fix(python): fix potential OverflowError in testing asserts with hu…

…ge `UInt64` diffs (#10437)

py-polars/polars/testing/asserts.py

@@ -12,6 +12,7 @@
     DataTypeClass,
     List,
     Struct,
+    UInt64,
     Utf8,
     dtype_to_py_type,
     unpack_dtypes,
@@ -407,7 +408,9 @@ def _assert_series_inner(
 
             if all(tp in UNSIGNED_INTEGER_DTYPES for tp in (left.dtype, right.dtype)):
                 # avoid potential "subtract-with-overflow" panic on uint math
-                s_diff = Series("diff", [abs(v1 - v2) for v1, v2 in zip(left, right)])
+                s_diff = Series(
+                    "diff", [abs(v1 - v2) for v1, v2 in zip(left, right)], dtype=UInt64
+                )
             else:
                 s_diff = (left - right).abs()
 

py-polars/tests/unit/test_testing.py

@@ -596,6 +596,19 @@ def test_assert_series_equal_uint_overflow() -> None:
         assert_series_equal(s1, s2, atol=0)
     assert_series_equal(s1, s2, atol=1)
 
+    # confirm no OverflowError in the below test case:
+    # as "(left-right).abs()" > max(Int64)
+    left = pl.Series(
+        values=[2810428175213635359],
+        dtype=pl.UInt64,
+    )
+    right = pl.Series(
+        values=[15807433754238349345],
+        dtype=pl.UInt64,
+    )
+    with pytest.raises(AssertionError):
+        assert_series_equal(left, right)
+
 
 @pytest.mark.parametrize(
     ("data1", "data2"),