TST: Different tests were collected between gw0 and gw1 (pandas-dev#37382)

Author: simonjayhawkins

pandas/tests/computation/test_eval.py

@@ -117,7 +117,7 @@ def _is_py3_complex_incompat(result, expected):
     return isinstance(expected, (complex, np.complexfloating)) and np.isnan(result)
 
 
-_good_arith_ops = set(ARITH_OPS_SYMS).difference(SPECIAL_CASE_ARITH_OPS_SYMS)
+_good_arith_ops = sorted(set(ARITH_OPS_SYMS).difference(SPECIAL_CASE_ARITH_OPS_SYMS))
 
 
 # TODO: using range(5) here is a kludge

pandas/tests/frame/apply/test_frame_transform.py

@@ -20,7 +20,7 @@ def test_transform_ufunc(axis, float_frame):
     tm.assert_frame_equal(result, expected)
 
 
-@pytest.mark.parametrize("op", transformation_kernels)
+@pytest.mark.parametrize("op", sorted(transformation_kernels))
 def test_transform_groupby_kernel(axis, float_frame, op):
     # GH 35964
     if op == "cumcount":
@@ -161,7 +161,7 @@ def test_transform_reducer_raises(all_reductions):
 
 # mypy doesn't allow adding lists of different types
 # https://github.com/python/mypy/issues/5492
-@pytest.mark.parametrize("op", [*transformation_kernels, lambda x: x + 1])
+@pytest.mark.parametrize("op", [*sorted(transformation_kernels), lambda x: x + 1])
 def test_transform_bad_dtype(op):
     # GH 35964
     df = DataFrame({"A": 3 * [object]})  # DataFrame that will fail on most transforms
@@ -182,7 +182,7 @@ def test_transform_bad_dtype(op):
             df.transform({"A": [op]})
 
 
-@pytest.mark.parametrize("op", transformation_kernels)
+@pytest.mark.parametrize("op", sorted(transformation_kernels))
 def test_transform_partial_failure(op):
     # GH 35964
     wont_fail = ["ffill", "bfill", "fillna", "pad", "backfill", "shift"]

pandas/tests/series/apply/test_series_transform.py

@@ -18,7 +18,7 @@ def test_transform_ufunc(string_series):
     tm.assert_series_equal(result, expected)
 
 
-@pytest.mark.parametrize("op", transformation_kernels)
+@pytest.mark.parametrize("op", sorted(transformation_kernels))
 def test_transform_groupby_kernel(string_series, op):
     # GH 35964
     if op == "cumcount":
@@ -144,7 +144,7 @@ def test_transform_reducer_raises(all_reductions):
 
 # mypy doesn't allow adding lists of different types
 # https://github.com/python/mypy/issues/5492
-@pytest.mark.parametrize("op", [*transformation_kernels, lambda x: x + 1])
+@pytest.mark.parametrize("op", [*sorted(transformation_kernels), lambda x: x + 1])
 def test_transform_bad_dtype(op):
     # GH 35964
     s = Series(3 * [object])  # Series that will fail on most transforms