Test summing a ChunkedArray with 0 chunks

Author: pitrou

cpp/src/arrow/compute/kernels/aggregate_test.cc

@@ -152,6 +152,10 @@ TYPED_TEST(TestNumericSumKernel, SimpleSum) {
   ValidateSum<TypeParam>(&this->ctx_, chunks,
                          Datum(std::make_shared<ScalarType>(static_cast<T>(5 * 6 / 2))));
 
+  chunks = {};
+  ValidateSum<TypeParam>(&this->ctx_, chunks,
+                         Datum(std::make_shared<ScalarType>()));  // null
+
   const T expected_result = static_cast<T>(14);
   ValidateSum<TypeParam>(&this->ctx_, "[1, null, 3, null, 3, null, 7]",
                          Datum(std::make_shared<ScalarType>(expected_result)));

cpp/src/arrow/testing/gtest_util.cc

@@ -231,7 +231,7 @@ std::shared_ptr<ChunkedArray> ChunkedArrayFromJSON(const std::shared_ptr<DataTyp
   for (const std::string& chunk_json : json) {
     out_chunks.push_back(ArrayFromJSON(type, chunk_json));
   }
-  return std::make_shared<ChunkedArray>(std::move(out_chunks));
+  return std::make_shared<ChunkedArray>(std::move(out_chunks), type);
 }
 
 std::shared_ptr<RecordBatch> RecordBatchFromJSON(const std::shared_ptr<Schema>& schema,

python/pyarrow/includes/libarrow.pxd

@@ -762,6 +762,7 @@ cdef extern from "arrow/api.h" namespace "arrow" nogil:
 
     cdef cppclass CScalar" arrow::Scalar":
         shared_ptr[CDataType] type
+        c_bool is_valid
 
     cdef cppclass CInt8Scalar" arrow::Int8Scalar"(CScalar):
         int8_t value

python/pyarrow/scalar.pxi

@@ -1031,6 +1031,7 @@ cdef class ScalarValue(Scalar):
     def __hash__(self):
         return hash(self.as_py())
 
+
 cdef class UInt8Scalar(ScalarValue):
     """
     Concrete class for uint8 scalars.
@@ -1041,7 +1042,7 @@ cdef class UInt8Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CUInt8Scalar* sp = <CUInt8Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class Int8Scalar(ScalarValue):
@@ -1054,7 +1055,7 @@ cdef class Int8Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CInt8Scalar* sp = <CInt8Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class UInt16Scalar(ScalarValue):
@@ -1067,7 +1068,7 @@ cdef class UInt16Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CUInt16Scalar* sp = <CUInt16Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class Int16Scalar(ScalarValue):
@@ -1080,7 +1081,7 @@ cdef class Int16Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CInt16Scalar* sp = <CInt16Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class UInt32Scalar(ScalarValue):
@@ -1093,7 +1094,7 @@ cdef class UInt32Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CUInt32Scalar* sp = <CUInt32Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class Int32Scalar(ScalarValue):
@@ -1106,7 +1107,7 @@ cdef class Int32Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CInt32Scalar* sp = <CInt32Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class UInt64Scalar(ScalarValue):
@@ -1119,7 +1120,7 @@ cdef class UInt64Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CUInt64Scalar* sp = <CUInt64Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class Int64Scalar(ScalarValue):
@@ -1132,7 +1133,7 @@ cdef class Int64Scalar(ScalarValue):
         Return this value as a Python int.
         """
         cdef CInt64Scalar* sp = <CInt64Scalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class FloatScalar(ScalarValue):
@@ -1145,7 +1146,7 @@ cdef class FloatScalar(ScalarValue):
         Return this value as a Python float.
         """
         cdef CFloatScalar* sp = <CFloatScalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef class DoubleScalar(ScalarValue):
@@ -1158,7 +1159,7 @@ cdef class DoubleScalar(ScalarValue):
         Return this value as a Python float.
         """
         cdef CDoubleScalar* sp = <CDoubleScalar*> self.sp_scalar.get()
-        return sp.value
+        return sp.value if sp.is_valid else None
 
 
 cdef dict _scalar_classes = {

python/pyarrow/tests/test_compute.py

@@ -63,6 +63,10 @@ def test_sum_array(arrow_type):
     assert arr.sum() == 10
     assert pa.compute.sum(arr) == 10
 
+    arr = pa.array([], type=arrow_type)
+    assert arr.sum() == None  # noqa: E711
+    assert pa.compute.sum(arr) == None  # noqa: E711
+
 
 @pytest.mark.parametrize('arrow_type', numerical_arrow_types)
 def test_sum_chunked_array(arrow_type):
@@ -81,6 +85,11 @@ def test_sum_chunked_array(arrow_type):
     ])
     assert pa.compute.sum(arr) == 10
 
+    arr = pa.chunked_array((), type=arrow_type)
+    print(arr, type(arr))
+    assert arr.num_chunks == 0
+    assert pa.compute.sum(arr) == None  # noqa: E711
+
 
 @pytest.mark.parametrize(('ty', 'values'), all_array_types)
 def test_take(ty, values):