7 @@
     _get_time_zone,
     to_datetime,
 )
+from skrub.conftest import polars_installed_without_pyarrow
 
 ISO = "%Y-%m-%dT%H:%M:%S"
 
@@ -44,6 +45,7 @@ def datetime_col(df_module):
     return sbd.col(df_module.example_dataframe, "datetime-col")
 
 
+@polars_installed_without_pyarrow
 @pytest.mark.parametrize(
     "format",
     [
@@ -85,6 +87,7 @@ def test_datetime_to_datetime(datetime_col):
     assert encoder.format_ is None
 
 
+@polars_installed_without_pyarrow
 def test_rejected_columns(df_module):
     with pytest.raises(ValueError, match=".*does not contain strings"):
         ToDatetime().fit_transform(sbd.col(df_module.example_dataframe, "float-col"))
@@ -96,6 +99,7 @@ def test_rejected_columns(df_module):
         )
 
 
+@polars_installed_without_pyarrow
 def test_transform_failures(datetime_col, df_module):
     encoder = ToDatetime().fit(strftime(datetime_col, ISO))
     test_col = df_module.make_column(
@@ -106,6 +110,7 @@ def test_transform_failures(datetime_col, df_module):
     assert sbd.to_list(sbd.fill_nulls(strftime(transformed, ISO), "????")) == expected
 
 
+@polars_installed_without_pyarrow
 def test_mixed_offsets(df_module):
     s = df_module.make_column(
         "when", ["2020-01-02T08:00:01+02:00", "2020-01-02T08:00:01+04:00"]

skrub/tests/test_to_float32.py

@@ -6,6 +6,7 @@
 from skrub._to_categorical import ToCategorical
 from skrub._to_datetime import ToDatetime
 from skrub._to_float32 import ToFloat32
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 def is_float32(df_module, column):
@@ -30,6 +31,7 @@ def test_to_float_32(values, df_module):
     assert is_float32(df_module, out)
 
 
+@polars_installed_without_pyarrow
 def test_rejected_columns(df_module):
     columns = [
         df_module.make_column("c", ["1", "2", "hello"]),

skrub/tests/test_to_str.py

@@ -6,6 +6,7 @@
 from skrub._apply_to_cols import RejectColumn
 from skrub._to_datetime import ToDatetime
 from skrub._to_str import ToStr
+from skrub.conftest import polars_installed_without_pyarrow
 
 
 def test_to_str(df_module):
@@ -26,6 +27,7 @@ def test_to_str(df_module):
     df_module.assert_column_equal(sbd.is_null(out), sbd.is_null(expected))
 
 
+@polars_installed_without_pyarrow
 def test_rejected_columns(df_module):
     columns = [
         ToDatetime().fit_transform(df_module.make_column("", ["2020-02-02"])),