Running our linter on a lot of files.

Change: 112920860

WORKSPACE

@@ -21,8 +21,8 @@ new_http_archive(
 
 new_http_archive(
   name = "eigen_archive",
-  url = "https://bitbucket.org/eigen/eigen/get/fb2fa05.tar.gz",
-  sha256 = "8aacd8065d52528af1a22d6b72925dbb6b9fb8f25e46769481dd06d3edf63bbd",
+  url = "https://bitbucket.org/eigen/eigen/get/c8e5d09.tar.gz",
+  sha256 = "be61d1ce686e950d9a3a61b2aa83562ba4582ce1813e475869146e0a05915857",
   build_file = "eigen.BUILD",
 )
 

configure

@@ -92,7 +92,7 @@ while true; do
 done
 
 cat > third_party/gpus/cuda/cuda.config <<EOF
-# CUDA_TOOLKIT_PATH refers to the CUDA toolkit. Tensorflow requries Cuda 7.0
+# CUDA_TOOLKIT_PATH refers to the CUDA toolkit. Tensorflow requires Cuda 7.0
 # at the moment.
 CUDA_TOOLKIT_PATH="$CUDA_TOOLKIT_PATH"
 

eigen.BUILD

@@ -1,6 +1,6 @@
 package(default_visibility = ["//visibility:public"])
 
-archive_dir = "eigen-eigen-fb2fa0527077"
+archive_dir = "eigen-eigen-c8e5d094f3a9"
 
 cc_library(
     name = "eigen",

tensorflow/cc/ops/cc_op_gen.cc

@@ -169,7 +169,7 @@ void WriteCCOp(const OpDef& op_def, WritableFile* h, WritableFile* cc) {
                      R"comment(  .WithName(StringPiece): Set the Node's name
   .WithDevice(StringPiece): Set the Node's requested device
   .WithControlInput(Node*) / .WithControlInputs({Node*, ...}):
-    Add control depencies on the specified Node(s).
+    Add control dependencies on the specified Node(s).
 
 Returns a pointer to the created Node)comment");
 

tensorflow/core/common_runtime/direct_session.cc

@@ -94,7 +94,7 @@ string GetRendezvousKey(const string& tensor_name,
 //
 // 2) Recv nodes always complete immediately: The inputs are sent into
 //    the local rendezvous before we start the executor, so the
-//    corresonding recvs will not block.
+//    corresponding recvs will not block.
 //
 // Based on these assumptions, we can use the same thread pool for
 // both "non-blocking" and "blocking" OpKernels on Android.
@@ -543,7 +543,7 @@ Status DirectSession::CreateGraphs(
     }
   }
 
-  for (auto partition : partitions) {
+  for (auto&& partition : partitions) {
     const string& partition_name = partition.first;
 
     GraphDef* graph_def = &partition.second;

tensorflow/core/common_runtime/direct_session_test.cc

@@ -95,7 +95,7 @@ TEST_F(DirectSessionMinusAXTest, RunSimpleNetwork) {
   ASSERT_OK(s);
 
   ASSERT_EQ(1, outputs.size());
-  // The first output should be initiailzed and have the correct
+  // The first output should be initialized and have the correct
   // output.
   auto mat = outputs[0].matrix<float>();
   ASSERT_TRUE(outputs[0].IsInitialized());

tensorflow/core/common_runtime/executor.cc

@@ -383,7 +383,7 @@ Status ExecutorImpl::InferAllocAttr(
   return s;
 }
 
-// The state associated with one invokation of ExecutorImpl::Run.
+// The state associated with one invocation of ExecutorImpl::Run.
 // ExecutorState dispatches nodes when they become ready and keeps
 // track of how many predecessors of a node have not done (pending_).
 class ExecutorState {

tensorflow/core/common_runtime/function.cc

@@ -479,7 +479,7 @@ Status FunctionLibraryRuntimeImpl::InstantiateSymbolicGradient(
   const auto& func = f->func();
   const FunctionDef* fdef = lib_def_->Find(func.name());
   if (fdef == nullptr) {
-    // f is a primitve op.
+    // f is a primitive op.
     gradient::Creator creator;
     TF_RETURN_IF_ERROR(gradient::GetOpGradientCreator(func.name(), &creator));
     if (creator == nullptr) {
@@ -1169,7 +1169,7 @@ class SymbolicGradientHelper {
 
   // 'ready' keeps track of nodes that have been completely
   // backpropped. Initially, for every output y of the function f, we
-  // add dy as an input of the the gradient function.
+  // add dy as an input of the gradient function.
   std::deque<Node*> ready_;
 
   // Makes a copy of fbody_ in gbody_.

tensorflow/core/common_runtime/function.h

@@ -90,7 +90,7 @@ bool RemoveListArrayConverter(Graph* g);
 // multiple times by calling ExpandInlineFunctions a few times.
 bool ExpandInlineFunctions(FunctionLibraryRuntime* lib, Graph* graph);
 
-// Applies graph rewrite optimzation such as inlining, dead code
+// Applies graph rewrite optimization such as inlining, dead code
 // removal, etc.
 //
 // **g is a graph constructed based on the runtime library 'lib'.

tensorflow/core/common_runtime/gpu/gpu_allocator_retry.h

@@ -32,7 +32,7 @@ class GPUAllocatorRetry {
   // then wait up to 'max_millis_to_wait' milliseconds, retrying each
   // time a call to DeallocateRaw() is detected, until either a good
   // pointer is returned or the deadline is exhausted.  If the
-  // deadline is exahusted, try one more time with 'verbose_failure'
+  // deadline is exhausted, try one more time with 'verbose_failure'
   // set to true.  The value returned is either the first good pointer
   // obtained from 'alloc_func' or nullptr.
   void* AllocateRaw(std::function<void*(size_t alignment, size_t num_bytes,

tensorflow/core/common_runtime/gpu/gpu_bfc_allocator.cc

@@ -441,7 +441,7 @@ void GPUBFCAllocator::DumpMemoryLog(size_t num_bytes) {
     }
   }
 
-  // Next show the the chunks that are in use, and also summarize their
+  // Next show the chunks that are in use, and also summarize their
   // number by size.
   std::map<size_t, int> in_use_by_size;
   for (auto& it : ptr_to_chunk_map_) {

tensorflow/core/common_runtime/gpu/gpu_event_mgr_test.cc

@@ -198,7 +198,7 @@ TEST(EventMgr, StreamSwitchingFlushesImmediately) {
   EXPECT_GT(initial_live_bytes, live_tensor_bytes);
 }
 
-TEST(EventMgr, ManySmallTensorsSeperateCallsFlushed) {
+TEST(EventMgr, ManySmallTensorsSeparateCallsFlushed) {
   auto stream_exec = GPUMachineManager()->ExecutorForDevice(0).ValueOrDie();
   EventMgr em(stream_exec, GPUOptions());
   TEST_EventMgrHelper th(&em);

tensorflow/core/common_runtime/gpu/gpu_region_allocator.cc

@@ -280,7 +280,7 @@ void GPURegionAllocator::CheckForMemoryLeaks() {
 }
 
 // Since there's no merging of chunks once allocated, we want to
-// maximize their reusablity (which argues for fewer, larger sizes),
+// maximize their reusability (which argues for fewer, larger sizes),
 // while minimizing waste (which argues for tight-fitting sizes).
 //
 // The smallest unit of allocation is 256 bytes.

tensorflow/core/common_runtime/gpu/gpu_stream_util.cc

@@ -61,7 +61,7 @@ Status AssignStreams(const Graph* graph, const AssignStreamsOpts& opts,
       }
     }
   }
-  // We perform stream assigmnent assuming a large number of
+  // We perform stream assignment assuming a large number of
   // stream IDs and then map these down to the required number of streams
   // using simple round-robin.
   // Stream Assignment strategy:

tensorflow/core/common_runtime/gpu/gpu_util.h

@@ -34,7 +34,7 @@ class GPUUtil {
   // "tensor" is GPU-local.  "dev" is the hosting GPU.
   // "device_context" should be the context of the GPU "_Send" op
   // which provides the Tensor.
-  // Sets all necessasry fields of "proto" by transferring value
+  // Sets all necessary fields of "proto" by transferring value
   // bytes from GPU to CPU RAM. "is_dead" indicates that the
   // tensor is dead with an uninit value.
   static void SetProtoFromGPU(const Tensor& tensor, Device* dev,

tensorflow/core/common_runtime/gpu/pool_allocator.cc

@@ -47,7 +47,7 @@ PoolAllocator::PoolAllocator(size_t pool_size_limit, bool auto_resize,
 PoolAllocator::~PoolAllocator() { Clear(); }
 
 namespace {
-// Pools contain Chunks allocatated from the underlying Allocator.
+// Pools contain Chunks allocated from the underlying Allocator.
 // Chunk alignment is always on kPoolAlignment boundaries.  Each Chunk
 // begins with a descriptor (ChunkPrefix) that gives its size and a
 // pointer to itself.  The pointer returned to the user is just past
@@ -56,7 +56,7 @@ namespace {
 // pointer and also re-write the ChunkPrefix.chunk_ptr value
 // immediately before it.  This way the Chunk address and size can be
 // recovered from the returned user pointer, regardless of alignment.
-// Note that this deferencing of the pointers means that we cannot
+// Note that this dereferencing of the pointers means that we cannot
 // handle GPU memory, only CPU memory.
 struct ChunkPrefix {
   size_t num_bytes;

tensorflow/core/common_runtime/gpu/process_state.cc

@@ -47,7 +47,7 @@ const bool FLAGS_brain_gpu_region_allocator_reset_to_nan = false;
 const bool FLAGS_brain_gpu_use_bfc_allocator = true;
 
 // If true, record attributes of memory allocations and
-// dyanmically check for appropriate use of registered memory.
+// dynamically check for appropriate use of registered memory.
 // Should only be true for debugging or diagnosis of
 // performance issues.
 bool FLAGS_brain_gpu_record_mem_types = false;

tensorflow/core/common_runtime/gpu/process_state.h

@@ -67,7 +67,7 @@ class ProcessState {
   MemDesc PtrType(const void* ptr);
 
   // Returns the one CPUAllocator used for the given numa_node.
-  // TEMPORY: ignores numa_node.
+  // TEMPORARY: ignores numa_node.
   Allocator* GetCPUAllocator(int numa_node);
 
   // Returns the one GPU allocator used for the indexed GPU.
@@ -80,7 +80,7 @@ class ProcessState {
   // used on that first call is used.
   //
   // "Allocator type" describes the type of algorithm to use for the
-  // underlying allocator.  REQURES: Must be a valid type (see
+  // underlying allocator.  REQUIRES: Must be a valid type (see
   // config.proto for the list of supported strings.).
   //
   // REQUIRES: gpu_id must be a valid ordinal for a GPU available in the
@@ -98,7 +98,7 @@ class ProcessState {
   // interface to be used for network device memory registration.
   // "bus_id" is platform-specific.  On many platforms it
   // should be 0.  On machines with multiple PCIe buses, it should be
-  // the index of one of the PCIe buses.  If the the bus_id is invalid,
+  // the index of one of the PCIe buses.  If the bus_id is invalid,
   // results are undefined.
   typedef std::function<void(void*, size_t)> AllocVisitor;
   void AddGPUAllocVisitor(int bus_id, AllocVisitor visitor);

tensorflow/core/common_runtime/kernel_benchmark_testlib.h

@@ -29,7 +29,7 @@ limitations under the License.
 namespace tensorflow {
 
 class Device;
-class SessionOptions;
+struct SessionOptions;
 
 namespace test {
 

tensorflow/core/common_runtime/local_device.h

@@ -22,7 +22,7 @@ limitations under the License.
 
 namespace tensorflow {
 
-class SessionOptions;
+struct SessionOptions;
 
 // This class is shared by ThreadPoolDevice and GPUDevice and
 // initializes a shared Eigen compute device used by both.  This

tensorflow/core/common_runtime/session_factory.h

@@ -25,7 +25,7 @@ limitations under the License.
 namespace tensorflow {
 
 class Session;
-class SessionOptions;
+struct SessionOptions;
 
 class SessionFactory {
  public:

tensorflow/core/common_runtime/simple_placer.cc

@@ -37,7 +37,7 @@ namespace {
 // types in 'supported_device_types' and returns the *first* subset of devices
 // that match.
 //
-// For example, if suported_device_types contains {GPU, CPU} and
+// For example, if supported_device_types contains {GPU, CPU} and
 // 'devices' contains CPU and GPU devices, the returned vector will
 // include *only* GPU devices, since that is higher in the priority
 // order in 'supported_device_types'.

tensorflow/core/framework/attr_value_util.h

@@ -36,7 +36,7 @@ string SummarizeAttrValue(const AttrValue& attr_value);
 // Generates an error if attr_value doesn't have the indicated attr type.
 Status AttrValueHasType(const AttrValue& attr_value, StringPiece type);
 
-// Converts a text proto value from "text" into the the field of *out
+// Converts a text proto value from "text" into the field of *out
 // indicated by "type" (e.g. from the type field of an AttrDef).
 // Examples:
 // * If type:"int" and text:"-14", then *out is set to "i: -14"

tensorflow/core/framework/function.proto

@@ -61,7 +61,7 @@ message FunctionDef {
     // 'attr' maps names defined by 'func's attr defs to attr values.
     // attr values may have placeholders which are substituted
     // recursively by concrete values when this node is instantiated.
-    // These placeholdes must name an attr listed in the FunctionDef's
+    // These placeholders must name an attr listed in the FunctionDef's
     // signature.
     map<string, AttrValue> attr = 5;
   }

tensorflow/core/framework/function_test.cc

@@ -296,8 +296,8 @@ REGISTER_OP("Cond")
 output = Cond(input) ? then_branch(input) : else_branch(input)
 
 cond: A function takes 'input' and returns a scalar.
-then_branch: A funcion takes 'input' and returns 'output'.
-else_branch: A funcion takes 'input' and returns 'output'.
+then_branch: A function takes 'input' and returns 'output'.
+else_branch: A function takes 'input' and returns 'output'.
 )doc");
 
 TEST(TFunc, Body_Array_List_Converter) {

tensorflow/core/framework/node_def_util.h

@@ -146,7 +146,7 @@ Status ValidateNodeDef(const NodeDef& node_def, const OpDef& op_def);
 
 // Computes the mapping from input/output argument name to the
 // corresponding input/output index range.  For example,
-// input "foo" coresponds to input indices
+// input "foo" corresponds to input indices
 //   [ (*inputs)["foo"].first, (*inputs)["foo"].second ).
 typedef std::unordered_map<string, std::pair<int, int>> NameRangeMap;
 Status NameRangesForNode(const NodeDef& node_def, const OpDef& op_def,

tensorflow/core/framework/op.h

@@ -72,7 +72,7 @@ class OpRegistry : public OpRegistryInterface {
   const OpDef* LookUp(const string& op_type_name,
                       Status* status) const override;
 
-  // Fills *ops with all registered OpDefss (except those with names
+  // Fills *ops with all registered OpDefs (except those with names
   // starting with '_' if include_internal == false).
   void Export(bool include_internal, OpList* ops) const;
 

tensorflow/core/framework/op_compatibility_test.cc

@@ -53,7 +53,7 @@ class OpCompatibilityTest : public OpsTestBase {
     ASSERT_OK(InOutTypesForNode(*node_def(), old_op_def, &old_in_types,
                                 &old_out_types));
 
-    // This should be all that is needed to get compatiblity.
+    // This should be all that is needed to get compatibility.
     const OpDef* new_op_def = RegisteredOpDef();
     AddDefaultsToNodeDef(*new_op_def, node_def());
 
@@ -93,7 +93,7 @@ class OpCompatibilityTest : public OpsTestBase {
     ASSERT_OK(InOutTypesForNode(*node_def(), old_op_def, &old_in_types,
                                 &old_out_types));
 
-    // This should be all that is needed to get compatiblity.
+    // This should be all that is needed to get compatibility.
     const OpDef* new_op_def = RegisteredOpDef();
     AddDefaultsToNodeDef(*new_op_def, node_def());
 
@@ -117,7 +117,7 @@ class OpCompatibilityTest : public OpsTestBase {
     ASSERT_OK(InOutTypesForNode(*node_def(), old_op_def, &old_in_types,
                                 &old_out_types));
 
-    // This should be all that is needed to get compatiblity.
+    // This should be all that is needed to get compatibility.
     const OpDef* new_op_def = RegisteredOpDef();
     AddDefaultsToNodeDef(*new_op_def, node_def());
 

tensorflow/core/framework/op_def_builder.h

@@ -64,7 +64,7 @@ class OpDefBuilder {
   // Perhaps by linking the type of the tensor to a type attr?
   OpDefBuilder& Attr(StringPiece spec);
 
-  // Adds an input or ouput to this OpDefBuilder (and returns *this).
+  // Adds an input or output to this OpDefBuilder (and returns *this).
   // The spec has form "<name>:<type-expr>" or "<name>:Ref(<type-expr>)"
   // where <name> matches regexp [a-z][a-z0-9_]* and <type-expr> can be:
   // * For a single tensor: <type>

tensorflow/core/framework/op_kernel.h

@@ -578,7 +578,7 @@ class OpKernelContext {
   Status mutable_input(const string& name, Tensor* tensor, bool lock_held);
 
   // Returns the named list-valued mutable input in "list", as defined
-  // in the OpDef.  If the named intput is not list-valued, returns a
+  // in the OpDef.  If the named input is not list-valued, returns a
   // one-element list. Must be used to access Ref inputs. The values
   // stored in the Tensor buffer may be modified, and modifications
   // will be visible to other Ops reading the same ref tensor.

tensorflow/core/framework/reader_interface.h

@@ -50,7 +50,7 @@ class ReaderInterface : public ResourceBase {
  public:
   // Read a single record into *key / *value.  May get more work from
   // *queue if the current work is complete.  Sets the status on
-  // *context with an OutOfRange Status if the the current work is
+  // *context with an OutOfRange Status if the current work is
   // complete and the queue is done (closed and empty).
   // This method may block.
   virtual void Read(QueueInterface* queue, string* key, string* value,

tensorflow/core/framework/rendezvous.h

@@ -44,7 +44,7 @@ class Rendezvous : public core::RefCounted {
     AllocatorAttributes alloc_attrs;
   };
 
-  // Constructs a rendezvouz key for the tensor of "name" sent from
+  // Constructs a rendezvous key for the tensor of "name" sent from
   // "src_device" to "dst_device". The tensor is generated in the frame
   // and iteration specified by "frame_iter".
   static string CreateKey(const string& src_device, uint64 src_incarnation,

tensorflow/core/framework/tensor.proto

@@ -24,8 +24,8 @@ message TensorProto {
   // to represent a constant Tensor with a single value.
   int32 version_number = 3;
 
-  // Serialized content from TensorBase::Serialize() This representation can be
-  // used for all tensor types.
+  // Serialized content from Tensor::AsProtoTensorContent(). This representation
+  // can be used for all tensor types.
   bytes tensor_content = 4;
 
   // Type specific representations that make it easy to create tensor protos in

tensorflow/core/framework/tensor_slice.h

@@ -38,7 +38,7 @@ class TensorSlice {
   // -- from just a dimension (in this case it will create a full slice)
   // -- from an array of pairs of integers.
   // -- from a TensorSliceProto protocol buffer
-  // -- from a string format of "start,lenth:start,length..." where each
+  // -- from a string format of "start,length:start,length..." where each
   //    "start,length" pair represents the slice on one dimension. We allow a
   //    special "-" that means "everything for this dimension". One such example
   //    is:  0,10:-:14,1:-:-

tensorflow/core/framework/tensor_testutil.h

@@ -84,7 +84,7 @@ template <typename T>
 void ExpectTensorEqual(const Tensor& x, const Tensor& y);
 
 // Expects "x" and "y" are tensors of the same type, same shape, and
-// approxmiate equal values, each within "abs_err".
+// approximate equal values, each within "abs_err".
 template <typename T>
 void ExpectTensorNear(const Tensor& x, const Tensor& y, const T& abs_err);