Fix modernize-use-equals-default nolint failures in torch/csrcs (#61142)

Summary:
Test-plan: Compile + clang-tidy

Pull Request resolved: #61142

Reviewed By: VitalyFedyunin

Differential Revision: D29529372

Pulled By: malfet

fbshipit-source-id: 2ccde7712a51c28243b16bbb4d1d68086e0414a6

.github/workflows/lint.yml

@@ -366,7 +366,7 @@ jobs:
           # FunctionsManual.cpp is excluded to keep this diff clean. It will be fixed
           # in a follow up PR.
           # /torch/csrc/generic/*.cpp is excluded because those files aren't actually built.
-          # deploy/interpreter files are excluded due to using macros and other techniquies
+          # deploy/interpreter files are excluded due to using macros and other techniques
           # that are not easily converted to accepted c++
           python3 tools/linter/clang_tidy.py \
             --parallel \

aten/src/ATen/core/dispatch/Dispatcher.cpp

@@ -32,8 +32,7 @@ class RegistrationListenerList final {
 };
 }
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-OpRegistrationListener::~OpRegistrationListener() {}
+OpRegistrationListener::~OpRegistrationListener()= default;
 
 Dispatcher::Dispatcher()
 : operators_()
@@ -42,8 +41,7 @@ Dispatcher::Dispatcher()
 , listeners_(std::make_unique<detail::RegistrationListenerList>())
 , mutex_() {}
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-Dispatcher::~Dispatcher() {}
+Dispatcher::~Dispatcher() = default;
 
 C10_EXPORT Dispatcher& Dispatcher::realSingleton() {
   static Dispatcher _singleton;

aten/src/ATen/native/RNN.cpp

@@ -729,9 +729,8 @@ struct Cell {
   using hidden_type = hidden_type_tmpl;
   using cell_params = cell_params_tmpl;
 
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  virtual ~Cell() {} // This is really dumb, but enables projects with
-                     // -Wnon-virtual-dtor to compile...
+  virtual ~Cell() = default; // This is really dumb, but enables projects with
+                             // -Wnon-virtual-dtor to compile...
 
   virtual hidden_type operator()(
       const Tensor& input,
@@ -846,9 +845,8 @@ template<typename io_type, typename hidden_type, typename param_type>
 struct Layer {
   using output_type = LayerOutput<io_type, hidden_type>;
 
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  virtual ~Layer() {} // This is really dumb, but enables projects with
-                      // -Wnon-virtual-dtor to compile...
+  virtual ~Layer() = default; // This is really dumb, but enables projects with
+                              // -Wnon-virtual-dtor to compile...
   virtual output_type operator()(
       const io_type& input,
       const hidden_type& input_hidden,

aten/src/ATen/native/metal/MetalGuardImpl.cpp

@@ -5,8 +5,7 @@ namespace at {
 namespace detail {
 
 struct MetalGuardImpl final : public c10::impl::DeviceGuardImplInterface {
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  MetalGuardImpl() {}
+  MetalGuardImpl() = default;
 
   explicit MetalGuardImpl(DeviceType t) {
     TORCH_INTERNAL_ASSERT(t == DeviceType::Metal);

aten/src/ATen/nnapi/nnapi_bind.cpp

@@ -44,16 +44,12 @@ MAKE_SMART_PTR(Execution)
 #undef MAKE_SMART_PTR
 
 struct NnapiCompilation : torch::jit::CustomClassHolder {
-  // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init,modernize-use-equals-default)
-  NnapiCompilation() {
-    // Could possibly call load_platform_library here, but error reporting
-    // can be complicated if the constructor is called during model loading.
-    // Instead, delay all work until the explicit init call.
-  }
+  // Could possibly call load_platform_library here, but error reporting
+  // can be complicated if the constructor is called during model loading.
+  // Instead, delay all work until the explicit init call.
+  NnapiCompilation() = default;
 
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  ~NnapiCompilation() {
-  }
+  ~NnapiCompilation() override = default;
 
   void init(
       at::Tensor serialized_model_tensor,

aten/src/ATen/quantized/Quantizer.cpp

@@ -208,8 +208,7 @@ Tensor PerChannelAffineFloatQParamsQuantizer::dequantize(const Tensor& qtensor)
   return rtensor;
 }
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-Quantizer::~Quantizer() {}
+Quantizer::~Quantizer() = default;
 
 C10_EXPORT void set_quantizer_(const Tensor& self, ConstQuantizerPtr quantizer) {
   get_qtensorimpl(self)->set_quantizer_(quantizer);

aten/src/ATen/test/variant_test.cpp

@@ -11,12 +11,9 @@ namespace enumtype {
   // error: default initialization of an object of const type 'const enumtype::Enum1'
   // without a user-provided default constructor
   // ```
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  struct Enum1 { Enum1() {}; };
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  struct Enum2 { Enum2() {}; };
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  struct Enum3 { Enum3() {}; };
+  struct Enum1 { Enum1() = default; };
+  struct Enum2 { Enum2() = default; };
+  struct Enum3 { Enum3()  = default; };
 } // namespace enumtype
 
 struct enum_name {

aten/src/ATen/test/vulkan_test.cpp

@@ -436,8 +436,7 @@ class Conv2d : public BaseOp {
 
 class OpsList {
  public:
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  OpsList() {}
+  OpsList() = default;
   OpsList(std::vector<std::unique_ptr<BaseOp>>& _ops) : ops(std::move(_ops)) {}
 
   auto runDual(at::Tensor& in, at::Tensor& vin) {

c10/core/Allocator.cpp

@@ -51,7 +51,6 @@ void reportMemoryUsageToProfiler(void* ptr, int64_t alloc_size, Device device) {
   }
 }
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-MemoryReportingInfoBase::MemoryReportingInfoBase() {}
+MemoryReportingInfoBase::MemoryReportingInfoBase() = default;
 
 } // namespace c10

c10/core/CPUAllocator.cpp

@@ -103,10 +103,7 @@ void free_cpu(void* data) {
 }
 
 struct C10_API DefaultCPUAllocator final : at::Allocator {
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  DefaultCPUAllocator() {}
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  ~DefaultCPUAllocator() override {}
+  DefaultCPUAllocator() = default;
   at::DataPtr allocate(size_t nbytes) const override {
     void* data = alloc_cpu(nbytes);
     profiledCPUMemoryReporter().New(data, nbytes);

c10/core/TensorImpl.cpp

@@ -449,8 +449,7 @@ at::DataPtr PlacementDeleteContext::makeDataPtr(
       device};
 }
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-AutogradMetaInterface::~AutogradMetaInterface() {}
+AutogradMetaInterface::~AutogradMetaInterface() = default;
 
 // Setting requires_grad to true on inference tensor outside InferenceMode
 // is forbidden.  Ideally it would also be illegal inside InferenceMode.

c10/cuda/CUDACachingAllocator.h

@@ -19,7 +19,7 @@ class C10_CUDA_API CUDAOutOfMemoryError : public c10::Error {
 // block inside of already allocated area.
 class C10_CUDA_API FreeMemoryCallback {
  public:
-  virtual ~FreeMemoryCallback(){};
+  virtual ~FreeMemoryCallback() = default;
   virtual bool Execute() = 0;
 };
 

c10/test/util/either_test.cpp

@@ -1161,20 +1161,16 @@ class ClassWithDestructorCallback {
  public:
   ClassWithDestructorCallback(const DestructorCallback* destructorCallback)
       : _destructorCallback(destructorCallback) {}
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  ClassWithDestructorCallback(const ClassWithDestructorCallback& rhs)
-      : _destructorCallback(rhs._destructorCallback) {}
 
   ~ClassWithDestructorCallback() {
     _destructorCallback->call();
   }
 
- private:
-  const DestructorCallback* _destructorCallback;
-
-  // NOLINTNEXTLINE(modernize-use-equals-delete)
   ClassWithDestructorCallback& operator=(
       const ClassWithDestructorCallback& rhs) = delete;
+
+ private:
+  const DestructorCallback* _destructorCallback;
 };
 class OnlyMoveableClassWithDestructorCallback {
  public:

c10/test/util/registry_test.cpp

@@ -13,8 +13,7 @@ class Foo {
   explicit Foo(int x) {
     // LOG(INFO) << "Foo " << x;
   }
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  virtual ~Foo() {}
+  virtual ~Foo() = default;
 };
 
 C10_DECLARE_REGISTRY(FooRegistry, Foo, int);

c10/test/util/typeid_test.cpp

@@ -76,8 +76,7 @@ TEST(TypeMetaTest, TypeMeta) {
 class ClassAllowAssignment {
  public:
   ClassAllowAssignment() : x(42) {}
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  ClassAllowAssignment(const ClassAllowAssignment& src) : x(src.x) {}
+  ClassAllowAssignment(const ClassAllowAssignment& src) = default;
   ClassAllowAssignment& operator=(const ClassAllowAssignment& src) = default;
   int x;
 };

c10/util/ThreadLocalDebugInfo.h

@@ -23,7 +23,7 @@ enum class C10_API_ENUM DebugInfoKind : uint8_t {
 class C10_API DebugInfoBase {
  public:
   DebugInfoBase() {}
-  virtual ~DebugInfoBase() {}
+  virtual ~DebugInfoBase() = default;
 };
 
 // Thread local debug information is propagated across the forward

tools/linter/clang_tidy.py

@@ -373,6 +373,10 @@ def main() -> None:
                 {"name": name, "lines": lines} for name, lines, in changed_files.items()
             ]
             files = list(changed_files.keys())
+            # Since header files are excluded, add .cpp file if it exists in the same folder
+            cpp_files = [f[:-1] + "cpp" for f in files if f.endswith(".h")]
+            cpp_files = [f for f in cpp_files if os.path.exists(f)]
+            files = list(set(files + cpp_files))
     else:
         line_filters = []
         files = get_all_files(paths)

torch/csrc/CudaIPCTypes.cpp

@@ -26,14 +26,12 @@ void warnProducerTerminatedBeforeSharedTensorsReleased() {
 
 struct CudaIPCGlobalEntities {
   std::mutex ref_counters_mutex_;
-  std::atomic<int64_t> sync_events_used_;
+  std::atomic<int64_t> sync_events_used_{0};
   std::map<std::string, std::shared_ptr<CudaIPCRefCountersFile>>
       ref_counters_files_;
   std::shared_ptr<CudaIPCRefCountersFile> next_available_ref_counters_file_;
   CudaIPCSentDataLimbo CudaIPCSentDataLimbo_;
-  // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  CudaIPCGlobalEntities() : ref_counters_files_() {}
+  CudaIPCGlobalEntities()  = default;
   ~CudaIPCGlobalEntities() {
     CudaIPCSentDataLimbo_.collect();
     // Clear shared blocks to avoid releasing shared blocks after
@@ -109,7 +107,6 @@ void CudaIPCSentDataLimbo::add(std::unique_ptr<CudaIPCSentData> shared_block) {
 }
 
 void CudaIPCSentDataDelete(void* ptr) {
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   std::unique_ptr<CudaIPCSentData> sent_data(
       static_cast<CudaIPCSentData*>(ptr));
   if (sent_data->counter_value() > 0) {
@@ -133,13 +130,12 @@ void ReturnRefCounter(const std::string& handle, uint64_t offset /* unused */) {
 
 } // namespace
 
-// NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
 CudaIPCSentData::CudaIPCSentData(
-    const std::string& handle,
+    std::string handle,
     int64_t offset,
     int64_t* counter_ptr,
     at::Device device)
-    : handle_(handle),
+    : handle_(std::move(handle)),
       offset_(offset),
       counter_ptr_(counter_ptr),
       original_ptr_(),
@@ -172,6 +168,7 @@ CudaIPCSentData::CudaIPCSentData(
   } else {
     auto stream = c10::cuda::getCurrentCUDAStream(device.index());
     C10_CUDA_CHECK(cudaStreamSynchronize(stream));
+    event_ = nullptr;
     event_sync_required_ = false;
   }
 #else

torch/csrc/CudaIPCTypes.h

@@ -29,7 +29,7 @@ struct CudaIPCSentData final {
   at::Device device_;
 
   CudaIPCSentData(
-      const std::string& handle,
+      std::string handle,
       int64_t offset,
       int64_t* counter_ptr,
       at::Device device);
@@ -78,13 +78,13 @@ struct CudaIPCSentDataLimbo final {
 
 struct CudaIPCRefCountersFile final {
   CudaIPCRefCountersFile(
-      const std::string& handle,
+      std::string handle,
       uint64_t size,
       at::DataPtr data_ptr)
       : next_offset_(0),
         size_(size),
         used_slots_(0),
-        handle_(handle),
+        handle_(std::move(handle)),
         refcounted_shared_mem_(std::move(data_ptr)) {}
 
   int64_t* counter_ptr() {
@@ -135,8 +135,6 @@ namespace c10 {
 namespace {
 class CudaIPCCollectCallback : public FreeMemoryCallback {
  public:
-  // NOLINTNEXTLINE(modernize-use-override,modernize-use-equals-default)
-  ~CudaIPCCollectCallback() {};
   bool Execute() override {
     return torch::CudaIPCCollect();
   }

torch/csrc/autograd/forward_grad.cpp

@@ -6,19 +6,11 @@ namespace {
     // See discussion in forward_grad.h for why these are global variables and not
     // thread local
 
-    // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
-    static std::mutex all_forward_levels_mutex_;
-    // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
-    static uint64_t next_forward_idx_ = 0;
-    // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
-    static std::vector<std::shared_ptr<ForwardADLevel>> all_forward_levels_;
+    std::mutex all_forward_levels_mutex_;
+    uint64_t next_forward_idx_ = 0;
+    std::vector<std::shared_ptr<ForwardADLevel>> all_forward_levels_;
 
     const static at::Tensor singleton_undefined_tensor;
-
-    // Temporary flag to disable forward mode
-    // TODO(alband) remove these when perf issues are solved
-    // NOLINTNEXTLINE(clang-diagnostic-unused-variable,cppcoreguidelines-avoid-non-const-global-variables)
-    static bool is_forward_grad_enabled = false;
 }
 
 uint64_t ForwardADLevel::get_next_idx() {

torch/csrc/autograd/forward_grad.h

@@ -84,7 +84,6 @@ struct ForwardGrad;
 #define EXPECTED_MAX_LEVEL 2
 
 struct TORCH_API ForwardADLevel {
-  // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
   ForwardADLevel(uint64_t idx) : idx_(idx) {}
   ~ForwardADLevel();
 
@@ -111,8 +110,7 @@ struct TORCH_API ForwardADLevel {
 };
 
 struct TORCH_API ForwardGrad : std::enable_shared_from_this<ForwardGrad> {
-  // NOLINTNEXTLINE(modernize-use-equals-default,cppcoreguidelines-pro-type-member-init)
-  ForwardGrad() {}
+  ForwardGrad() = default;
 
   // This function must only be called when AutogradMeta or SavedVariable is
   // being destructed as it ensures that:

torch/csrc/autograd/functions/basic_ops.h

@@ -63,8 +63,7 @@ struct TORCH_API UndefinedGradBackward : public Node {
   UndefinedGradBackward(edge_list&& next_edges)
     : Node(std::move(next_edges)) {}
 
-  // NOLINTNEXTLINE(modernize-use-equals-default)
-  UndefinedGradBackward() {}
+  UndefinedGradBackward() = default;
 
   variable_list apply(variable_list&& inputs) override;
 };

torch/csrc/autograd/functions/comm.cpp

@@ -31,8 +31,7 @@ Scatter::Scatter(
       streams_(streams),
       unsqueeze_scalars_(unsqueeze_scalars) {}
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-Scatter::~Scatter() {}
+Scatter::~Scatter() = default;
 
 variable_list Scatter::apply(variable_list&& inputs) {
   AT_ASSERT(inputs.size() == 1);
@@ -52,7 +51,6 @@ variable_list Scatter::apply(variable_list&& inputs) {
       // NOLINTNEXTLINE(performance-move-const-arg)
       std::move(input), device_indices, chunk_sizes_, dim_, streams_);
 
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   std::vector<Variable> variables;
   variables.reserve(tensors.size());
   for (auto& tensor : tensors) {
@@ -75,8 +73,7 @@ variable_list Scatter::apply(variable_list&& inputs) {
 Gather::Gather(const at::Device& destination_device, int64_t dim)
     : destination_device_(destination_device), dim_(dim) {}
 
-// NOLINTNEXTLINE(modernize-use-equals-default)
-Gather::~Gather() {}
+Gather::~Gather() = default;
 
 variable_list Gather::apply(variable_list&& inputs) {
   bool all_are_zero_dim = true;