[2/N] Fix clang-tidy warnings in torch/csrc/autograd (#133295)

Follows #133180
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133295
Approved by: https://github.com/Skylion007

torch/csrc/autograd/VariableTypeManual.cpp

@@ -20,8 +20,8 @@ using torch::autograd::as_view;
 using torch::autograd::CreationMeta;
 
 namespace torch {
-namespace autograd {
-namespace VariableType {
+
+namespace autograd::VariableType {
 
 static std::vector<at::DeprecatedTypeProperties*> allTypesForBackends(
     at::ArrayRef<at::Backend> backends) {
@@ -372,8 +372,7 @@ TORCH_LIBRARY_IMPL(aten, Autograd, m) {
 }
 
 } // namespace
-} // namespace VariableType
-} // namespace autograd
+} // namespace autograd::VariableType
 
 namespace ADInplaceOrView {
 #define CREATION_META_DEFINITION                            \

torch/csrc/autograd/VariableTypeUtils.h

@@ -29,8 +29,7 @@
 #endif
 #endif
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 enum class can_mutate_inplace_result {
   success,
   non_default_backward_view,
@@ -438,5 +437,4 @@ Return run_jit_decomposition_with_args_for_jvp(
 
 } // namespace impl
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/anomaly_mode.cpp

@@ -4,8 +4,7 @@
 #include <torch/csrc/autograd/function.h>
 #include <mutex>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 bool AnomalyMode::_enabled = false;
 bool AnomalyMode::_check_nan = true;
@@ -75,5 +74,4 @@ void AnomalyMetadata::assign_parent(const std::shared_ptr<Node>& parent_node) {
   parent_ = parent_node;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/autograd.cpp

@@ -14,8 +14,7 @@
 
 #include <c10/util/irange.h>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 // NB: This code duplicates existing logic at torch/autograd/__init__.py and
 // torch._C._EngineBase.run_backward in torch/csrc/autograd/python_engine.cpp
@@ -215,5 +214,4 @@ void exit_dual_level(uint64_t level) {
 
 } // namespace forward_ad
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/autograd_meta.cpp

@@ -10,8 +10,7 @@
 #include <ATen/ops/zeros.h>
 #endif
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 using at::Tensor;
 
@@ -315,5 +314,4 @@ const Variable& AutogradMeta::fw_grad(
   return direct_fw_grad;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/autograd_not_implemented_fallback.cpp

@@ -17,8 +17,7 @@
 #include <utility>
 #include <vector>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 namespace {
 
@@ -630,5 +629,4 @@ torch::CppFunction autogradNotImplementedInplaceOrViewFallback() {
       &autogradNotImplementedInplaceOrViewFallbackImpl>();
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/cpp_hook.cpp

@@ -19,8 +19,7 @@ void check_single_result(
 }
 } // namespace
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 CppFunctionTensorPreHook::CppFunctionTensorPreHook(
     std::shared_ptr<hooks_list> hooks,
@@ -65,5 +64,4 @@ variable_list CppFunctionSingleTensorPreHook::operator()(
   return results;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/custom_function.cpp

@@ -97,7 +97,6 @@ static void _process_forward_mode_AD(
     forward_grads = jvp_user_function(inputs, std::move(input_grads));
   }
 
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   const auto num_forward_grads = forward_grads.size();
   // contrary to backward mode, we don't allow returning too many gradients
   TORCH_CHECK(

torch/csrc/autograd/custom_function.h

@@ -124,7 +124,6 @@ struct TORCH_API AutogradContext {
   AutogradContext& operator=(const AutogradContext& other) = delete;
 
   /// Can be used to save non-variable data for `backward`.
-  // NOLINTNEXTLINE(cppcoreguidelines-non-private-member-variables-in-classes)
   ska::flat_hash_map<std::string, at::IValue> saved_data;
 
   /// Saves the list of variables for a future call to `backward`. This
@@ -248,7 +247,6 @@ struct ExtractVariables : IterArgs<ExtractVariables> {
   ExtractVariables(std::vector<bool>& is_var, variable_list& list)
       : is_var_(is_var), list_(list) {}
   void operator()(const std::optional<at::Tensor>& x) {
-    // NOLINTNEXTLINE(bugprone-branch-clone)
     if (x.has_value() && x.value().defined()) {
       is_var_.push_back(true);
       list_.emplace_back(x.value());
@@ -283,7 +281,6 @@ inline void extract_vars(
 template <typename T>
 std::enable_if_t<std::is_same_v<T, variable_list>, T> to_output_type(
     std::vector<std::optional<Variable>>& output_list) {
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   variable_list result;
   std::transform(
       output_list.begin(),
@@ -304,7 +301,6 @@ inline std::vector<std::optional<Variable>> to_optional(Variable& output) {
 }
 
 inline std::vector<std::optional<Variable>> to_optional(variable_list& output) {
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   std::vector<std::optional<Variable>> result;
   std::transform(
       output.begin(),
@@ -327,7 +323,6 @@ auto Function<T>::apply(Args&&... args)
   }
 
   std::shared_ptr<CppNode<T>> node(new CppNode<T>(), deleteNode);
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   variable_list input_vars;
 
   const size_t num_inputs = sizeof...(Args);
@@ -336,7 +331,6 @@ auto Function<T>::apply(Args&&... args)
   // TODO Add tracing here
   extract_vars(node->is_variable_input_, input_vars, args...);
 
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   bool is_executable =
       GradMode::is_enabled() && any_variable_requires_grad(input_vars);
   auto next_edges =
@@ -351,7 +345,6 @@ auto Function<T>::apply(Args&&... args)
   }
 
   using forward_return_t = forward_t<X, Args...>;
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   forward_return_t outputs;
   {
     AutoGradMode grad_mode(false);
@@ -406,7 +399,6 @@ variable_list CppNode<T>::apply(variable_list&& inputs) {
   at::OptionalDeviceGuard _device_guard;
 
   auto num_inputs = inputs.size();
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   variable_list backward_inputs;
   backward_inputs.reserve(num_inputs);
   for (const auto i : c10::irange(num_inputs)) {
@@ -449,7 +441,6 @@ variable_list CppNode<T>::apply(variable_list&& inputs) {
     throw std::runtime_error(msg);
   }
 
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   variable_list results;
   results.reserve(num_outputs);
   for (const auto i : c10::irange(num_outputs)) {

torch/csrc/autograd/forward_grad.cpp

@@ -1,7 +1,6 @@
 #include <torch/csrc/autograd/forward_grad.h>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 namespace {
 // See discussion in forward_grad.h for why these are global variables and not
@@ -76,5 +75,4 @@ const at::Tensor& ForwardGrad::undef_grad() {
   return singleton_undefined_tensor;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/function.cpp

@@ -17,8 +17,7 @@
 #include <ATen/ops/zeros.h>
 #endif
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 // The current evaluating node. This is useful to assign the current node as a
 // parent of new nodes created during the evaluation of this node in anomaly
@@ -110,5 +109,4 @@ at::Tensor TypeAndSize::zeros() {
   return at::zeros_symint(sym_sizes, options);
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/init.cpp

@@ -1280,7 +1280,7 @@ PyObject* THPModule_increment_version(
 }
 
 // autograd methods on torch._C
-static PyMethodDef methods[] = { // NOLINT
+static PyMethodDef methods[] = {
     {"_set_grad_enabled",
      castPyCFunctionWithKeywords(set_grad_enabled),
      METH_VARARGS | METH_KEYWORDS,

torch/csrc/autograd/input_buffer.cpp

@@ -17,8 +17,7 @@
 #include <utility>
 #include <vector>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 namespace {
 // look what you made me do >.<
@@ -246,5 +245,4 @@ auto InputBuffer::variables(InputBuffer&& g) -> std::vector<Variable> {
   return result;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/input_metadata.cpp

@@ -3,8 +3,7 @@
 // TODO: we may be able to move some imports from input_metadata.h to here, but
 // it seems that function.h transitively depends on some of them.
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 namespace {
 
@@ -158,7 +157,6 @@ std::stringstream InputMetadata::incompatible_shape_error_message(
 }
 
 bool InputMetadata::is_cpp_nested_tensor() const {
-  // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
   bool ret = std::holds_alternative<at::Tensor>(shape_);
   TORCH_INTERNAL_ASSERT(ret == (is_nested_ && !is_tensor_subclass_))
   return ret;
@@ -203,5 +201,4 @@ bool InputMetadata::maybe_expandable_to(const at::Tensor& grad) const {
   }
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/jit_decomp_interface.cpp

@@ -1,8 +1,6 @@
 #include <torch/csrc/autograd/jit_decomp_interface.h>
 
-namespace torch {
-namespace autograd {
-namespace impl {
+namespace torch::autograd::impl {
 
 namespace {
 JitDecompInterface* impl = nullptr;
@@ -16,6 +14,4 @@ JitDecompInterface* getJitDecompImpl() {
   return impl;
 }
 
-} // namespace impl
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::impl

torch/csrc/autograd/profiler_legacy.cpp

@@ -645,7 +645,7 @@ RecordProfile::RecordProfile(std::ostream& out) : out_(out) {
 }
 
 RecordProfile::RecordProfile(const std::string& filename)
-    : file_(new std::ofstream(filename)), out_(*file_) {
+    : file_(std::make_unique<std::ofstream>(filename)), out_(*file_) {
   init();
 }
 

torch/csrc/autograd/profiler_legacy.h

@@ -12,11 +12,7 @@
 #include <torch/csrc/profiler/stubs/base.h>
 #include <torch/csrc/profiler/util.h>
 
-namespace torch::autograd {
-
-struct Node;
-
-namespace profiler {
+namespace torch::autograd::profiler {
 
 enum class C10_API_ENUM EventKind : uint16_t {
   Mark,
@@ -402,5 +398,4 @@ struct TORCH_API TLSLegacyProfilerGuard {
   const std::optional<ProfilerDisableOptions> profilerDisableOptions_;
 };
 
-} // namespace profiler
-} // namespace torch::autograd
+} // namespace torch::autograd::profiler

torch/csrc/autograd/python_anomaly_mode.cpp

@@ -8,8 +8,7 @@
 #include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/utils/python_strings.h>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 void PyAnomalyMetadata::store_stack() {
   pybind11::gil_scoped_acquire gil;
@@ -125,5 +124,4 @@ void _print_stack(
   }
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/python_anomaly_mode.h

@@ -5,8 +5,7 @@
 #include <torch/csrc/python_headers.h>
 #include <torch/csrc/utils/pybind.h>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 struct PyAnomalyMetadata : public AnomalyMetadata {
   static constexpr const char* ANOMALY_TRACE_KEY = "traceback_";
@@ -40,5 +39,4 @@ void _print_stack(
     const std::string& current_node_name,
     bool is_parent);
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/python_cpp_function.cpp

@@ -297,8 +297,7 @@ PyObject* functionToPyObject(const std::shared_ptr<Node>& cdata) {
   } else {
     auto& fn = *cdata;
     auto it = cpp_function_types_map.find(std::type_index(typeid(fn)));
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    PyTypeObject* type;
+    PyTypeObject* type = nullptr;
     if (it == cpp_function_types_map.end()) {
       type = get_default_type();
     } else {

torch/csrc/autograd/python_function.cpp

@@ -413,8 +413,7 @@ PyObject* PyNode::to_py_args(
     throw_python_error();
   auto& output_info = py_fn->output_info;
   for (const auto i : c10::irange(num_inputs)) {
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    PyObject* input;
+    PyObject* input = nullptr;
     if (inputs[i].defined() || !py_fn->materialize_grads ||
         (input_metadata(i).was_default_constructed() &&
          !py_fn->materialize_non_diff_grads)) {
@@ -1648,8 +1647,8 @@ PyObject* THPFunction_metadata(THPFunction* self, void* _unused) {
   END_HANDLE_TH_ERRORS
 }
 
-typedef PyObject* (*getter)(PyObject*, void*);
-typedef int (*setter)(PyObject*, PyObject*, void*);
+using getter = PyObject* (*)(PyObject*, void*);
+using setter = int (*)(PyObject*, PyObject*, void*);
 
 namespace {
 

torch/csrc/autograd/python_hook.cpp

@@ -26,8 +26,7 @@ static void check_single_result(
     PyObject* result,
     PyObject* hook);
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 namespace {
 
@@ -253,8 +252,7 @@ void PyFunctionTensorPostAccGradHooks::apply_with_saved(
   }
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd
 
 static PyObject* wrap_variables(const variable_list& c_variables) {
   size_t num_vars = c_variables.size();

torch/csrc/autograd/python_legacy_variable.cpp

@@ -10,8 +10,7 @@
 
 using namespace at;
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 static PyObject* THPVariable_pynew(
     PyTypeObject* type,
@@ -163,5 +162,4 @@ void init_legacy_variable(PyObject* module) {
   }
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/python_nested_functions_manual.cpp

@@ -3,8 +3,7 @@
 #include <torch/csrc/utils/python_arg_parser.h>
 #include <torch/torch.h>
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 static PyObject* THPVariable_nested_tensor(
     PyObject* /*self*/,
@@ -40,5 +39,4 @@ PyMethodDef* get_nested_functions_manual() {
   return nested_functions_manual;
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/python_torch_functions_manual.cpp

@@ -41,8 +41,7 @@ using at::TensorOptions;
 using torch::utils::check_out_type_matches;
 using namespace torch::autograd::utils;
 
-namespace torch {
-namespace autograd {
+namespace torch::autograd {
 
 // NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)
 PyObject* THPVariableFunctionsModule = nullptr;
@@ -796,5 +795,4 @@ void initTorchFunctions(PyObject* module) {
       });
 }
 
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd

torch/csrc/autograd/python_variable_indexing.h

@@ -43,8 +43,7 @@ inline UnpackedSlice __PySlice_Unpack(PyObject* _r) {
     if (torch::is_symint(r->step)) {
       step_sym = py::handle(r->step).cast<c10::SymInt>();
     } else {
-      // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-      Py_ssize_t step;
+      Py_ssize_t step = 0;
       if (!_PyEval_SliceIndex(r->step, &step)) {
         throw python_error();
       }
@@ -62,8 +61,7 @@ inline UnpackedSlice __PySlice_Unpack(PyObject* _r) {
   } else if (r->start == Py_None) {
     start_sym = c10::SymInt(step_sym < 0 ? PY_SSIZE_T_MAX : 0);
   } else {
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    Py_ssize_t start;
+    Py_ssize_t start = 0;
     if (!_PyEval_SliceIndex(r->start, &start)) {
       throw python_error();
     }
@@ -77,8 +75,7 @@ inline UnpackedSlice __PySlice_Unpack(PyObject* _r) {
     stop_sym = c10::SymInt(
         step_sym < 0 ? c10::SymInt::min_representable_int() : PY_SSIZE_T_MAX);
   } else {
-    // NOLINTNEXTLINE(cppcoreguidelines-init-variables)
-    Py_ssize_t stop;
+    Py_ssize_t stop = 0;
     if (!_PyEval_SliceIndex(r->stop, &stop)) {
       throw python_error();
     }

torch/csrc/autograd/record_function_ops.cpp

@@ -12,9 +12,7 @@ namespace caffe2 {
 CAFFE_KNOWN_TYPE(at::RecordFunction);
 } // namespace caffe2
 
-namespace torch {
-namespace autograd {
-namespace profiler {
+namespace torch::autograd::profiler {
 
 // Creates a new profiling scope using RecordFunction and invokes its starting
 // callbacks.
@@ -165,6 +163,4 @@ TORCH_LIBRARY_FRAGMENT(profiler, m) {
       c10::AliasAnalysisKind::FROM_SCHEMA));
 }
 
-} // namespace profiler
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::profiler

torch/csrc/autograd/utils/error_messages.h

@@ -2,9 +2,7 @@
 
 #include <sstream>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 inline std::string requires_grad_leaf_error(bool requires_grad) {
   std::ostringstream oss;
@@ -17,6 +15,4 @@ inline std::string requires_grad_leaf_error(bool requires_grad) {
   return oss.str();
 }
 
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils

torch/csrc/autograd/utils/grad_layout_contract.h

@@ -2,9 +2,7 @@
 
 #include <ATen/Tensor.h>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 // Helper functions to enforce the "Gradient Layout Contract" described in
 // torch/csrc/autograd/functions/accumulate_grad.h.
@@ -75,6 +73,4 @@ inline at::Tensor clone_obey_contract(
   }
 }
 
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils

torch/csrc/autograd/utils/python_arg_parsing.h

@@ -5,9 +5,7 @@
 
 #include <torch/csrc/utils/python_arg_parser.h>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 // The parameter allow_copy is to accept copy for Tensor.to (and by proxy
 // PackedSequences.to) but not nn.Module.to.
@@ -48,6 +46,4 @@ parse_to_conversion(PythonArgs& r, bool allow_copy) {
         r.memoryformatOptional(3));
   }
 }
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils

torch/csrc/autograd/utils/warnings.cpp

@@ -1,8 +1,6 @@
 #include <torch/csrc/autograd/utils/warnings.h>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 void DelayWarningHandler::process(const c10::Warning& warning) {
   std::lock_guard<std::mutex> lock(mutex_);
@@ -19,6 +17,4 @@ void DelayWarningHandler::replay_warnings() {
   }
 }
 
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils

torch/csrc/autograd/utils/warnings.h

@@ -4,9 +4,7 @@
 #include <mutex>
 #include <vector>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 // Warning handler for multi-threaded contexts. Gather warnings from
 // all threads into a single queue, then process together at the end
@@ -23,6 +21,4 @@ class DelayWarningHandler : public at::WarningHandler {
   std::mutex mutex_;
 };
 
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils

torch/csrc/autograd/utils/wrap_outputs.h

@@ -18,9 +18,7 @@
 #include <torch/csrc/utils/python_numbers.h>
 #include <torch/csrc/utils/tensor_qschemes.h>
 
-namespace torch {
-namespace autograd {
-namespace utils {
+namespace torch::autograd::utils {
 
 inline PyObject* wrap(bool value) {
   if (value) {
@@ -148,6 +146,4 @@ PyObject* wrap(PyTypeObject* type, std::tuple<Ts...> values) {
   return r.release();
 }
 
-} // namespace utils
-} // namespace autograd
-} // namespace torch
+} // namespace torch::autograd::utils