void TF_AbstractOpSetOpType(TF_AbstractOp* op, const char* const op_type,
                            TF_Status* s) {
  tsl::Set_TF_Status_from_Status(
      s, unwrap(op)->Reset(op_type,
                           /*raw_device_name=*/nullptr));
}

void TF_AbstractOpSetOpName(TF_AbstractOp* op, const char* const op_name,
                            TF_Status* s) {
  TracingOperation* tracing_op = dyn_cast<TracingOperation>(unwrap(op));

                                   absl::Span<AbstractTensorHandle* const>,
                                   absl::Span<AbstractTensorHandle*>)>;

// Runs `model` maybe wrapped in a function call op. This can be thought as
// being equivalent to the following python code.
//
// if use_function:
//   outputs = tf.function(model)(inputs)
// else:
//   outputs = model(inputs)

GradientFunction* ReluRegisterer(const ForwardOperation& op) {
  return new ReluGradientFunction(op.outputs);
}

GradientFunction* SparseSoftmaxCrossEntropyWithLogitsRegisterer(
    const ForwardOperation& op) {
  return new SparseSoftmaxCrossEntropyWithLogitsGradientFunction(op.outputs);
}

GradientFunction* BiasAddRegisterer(const ForwardOperation& op) {
  return new BiasAddGradientFunction(op.attrs);
}

}  // namespace gradients

        "//tensorflow/c/eager:c_api",
        "//tensorflow/c/eager:c_api_experimental",
        "//tensorflow/c/eager:tfe_cancellation_manager_internal",
        "//tensorflow/c/eager:tfe_op_internal",
        "//tensorflow/c/eager:tfe_tensorhandle_internal",
        "//tensorflow/core:framework",
        "//tensorflow/core:lib",
        "@com_google_absl//absl/types:optional",

namespace tracing {

Status MaybeSetOpName(AbstractOperation* op, const char* op_name) {
  if (isa<TracingOperation>(op)) {
    TF_RETURN_IF_ERROR(dyn_cast<TracingOperation>(op)->SetOpName(op_name));
  }
  if (isa<gradients::TapeOperation>(op)) {
    TF_RETURN_IF_ERROR(MaybeSetOpName(
        dyn_cast<gradients::TapeOperation>(op)->GetBackingOperation(),
        op_name));
  }
  return absl::OkStatus();
}
}  // namespace tracing

    }

    AbstractOperationPtr fn_op(ctx->CreateOperation());
    TF_RETURN_IF_ERROR(fn_op->Reset(fn_name, /*raw_device_name=*/nullptr));
    for (auto input : inputs) {
      TF_RETURN_IF_ERROR(fn_op->AddInput(input));
    }
    int retvals = outputs.size() - null_indices.size();
    std::vector<AbstractTensorHandle*> fn_outputs(retvals);
    TF_RETURN_IF_ERROR(fn_op->Execute(

// Return an add op multiplying `a` by `b`.
TFE_Op* AddOp(TFE_Context* ctx, TFE_TensorHandle* a, TFE_TensorHandle* b);

// Return a matmul op multiplying `a` by `b`.
TFE_Op* MatMulOp(TFE_Context* ctx, TFE_TensorHandle* a, TFE_TensorHandle* b);

// Return an identity op.
TFE_Op* IdentityOp(TFE_Context* ctx, TFE_TensorHandle* a);

// Return a shape op fetching the shape of `a`.

  }

  AbstractOperationPtr check_numerics_op(ctx->CreateOperation());
  ForwardOperation forward_op;
  Status s = Reset(check_numerics_op.get(), "CheckNumerics",
                   /*raw_device_name=*/nullptr, &forward_op);
  ASSERT_EQ(errors::OK, s.code()) << s.message();
  if (isa<TracingOperation>(check_numerics_op.get())) {
    s = dyn_cast<TracingOperation>(check_numerics_op.get())
            ->SetOpName("check_numerics");

      }
      grad_inputs[i] = grad_input;
    }
    return absl::OkStatus();
  }
  ~IdentityNGradientFunction() override {}
};
}  // namespace

GradientFunction* IdentityNRegisterer(const ForwardOperation& op) {
  return new IdentityNGradientFunction;
}

}  // namespace gradients

inspected and debugged and it is intended to be used during the development
phase.

As part of the differences that make Eager mode easier to debug, the [shape
inference
functions](https://www.tensorflow.org/guide/create_op#define_the_op_interface)
are skipped, and any checks implemented inside the shape inference code are not
executed.

The security impact of skipping those checks should be low, since the attack