ARROW-4563: [Python] Validate decimal128() precision input

Also add a debug check on the C++ side.

Author: pitrou

cpp/src/arrow/type.cc

@@ -292,6 +292,15 @@ int StructType::GetChildIndex(const std::string& name) const {
   return GetFieldIndex(name);
 }
 
+// ----------------------------------------------------------------------
+// Decimal128 type
+
+Decimal128Type::Decimal128Type(int32_t precision, int32_t scale)
+    : DecimalType(16, precision, scale) {
+  DCHECK_GE(precision, 1);
+  DCHECK_LE(precision, 38);
+}
+
 // ----------------------------------------------------------------------
 // DictionaryType
 

cpp/src/arrow/type.h

@@ -550,8 +550,7 @@ class ARROW_EXPORT Decimal128Type : public DecimalType {
  public:
   static constexpr Type::type type_id = Type::DECIMAL;
 
-  explicit Decimal128Type(int32_t precision, int32_t scale)
-      : DecimalType(16, precision, scale) {}
+  explicit Decimal128Type(int32_t precision, int32_t scale);
 
   Status Accept(TypeVisitor* visitor) const override;
   std::string ToString() const override;

cpp/src/gandiva/expression_registry.cc

@@ -137,7 +137,7 @@ void ExpressionRegistry::AddArrowTypesToVector(arrow::Type::type& type,
       vector.push_back(arrow::null());
       break;
     case arrow::Type::type::DECIMAL:
-      vector.push_back(arrow::decimal(0, 0));
+      vector.push_back(arrow::decimal(38, 0));
       break;
     case arrow::Type::type::FIXED_SIZE_BINARY:
     case arrow::Type::type::MAP:

cpp/src/gandiva/function_registry_common.h

@@ -53,7 +53,7 @@ inline DataTypePtr time32() { return arrow::time32(arrow::TimeUnit::MILLI); }
 inline DataTypePtr time64() { return arrow::time64(arrow::TimeUnit::MICRO); }
 
 inline DataTypePtr timestamp() { return arrow::timestamp(arrow::TimeUnit::MILLI); }
-inline DataTypePtr decimal128() { return arrow::decimal(0, 0); }
+inline DataTypePtr decimal128() { return arrow::decimal(38, 0); }
 
 struct KeyHash {
   std::size_t operator()(const FunctionSignature* k) const { return k->Hash(); }

python/pyarrow/tests/test_types.py

@@ -402,6 +402,14 @@ def test_decimal_properties():
     assert ty.scale == 4
 
 
+def test_decimal_overflow():
+    pa.decimal128(1, 0)
+    pa.decimal128(38, 0)
+    for i in (0, -1, 39):
+        with pytest.raises(ValueError):
+            pa.decimal128(39, 0)
+
+
 def test_type_equality_operators():
     many_types = get_many_types()
     non_pyarrow = ('foo', 16, {'s', 'e', 't'})

python/pyarrow/types.pxi

@@ -1321,6 +1321,8 @@ cpdef DataType decimal128(int precision, int scale=0):
     decimal_type : Decimal128Type
     """
     cdef shared_ptr[CDataType] decimal_type
+    if precision < 1 or precision > 38:
+        raise ValueError("precision should be between 1 and 38")
     decimal_type.reset(new CDecimal128Type(precision, scale))
     return pyarrow_wrap_data_type(decimal_type)