Status TestTensorHandleWithDims(AbstractContext* ctx, const T* data,
                                const int64_t* dims, int num_dims,
                                AbstractTensorHandle** tensor) {
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);
  TFE_Context* eager_ctx =
      TF_ExecutionContextGetTFEContext(wrap(ctx), status.get());

  ASSERT_EQ(TF_OK, TF_GetCode(s_)) << TF_Message(s_);

  EXPECT_TF_META("v", list.size(), TF_ATTR_STRING, list_total_size);
  std::unique_ptr<void*[]> values(new void*[list.size()]);
  std::unique_ptr<size_t[]> lens(new size_t[list.size()]);
  std::unique_ptr<char[]> storage(new char[list_total_size]);
  TF_OperationGetAttrStringList(oper, "v", values.get(), lens.get(),

#include "tensorflow/core/lib/llvm_rtti/llvm_rtti.h"
#include "tensorflow/core/platform/errors.h"

namespace tensorflow {

AbstractContext* BuildFunction(const char* fn_name) {
  std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)> status(
      TF_NewStatus(), TF_DeleteStatus);
  TF_ExecutionContext* graph_ctx = TF_CreateFunction(fn_name, status.get());
  return unwrap(graph_ctx);
}

  Tape tape(/*persistent=*/false);
  tape.Watch(inputs[0]);  // Watch x.
  AbstractTensorHandle* exp_output;
  TF_RETURN_IF_ERROR(ops::Exp(ctx, inputs[0], &exp_output, "Exp"));
  std::unique_ptr<GradientFunction> gradient_function(
      new PassThroughGradientFunction);
  tape.RecordOperation(inputs, {exp_output}, gradient_function.release());
  TF_RETURN_IF_ERROR(tape.ComputeGradient(ctx,

    }
    if (op_type_.empty()) {
      return errors::FailedPrecondition(
          "GraphOperation::Reset must be called before calling SetOpName.");
    }
    // TODO(b/145674566): We use Graph::NewName to get a unique name here but
    // this may not be consistent with python's naming policy.
    mutex_lock l(g_->mu);
    op_.reset(new TF_OperationDescription(g_, op_type_.c_str(),

  TFE_OpAddInput(op, var_handle, status);

  // Convert 'value' to a TF_Tensor then a TFE_TensorHandle.
  std::unique_ptr<TF_Tensor, decltype(&TF_DeleteTensor)> t(
      TF_AllocateTensor(TF_FLOAT, nullptr, 0, sizeof(value)), TF_DeleteTensor);
  memcpy(TF_TensorData(t.get()), &value, TF_TensorByteSize(t.get()));

  std::unique_ptr<TFE_TensorHandle, decltype(&TFE_DeleteTensorHandle)>
      value_handle(TFE_NewTensorHandle(t.get(), status),

    return nullptr;
  }
  std::vector<std::unique_ptr<tensorflow::Device>> devices;
  status->status = tensorflow::DeviceFactory::AddDevices(
      opts->session_options.options, "/job:localhost/replica:0/task:0",
      &devices);
  if (!status->status.ok()) return nullptr;
  std::unique_ptr<tensorflow::DeviceMgr> device_mgr(
      new tensorflow::DynamicDeviceMgr(std::move(devices)));

  # Column 1 is left only, Column 2 is right only, Column 3 is shared lines
  # Select lines unique to actual_licenses, i.e. extra licenses.
  comm -1 -3 $BATS_TEST_TMPDIR/expected_licenses $BATS_TEST_TMPDIR/actual_licenses | grep -v -f $BATS_TEST_TMPDIR/allowed_to_be_extra > $BATS_TEST_TMPDIR/actual_extra_licenses || true
  # Select lines unique to expected_licenses, i.e. missing licenses

using tensorflow::FunctionDef;
using tensorflow::Node;
using tensorflow::NodeBuilder;
using tensorflow::Status;
using tensorflow::errors::InvalidArgument;

namespace {
typedef std::unique_ptr<TF_Function, decltype(&TF_DeleteFunction)>
    UniqueFuncPtr;
}

// struct TF_Operation { tensorflow::Node node; };
static TF_Operation* ToTF_Operation(Node* node) {

  // TODO(apassos) consider narrowing down this interface.
  BackwardFunction* backward_function;

  // Should be called before deleting the backward function. TODO(apassos) use
  // unique_ptrs to ensure this happens.
  std::function<void(BackwardFunction*)> backward_function_deleter;
};

// Map from tensor_id to internally-defined operation-id of the operation which