ci: Fix TPCH test data generation in CI (#2470)

* ci: Fix TPCH test data generation in CI

* pandas-nullable fixup

Author: MarcoGorelli

tests/expr_and_series/is_finite_test.py

@@ -21,6 +21,7 @@ def test_is_finite_expr(constructor: Constructor) -> None:
     ):
         expected = {"a": [False, False, True, False]}
     else:  # pandas_nullable_constructor, pandas_pyarrow_constructor, modin_pyarrrow_constructor
+        # Here, the 'nan' and None get mangled upon dataframe construction.
         expected = {"a": [None, False, True, None]}
 
     df = nw.from_native(constructor(data))

tests/expr_and_series/str/to_uppercase_to_lowercase_test.py

@@ -20,31 +20,27 @@
                     "ςpecial caσe",  # noqa: RUF001
                 ]
             },
-            {"a": ["SPECIAL CASE SS", "ΣPECIAL CAΣE"]},
+            {"a": ["SPECIAL CASE ẞ", "ΣPECIAL CAΣE"]},
         ),
     ],
 )
 def test_str_to_uppercase(
     constructor: Constructor,
     data: dict[str, list[str]],
     expected: dict[str, list[str]],
-    request: pytest.FixtureRequest,
 ) -> None:
-    if any("ß" in s for value in data.values() for s in value) & (
-        constructor.__name__
-        in {
-            "pandas_pyarrow_constructor",
-            "pyarrow_table_constructor",
-            "modin_pyarrow_constructor",
-            "duckdb_lazy_constructor",
-            "sqlframe_pyspark_lazy_constructor",
-        }
-        or ("dask" in str(constructor) and PYARROW_VERSION >= (12,))
+    if "dask" in str(constructor) and PYARROW_VERSION < (12,):
+        pytest.skip()
+
+    if (
+        any(
+            x in str(constructor)
+            for x in ("pandas_constructor", "pandas_nullable", "polars")
+        )
+        and "ẞ" in expected["a"][0]
     ):
-        # We are marking it xfail for these conditions above
-        # since the pyarrow backend will convert
-        # smaller cap 'ß' to upper cap 'ẞ' instead of 'SS'
-        request.applymarker(pytest.mark.xfail)
+        expected = {"a": ["SPECIAL CASE SS", "ΣPECIAL CAΣE"]}
+
     df = nw.from_native(constructor(data))
     result_frame = df.select(nw.col("a").str.to_uppercase())
 
@@ -62,33 +58,25 @@ def test_str_to_uppercase(
                     "ςpecial caσe",  # noqa: RUF001
                 ]
             },
-            {"a": ["SPECIAL CASE SS", "ΣPECIAL CAΣE"]},
+            {"a": ["SPECIAL CASE ẞ", "ΣPECIAL CAΣE"]},
         ),
     ],
 )
 def test_str_to_uppercase_series(
     constructor_eager: ConstructorEager,
     data: dict[str, list[str]],
     expected: dict[str, list[str]],
-    request: pytest.FixtureRequest,
 ) -> None:
     df = nw.from_native(constructor_eager(data), eager_only=True)
 
-    if any("ß" in s for value in data.values() for s in value) & (
-        constructor_eager.__name__
-        not in {
-            "pandas_constructor",
-            "pandas_nullable_constructor",
-            "polars_eager_constructor",
-            "cudf_constructor",
-            "duckdb_lazy_constructor",
-            "modin_constructor",
-        }
+    if (
+        any(
+            x in str(constructor_eager)
+            for x in ("pandas_constructor", "pandas_nullable", "polars")
+        )
+        and "ẞ" in expected["a"][0]
     ):
-        # We are marking it xfail for these conditions above
-        # since the pyarrow backend will convert
-        # smaller cap 'ß' to upper cap 'ẞ' instead of 'SS'
-        request.applymarker(pytest.mark.xfail)
+        expected = {"a": ["SPECIAL CASE SS", "ΣPECIAL CAΣE"]}
 
     result_series = df["a"].str.to_uppercase()
     assert_equal_data({"a": result_series}, expected)

tests/selectors_test.py

@@ -224,6 +224,7 @@ def test_set_ops(
     request: pytest.FixtureRequest,
 ) -> None:
     if ("duckdb" in str(constructor) or "sqlframe" in str(constructor)) and not expected:
+        # https://github.com/narwhals-dev/narwhals/issues/2469
         request.applymarker(pytest.mark.xfail)
     df = nw.from_native(constructor(data))
     result = df.select(selector).collect_schema().names()

tpch/generate_data.py

@@ -30,7 +30,7 @@
     tbl_arrow = tbl.to_arrow_table()
     new_schema = []
     for field in tbl_arrow.schema:
-        if isinstance(field.type, type(pa.decimal128(1))):
+        if isinstance(field.type, type(pa.decimal64(1))):
             new_schema.append(pa.field(field.name, pa.float64()))
         elif field.type == pa.date32():
             new_schema.append(pa.field(field.name, pa.timestamp("ns")))