Fix all warnings & Add option TREAT_WARING_AS_ERROR to cmake

.github/workflows/test.yml

@@ -134,25 +134,25 @@ jobs:
         include:
           - test_suite: cuda
             cuda_version: 10.2
-            extra_flags: --extra_oneflow_cmake_args=-DCUDA_NVCC_GENCODES=arch=compute_61,code=sm_61 --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple
+            extra_flags: --extra_oneflow_cmake_args=-DCUDA_NVCC_GENCODES=arch=compute_61,code=sm_61 --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple --extra_oneflow_cmake_args=-DTREAT_WARNINGS_AS_ERRORS=OFF
             os: [self-hosted, linux, build]
             allow_fail: false
             python_version: 3.6,3.7
           - test_suite: cpu
             cuda_version: 10.2
-            extra_flags: --extra_oneflow_cmake_args=-DBUILD_SHARED_LIBS=OFF --extra_oneflow_cmake_args=-DRPC_BACKEND=LOCAL --cpu --gcc7
+            extra_flags: --extra_oneflow_cmake_args=-DBUILD_SHARED_LIBS=OFF --extra_oneflow_cmake_args=-DRPC_BACKEND=LOCAL --cpu --gcc7 --extra_oneflow_cmake_args=-DTREAT_WARNINGS_AS_ERRORS=OFF
             os: [self-hosted, linux, build]
             allow_fail: false
             python_version: 3.6,3.7
           - test_suite: xla
             cuda_version: 10.1
-            extra_flags: --extra_oneflow_cmake_args=-DCUDA_NVCC_GENCODES=arch=compute_61,code=sm_61 --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --xla --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple
+            extra_flags: --extra_oneflow_cmake_args=-DCUDA_NVCC_GENCODES=arch=compute_61,code=sm_61 --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --xla --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple --extra_oneflow_cmake_args=-DTREAT_WARNINGS_AS_ERRORS=OFF
             os: [self-hosted, linux, build]
             allow_fail: true
             python_version: 3.6
           - test_suite: xla_cpu
             cuda_version: 10.1
-            extra_flags: --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --xla --cpu --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple
+            extra_flags: --extra_oneflow_cmake_args=-DRPC_BACKEND=GRPC,LOCAL --xla --cpu --extra_oneflow_cmake_args=-DPIP_INDEX_MIRROR=https://pypi.tuna.tsinghua.edu.cn/simple --extra_oneflow_cmake_args=-DTREAT_WARNINGS_AS_ERRORS=OFF
             os: [self-hosted, linux, build]
             allow_fail: true
             python_version: 3.6

CMakeLists.txt

@@ -27,6 +27,7 @@ option(WITH_COCOAPI "Option to build with COCO API" ON)
 option(BUILD_GIT_VERSION "" ON)
 option(BUILD_PROFILER "" OFF)
 option(OF_SOFTMAX_USE_FAST_MATH "" ON)
+option(TREAT_WARNINGS_AS_ERRORS "" ON)
 set(RPC_BACKEND "GRPC,LOCAL" CACHE STRING "")
 set(THIRD_PARTY_MIRROR "" CACHE STRING "")
 set(PIP_INDEX_MIRROR "" CACHE STRING "")
@@ -126,10 +127,7 @@ if(WIN32)
   #set(CMAKE_EXE_LINKER_FLAGS_DEBUG "${CMAKE_EXE_LINKER_FLAGS} /DEBUG:FASTLINK")
   set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} /D_ITERATOR_DEBUG_LEVEL=0")
 else()
-  set(EXTRA_CXX_FLAGS "-std=c++11 -Wall -Wno-sign-compare -Wno-unused-function -fPIC -Werror=return-type")
-  if (APPLE)
-     set(EXTRA_CXX_FLAGS "${EXTRA_CXX_FLAGS} -Wno-deprecated-declarations -Wno-mismatched-tags")
-  endif()
+  set(EXTRA_CXX_FLAGS "-std=c++11 -Wall -Wno-sign-compare -Wno-unused-function -fPIC")
   set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${EXTRA_CXX_FLAGS}")
   set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} ${EXTRA_CXX_FLAGS}")
   set(CMAKE_CXX_FLAGS_RELEASE "${CMAKE_CXX_FLAGS_RELEASE} ${EXTRA_CXX_FLAGS}")

cmake/oneflow.cmake

@@ -237,6 +237,8 @@ if(BUILD_CUDA)
   add_dependencies(of_cudaobj of_protoobj of_cfgobj prepare_oneflow_third_party)
   target_link_libraries(of_cudaobj ${oneflow_third_party_libs})
   set(ONEFLOW_CUDA_LIBS of_cudaobj)
+
+  target_compile_options(of_cudaobj PRIVATE -Werror=return-type)
 endif()
 
 # cc obj lib
@@ -254,6 +256,33 @@ if (BUILD_SHARED_LIBS)
   target_link_libraries(of_ccobj of_protoobj of_cfgobj ${ONEFLOW_CUDA_LIBS} glog_imported)
 endif()
 
+target_compile_options(of_ccobj PRIVATE -Werror=return-type)
+
+if (TREAT_WARNINGS_AS_ERRORS)
+  target_compile_options(of_ccobj PRIVATE -Werror)
+
+  # TODO: remove it while fixing all deprecated call
+  target_compile_options(of_ccobj PRIVATE -Wno-error=deprecated-declarations)
+
+  # disable unused-* for different compile mode (maybe unused in cpu.cmake, but used in cuda.cmake)
+  target_compile_options(of_ccobj PRIVATE -Wno-error=unused-const-variable)
+  target_compile_options(of_ccobj PRIVATE -Wno-error=unused-variable)
+  if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+    target_compile_options(of_ccobj PRIVATE -Wno-error=unused-private-field)
+    target_compile_options(of_ccobj PRIVATE -Wno-error=unused-local-typedef)
+    target_compile_options(of_ccobj PRIVATE -Wno-error=unused-lambda-capture)
+    target_compile_options(of_ccobj PRIVATE -Wno-error=instantiation-after-specialization)
+  endif()
+
+  if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
+    # the mangled name between `struct X` and `class X` is different in MSVC ABI, remove it while windows is supported (in MSVC/cl or clang-cl)
+    target_compile_options(of_ccobj PRIVATE -Wno-error=mismatched-tags)
+
+    # TODO: remove it while `oneflow/user/kernels/upsample_kernel.h:141:9: error: implicit conversion from 'double' to 'int' changes value from -0.75 to 0 [-Wliteral-conversion]` is fixed
+    target_compile_options(of_ccobj PRIVATE -Wno-error=literal-conversion)
+  endif()
+endif()
+
 # py ext lib
 add_library(of_pyext_obj ${of_pyext_obj_cc})
 target_include_directories(of_pyext_obj PRIVATE ${Python_INCLUDE_DIRS} ${Python_NumPy_INCLUDE_DIRS})
@@ -262,6 +291,7 @@ if(BUILD_SHARED_LIBS AND APPLE)
   target_link_libraries(of_pyext_obj ${Python3_LIBRARIES})
 endif()
 add_dependencies(of_pyext_obj of_ccobj)
+target_compile_options(of_pyext_obj PRIVATE -Werror=return-type)
 
 if(APPLE)
   set(of_libs -Wl,-force_load ${ONEFLOW_CUDA_LIBS} of_ccobj of_protoobj of_cfgobj)

oneflow/core/autograd/gradient_funcs/adaptive_pool.cpp

@@ -29,6 +29,8 @@ struct AdaptivePoolInterpState : public OpExprInterpState {
 
 class AdaptivePoolNdGrad : public OpExprGradFunction<AdaptivePoolInterpState> {
  public:
+  using OpExprGradFunction<AdaptivePoolInterpState>::Init;
+
   Maybe<void> Init(const OpExpr& op, std::string mode, const int& ndims);
   Maybe<void> Capture(AdaptivePoolInterpState* ctx, const TensorTuple& inputs,
                       const TensorTuple& outputs, const AttrMap& attrs) const override;

oneflow/core/autograd/gradient_funcs/avg_pooling.cpp

@@ -43,7 +43,7 @@ struct AvgPoolingInterpState : public OpExprInterpState {
 class AvgPoolingNdGrad : public OpExprGradFunction<AvgPoolingInterpState> {
  public:
   virtual ~AvgPoolingNdGrad() = default;
-  Maybe<void> Init(const OpExpr& op);
+  Maybe<void> Init(const OpExpr& op) override;
   Maybe<void> Capture(AvgPoolingInterpState* ctx, const TensorTuple& inputs,
                       const TensorTuple& outputs, const AttrMap& attrs) const override;
   Maybe<void> Apply(const AvgPoolingInterpState* ctx, const TensorTuple& out_grads,

oneflow/core/autograd/gradient_funcs/broadcast_binary_ops.cpp

@@ -55,7 +55,7 @@ class BroadcastBinaryGrad : public OpExprGradFunction<OpExprInterpState> {
   BroadcastBinaryGrad() = default;
   virtual ~BroadcastBinaryGrad() = default;
 
-  virtual Maybe<void> Init(const OpExpr& op) { return Maybe<void>::Ok(); }
+  virtual Maybe<void> Init(const OpExpr& op) override { return Maybe<void>::Ok(); }
 
   Maybe<void> Capture(OpExprInterpState* ctx, const TensorTuple& inputs, const TensorTuple& outputs,
                       const AttrMap& attrs) const override {

oneflow/core/autograd/gradient_funcs/pool.cpp

@@ -43,6 +43,9 @@ struct PoolInterpState : public OpExprInterpState {
 class PoolNdGrad : public OpExprGradFunction<PoolInterpState> {
  public:
   virtual ~PoolNdGrad() = default;
+
+  using OpExprGradFunction<PoolInterpState>::Init;
+
   Maybe<void> Init(const OpExpr& op, const std::string& mode);
   Maybe<void> Capture(PoolInterpState* ctx, const TensorTuple& inputs, const TensorTuple& outputs,
                       const AttrMap& attrs) const override;

oneflow/core/autograd/gradient_funcs/pooling.cpp

@@ -45,6 +45,9 @@ struct PoolingInterpState : public OpExprInterpState {
 class PoolingNdGrad : public OpExprGradFunction<PoolingInterpState> {
  public:
   virtual ~PoolingNdGrad() = default;
+
+  using OpExprGradFunction<PoolingInterpState>::Init;
+
   Maybe<void> Init(const OpExpr& op, const std::string& mode);
   Maybe<void> Capture(PoolingInterpState* ctx, const TensorTuple& inputs,
                       const TensorTuple& outputs, const AttrMap& attrs) const override;

oneflow/core/common/cached_object_msg_allocator.h

@@ -22,7 +22,7 @@ limitations under the License.
 
 namespace oneflow {
 
-class ObjMsgChunk;
+struct ObjMsgChunk;
 
 struct ObjMsgMemBlock final {
  public:
@@ -41,7 +41,7 @@ struct ObjMsgMemBlock final {
   char mem_ptr_[0];
 };
 
-class ObjMsgSizedMemPool;
+struct ObjMsgSizedMemPool;
 
 // clang-format off
 OBJECT_MSG_BEGIN(ObjMsgChunk);

oneflow/core/common/maybe.h

@@ -294,7 +294,7 @@ inline bool MaybeIsOk(Maybe<void>&& maybe) {
   })(__FUNCTION__)                                                             \
       .Data_YouAreNotAllowedToCallThisFuncOutsideThisFile()
 
-#define CHECK_OK(...) CHECK(MaybeIsOk(std::move(__VA_ARGS__)))
+#define CHECK_OK(...) CHECK(MaybeIsOk(__VA_ARGS__))
 
 #define OF_RETURN_IF_ERROR(...)                                                              \
   for (MAYBE_CONST_AUTO_REF maybe_##__LINE__ = __MaybeErrorStackCheckWrapper__(__VA_ARGS__); \

oneflow/core/common/symbol.h

@@ -28,7 +28,7 @@ limitations under the License.
 namespace oneflow {
 
 template<typename T>
-class SymbolUtil;
+struct SymbolUtil;
 
 template<typename T>
 class Symbol final {

oneflow/core/eager/opkernel_instruction_type.h

@@ -17,6 +17,7 @@ limitations under the License.
 #define ONEFLOW_CORE_EAGER_CALL_OPKERNEL_INSTRUCTION_H_
 
 #include "oneflow/core/eager/opkernel_instruction.msg.h"
+#include "oneflow/core/vm/instr_type_id.h"
 #include "oneflow/core/vm/instruction.msg.h"
 #include "oneflow/core/vm/instruction_type.h"
 #include "oneflow/core/memory/memory_case.pb.h"
@@ -60,6 +61,9 @@ class UserStatelessCallOpKernelInstructionType : public vm::InstructionType {
   void Infer(vm::Instruction* instruction) const override;
   void Compute(vm::Instruction* instruction) const override;
 
+  using vm::InstructionType::Compute;
+  using vm::InstructionType::Infer;
+
  protected:
   UserStatelessCallOpKernelInstructionType() = default;
   virtual ~UserStatelessCallOpKernelInstructionType() = default;
@@ -80,6 +84,9 @@ class SystemStatelessCallOpKernelInstructionType : public vm::InstructionType {
   virtual std::shared_ptr<MemoryCase> GetOutBlobMemCase(const DeviceType device_type,
                                                         const int64_t device_id) const;
 
+  using vm::InstructionType::Compute;
+  using vm::InstructionType::Infer;
+
  protected:
   SystemStatelessCallOpKernelInstructionType() = default;
   virtual ~SystemStatelessCallOpKernelInstructionType() = default;

oneflow/core/eager/opkernel_object.h

@@ -24,7 +24,7 @@ limitations under the License.
 
 namespace oneflow {
 
-class KernelCtx;
+struct KernelCtx;
 class Blob;
 class ParallelContext;
 

oneflow/core/framework/config_def.cpp

@@ -81,9 +81,9 @@ const ConfigDefBuidler<config_def_type>& ConfigDefBuidler<config_def_type>::List
   return *this;
 }
 
-template class ConfigDefBuidler<kEnvConfigDefType>;
-template class ConfigDefBuidler<kSessionConfigDefType>;
-template class ConfigDefBuidler<kFunctionConfigDefType>;
-template class ConfigDefBuidler<kScopeConfigDefType>;
+template struct ConfigDefBuidler<kEnvConfigDefType>;
+template struct ConfigDefBuidler<kSessionConfigDefType>;
+template struct ConfigDefBuidler<kFunctionConfigDefType>;
+template struct ConfigDefBuidler<kScopeConfigDefType>;
 
 }  // namespace oneflow

oneflow/core/framework/device.h

@@ -33,7 +33,7 @@ class Device final {
   Device(const Device&) = default;
   Device(Device&&) = default;
   ~Device() = default;
-  Device& operator=(const Device&) = default;
+  Device& operator=(const Device&) = delete;
   const std::string& type() const { return type_; }
   Maybe<const std::string&> of_type() const;
   int64_t device_id() const { return device_id_; }

oneflow/core/framework/instructions_builder.cpp

@@ -1166,7 +1166,7 @@ Maybe<void> InstructionsBuilder::StatefulCall(
   };
 
   const auto GetDelegateBlobObject =
-      [this, &FetchDelegateBlobObject](
+      [&FetchDelegateBlobObject](
           const std::shared_ptr<compatible_py::BlobObject>& blob_object,
           const std::shared_ptr<compatible_py::OpArgParallelAttribute>& op_arg_parallel_attr)
       -> Maybe<compatible_py::BlobObject> {
@@ -1199,7 +1199,7 @@ Maybe<void> InstructionsBuilder::StatelessCall(
   };
 
   const auto GetDelegateBlobObject =
-      [this, &FetchDelegateBlobObject](
+      [&FetchDelegateBlobObject](
           const std::shared_ptr<compatible_py::BlobObject>& blob_object,
           const std::shared_ptr<compatible_py::OpArgParallelAttribute>& op_arg_parallel_attr)
       -> Maybe<compatible_py::BlobObject> {
@@ -1242,7 +1242,7 @@ Maybe<void> InstructionsBuilder::NoBoxingStatelessCall(
   };
 
   const auto GetDirectOr121BlobObject =
-      [this, &FetchDelegateBlobObject](
+      [&FetchDelegateBlobObject](
           const std::shared_ptr<compatible_py::BlobObject>& blob_object,
           const std::shared_ptr<compatible_py::OpArgParallelAttribute>& op_arg_parallel_attr)
       -> Maybe<compatible_py::BlobObject> {

oneflow/core/framework/op_expr.h

@@ -120,7 +120,7 @@ class ConsistentTensorInferCache;
 
 class UserOpExpr final : public BuiltinOpExprImpl<UserOpConf> {
  public:
-  UserOpExpr() = default;
+  UserOpExpr() = delete;
   virtual ~UserOpExpr() = default;
 
   static Maybe<UserOpExpr> New(const std::string& op_name, UserOpConf&& op_proto,

oneflow/core/framework/tensor_rpc_util.h

@@ -22,7 +22,7 @@ limitations under the License.
 
 namespace oneflow {
 
-class FlatTensorConsistency;
+struct FlatTensorConsistency;
 
 class CheckConsistencyAsyncTransportCtx : public AsyncTransportCtx {
  public:

oneflow/core/framework/user_op_conf.cpp

@@ -249,8 +249,7 @@ UserOpConfWrapper& BackwardOpConfContext::GetOp(const std::string& op_name) {
     CHECK(fn_it != op_builder_fns_.end()) << " op_name " << op_name << " has no builder function.";
     CHECK(fn_it->second != nullptr) << " op_name " << op_name << " builder function is null.";
     UserOpConfWrapperBuilder builder(op_name);
-    auto ret =
-        op_builder_results_.emplace(std::make_pair(op_name, std::move(fn_it->second(builder))));
+    auto ret = op_builder_results_.emplace(std::make_pair(op_name, fn_it->second(builder)));
     CHECK(ret.second == true) << " op_name " << op_name << " build result insert failed.";
 
     // add new op conf

oneflow/core/framework/vm_local_dep_object.cpp

@@ -88,7 +88,7 @@ ObjectMsgPtr<LocalDepObject> GetRecycledLocalDepObject(const ParallelDesc& paral
   ObjectMsgPtr<LocalDepObject> object = thread_local_free_list->Begin();
   thread_local_free_list->Erase(object.Mutable());
   CHECK_EQ(object->ref_cnt(), 1);  // hold by `object` only
-  return std::move(object);
+  return object;
 }
 
 void MoveLocalDepObjectToZombieList(ObjectMsgPtr<LocalDepObject>&& local_dep_object) {

oneflow/core/functional/packed_functor.h

@@ -33,6 +33,7 @@ using remove_cvref_t = oneflow::detail::remove_cvref_t<T>;
 
 struct FunctionBody {
   virtual operator void*() = 0;
+  virtual ~FunctionBody() = default;
 };
 
 template<typename T>

oneflow/core/job/job_build_and_infer_ctx_mgr.cpp

@@ -24,7 +24,7 @@ limitations under the License.
 namespace oneflow {
 
 Maybe<void> JobBuildAndInferCtxMgr::OpenJobBuildAndInferCtx(const std::string& job_name) {
-  CHECK_OR_RETURN(!has_cur_job_) << Error::UnknownJobBuildAndInferError
+  CHECK_OR_RETURN(!has_cur_job_) << Error::UnknownJobBuildAndInferError()
                                  << "cur job not leave before you enter this job_name:" << job_name;
   CHECK_OR_RETURN(!job_name.empty()) << Error::JobNameEmptyError();
   CHECK_OR_RETURN(job_name2infer_ctx_.find(job_name) == job_name2infer_ctx_.end())

oneflow/core/job/job_build_and_infer_ctx_mgr.h

@@ -63,7 +63,7 @@ class LazyJobBuildAndInferCtxMgr : public JobBuildAndInferCtxMgr {
   friend class Global<LazyJobBuildAndInferCtxMgr>;
 
   Maybe<void> VirtualCloseJob() override;
-  JobBuildAndInferCtx* NewJobBuildAndInferCtx(Job* job, int64_t job_id) const;
+  JobBuildAndInferCtx* NewJobBuildAndInferCtx(Job* job, int64_t job_id) const override;
 };
 
 class EagerJobBuildAndInferCtxMgr : public JobBuildAndInferCtxMgr {
@@ -76,7 +76,7 @@ class EagerJobBuildAndInferCtxMgr : public JobBuildAndInferCtxMgr {
   friend class Global<EagerJobBuildAndInferCtxMgr>;
 
   Maybe<void> VirtualCloseJob() override;
-  JobBuildAndInferCtx* NewJobBuildAndInferCtx(Job* job, int64_t job_id) const;
+  JobBuildAndInferCtx* NewJobBuildAndInferCtx(Job* job, int64_t job_id) const override;
 };
 
 bool EagerExecutionEnabled();

oneflow/core/job/session_global_objects_scope.cpp

@@ -145,7 +145,7 @@ Maybe<void> SessionGlobalObjectsScope::EagerInit(const ConfigProto& config_proto
   Global<ResourceDesc, ForSession>::Delete();
   DumpVersionInfo();
   Global<ResourceDesc, ForSession>::New(config_proto.resource());
-  for (const std::string lib_path : config_proto.load_lib_path()) { JUST(LoadLibrary(lib_path)); }
+  for (const std::string& lib_path : config_proto.load_lib_path()) { JUST(LoadLibrary(lib_path)); }
   return Maybe<void>::Ok();
 }
 

oneflow/core/job_rewriter/autotick.h

@@ -37,6 +37,7 @@ class MutOpConTickInputHelper {
   virtual bool VirtualIsTickInputBound() const = 0;
   virtual OperatorConf NewTickInputBoundOpConf(const std::string& lbn) const = 0;
   void InitFromOpConf(const OperatorConf& op_conf) { op_conf_ = &op_conf; }
+  virtual ~MutOpConTickInputHelper() = default;
 
  protected:
   MutOpConTickInputHelper() : op_conf_(nullptr) {}

oneflow/core/job_rewriter/job_pass.h

@@ -55,7 +55,8 @@ class JobPassCtx {
     const auto& iter = key2state_.find(key);
     CHECK_OR_RETURN(iter != key2state_.end());
     const T* ptr = dynamic_cast<T*>(iter->second.get());
-    CHECK_NOTNULL_OR_RETURN(ptr) << typeid(*iter->second).name();
+    const auto& origin_obj = *iter->second;
+    CHECK_NOTNULL_OR_RETURN(ptr) << typeid(origin_obj).name();
     return *ptr;
   }
 
@@ -64,7 +65,8 @@ class JobPassCtx {
     const auto& iter = key2state_.find(key);
     CHECK_OR_RETURN(iter != key2state_.end());
     T* ptr = dynamic_cast<T*>(iter->second.get());
-    CHECK_NOTNULL_OR_RETURN(ptr) << typeid(*iter->second).name();
+    const auto& origin_obj = *iter->second;
+    CHECK_NOTNULL_OR_RETURN(ptr) << typeid(origin_obj).name();
     return ptr;
   }
 

oneflow/core/kernel/slice_boxing_kernel.cpp

@@ -48,7 +48,7 @@ class SliceBoxingCopyKernel final : public SliceBoxingKernel<device_type, T> {
   ~SliceBoxingCopyKernel() override = default;
 
  private:
-  virtual const SliceBoxingConf& GetCustomizedBoxingConf() const;
+  virtual const SliceBoxingConf& GetCustomizedBoxingConf() const override;
   void ForwardDataContent(const KernelCtx&,
                           std::function<Blob*(const std::string&)>) const override;
 };
@@ -61,7 +61,7 @@ class SliceBoxingAddKernel final : public SliceBoxingKernel<device_type, T> {
   ~SliceBoxingAddKernel() override = default;
 
  private:
-  virtual const SliceBoxingConf& GetCustomizedBoxingConf() const;
+  virtual const SliceBoxingConf& GetCustomizedBoxingConf() const override;
   void ForwardDataContent(const KernelCtx&,
                           std::function<Blob*(const std::string&)>) const override;
 };