fix uint64 overflow

Author: Rohan Jain

pandas/core/arrays/arrow/array.py

@@ -129,15 +129,22 @@ def floordiv_compat(
     ) -> pa.ChunkedArray:
         divided = pc.divide(left, right)
         if pa.types.is_integer(divided.type):
-            # GH 56676, avoid storing intermediate calculating in floating point type.
+            # GH 56676: avoid storing intermediate calculating in floating point type.
             has_remainder = pc.not_equal(pc.multiply(divided, right), left)
             result = pc.if_else(
-                pc.and_(pc.less(divided, 0), has_remainder),
+                # Pass a typed arrow scalar rather than stdlib int
+                # which always inferred as int64, to prevent overflow
+                # in case of large uint64 values.
+                pc.and_(
+                    pc.less(divided, pa.scalar(0, type=divided.type)), has_remainder
+                ),
+                # GH 55561: floordiv should round towards negative infinity.
+                # pv.divide for integral types rounds towards 0.
                 # Avoid using subtract_checked which would incorrectly raise
                 # for -9223372036854775808 // 1, because if integer overflow
                 # occurs, then has_remainder should be false, and overflowed
                 # value is discarded.
-                pc.subtract(divided, 1),
+                pc.subtract(divided, pa.scalar(1, type=divided.type)),
                 divided,
             )
             # Ensure compatibility with older versions of pandas where

pandas/tests/extension/test_arrow.py

@@ -3253,6 +3253,14 @@ def test_arrow_floordiv_large_integral_result():
     tm.assert_series_equal(result, a)
 
 
+def test_arrow_floordiv_no_overflow():
+    # GH 56676
+    a = pd.Series([9223372036854775808], dtype="uint64[pyarrow]")
+    b = pd.Series([1], dtype="uint64[pyarrow]")
+    result = a // b
+    tm.assert_series_equal(result, a)
+
+
 def test_string_to_datetime_parsing_cast():
     # GH 56266
     string_dates = ["2020-01-01 04:30:00", "2020-01-02 00:00:00", "2020-01-03 00:00:00"]