Merge pull request #61364 from ClickHouse/vdimir/fix_crash_objson_array_null

Fix crash in ObjectJson parsing array with nulls

src/Columns/ColumnObject.cpp

@@ -20,12 +20,12 @@ namespace DB
 
 namespace ErrorCodes
 {
-    extern const int ILLEGAL_COLUMN;
+    extern const int ARGUMENT_OUT_OF_BOUND;
     extern const int DUPLICATE_COLUMN;
+    extern const int EXPERIMENTAL_FEATURE_ERROR;
+    extern const int ILLEGAL_COLUMN;
     extern const int NUMBER_OF_DIMENSIONS_MISMATCHED;
     extern const int SIZES_OF_COLUMNS_DOESNT_MATCH;
-    extern const int ARGUMENT_OUT_OF_BOUND;
-    extern const int EXPERIMENTAL_FEATURE_ERROR;
 }
 
 namespace
@@ -334,7 +334,18 @@ void ColumnObject::Subcolumn::insert(Field field, FieldInfo info)
     if (type_changed || info.need_convert)
         field = convertFieldToTypeOrThrow(field, *least_common_type.get());
 
-    data.back()->insert(field);
+    if (!data.back()->tryInsert(field))
+    {
+        /** Normalization of the field above is pretty complicated (it uses several FieldVisitors),
+          * so in the case of a bug, we may get mismatched types.
+          * The `IColumn::insert` method does not check the type of the inserted field, and it can lead to a segmentation fault.
+          * Therefore, we use the safer `tryInsert` method to get an exception instead of a segmentation fault.
+          */
+        throw Exception(ErrorCodes::EXPERIMENTAL_FEATURE_ERROR,
+            "Cannot insert field {} to column {}",
+            field.dump(), data.back()->dumpStructure());
+    }
+
     ++num_rows;
 }
 

src/Columns/ColumnVector.cpp

@@ -460,6 +460,28 @@ Float32 ColumnVector<T>::getFloat32(size_t n [[maybe_unused]]) const
         throw Exception(ErrorCodes::NOT_IMPLEMENTED, "Cannot get the value of {} as Float32", TypeName<T>);
 }
 
+template <typename T>
+bool ColumnVector<T>::tryInsert(const DB::Field & x)
+{
+    NearestFieldType<T> value;
+    if (!x.tryGet<NearestFieldType<T>>(value))
+    {
+        if constexpr (std::is_same_v<T, UInt8>)
+        {
+            /// It's also possible to insert boolean values into UInt8 column.
+            bool boolean_value;
+            if (x.tryGet<bool>(boolean_value))
+            {
+                data.push_back(static_cast<T>(boolean_value));
+                return true;
+            }
+        }
+        return false;
+    }
+    data.push_back(static_cast<T>(value));
+    return true;
+}
+
 template <typename T>
 void ColumnVector<T>::insertRangeFrom(const IColumn & src, size_t start, size_t length)
 {

src/Columns/ColumnVector.h

@@ -224,14 +224,8 @@ public:
         data.push_back(static_cast<T>(x.get<T>()));
     }
 
-    bool tryInsert(const DB::Field & x) override
+    bool tryInsert(const DB::Field & x) override;
-    {
+
-        NearestFieldType<T> value;
-        if (!x.tryGet<NearestFieldType<T>>(value))
-            return false;
-        data.push_back(static_cast<T>(value));
-        return true;
-    }
     void insertRangeFrom(const IColumn & src, size_t start, size_t length) override;
 
     ColumnPtr filter(const IColumn::Filter & filt, ssize_t result_size_hint) const override;

src/DataTypes/ObjectUtils.cpp

@@ -1054,6 +1054,14 @@ Field FieldVisitorFoldDimension::operator()(const Array & x) const
     return res;
 }
 
+Field FieldVisitorFoldDimension::operator()(const Null & x) const
+{
+    if (num_dimensions_to_fold == 0)
+        return x;
+
+    return Array();
+}
+
 void setAllObjectsToDummyTupleType(NamesAndTypesList & columns)
 {
     for (auto & column : columns)

src/DataTypes/ObjectUtils.h

@@ -149,7 +149,7 @@ public:
 
     Field operator()(const Array & x) const;
 
-    Field operator()(const Null & x) const { return x; }
+    Field operator()(const Null & x) const;
 
     template <typename T>
     Field operator()(const T & x) const

tests/queries/0_stateless/02287_type_object_convert.reference

@@ -16,3 +16,4 @@
 {"x":[[],[1,2]]}
 {"x":[[],[[1],[2]]]}
 {"x":[[],[[],[2]]]}
+{"a.a":[[1],[]]}

tests/queries/0_stateless/02287_type_object_convert.sql

@@ -32,3 +32,4 @@ SELECT CAST('{"x" : [ 1 , [ 1 , 2] ]}', 'Object(\'json\')');
 SELECT CAST('{"x" : [ {} , [ 1 , 2] ]}', 'Object(\'json\')');
 SELECT CAST('{"x" : [ {} , [ 1 , [2]] ]}', 'Object(\'json\')');
 SELECT CAST('{"x" : [ {} , [ {} , [2]] ]}', 'Object(\'json\')');
+SELECT CAST(' {"a": { "a": [ [1],  null ] } }', 'Object(Nullable(\'json\'))');