overlapping-outputs-input-comparison.png...

    handle.set_hash_code(type_index.hash_code());
    handle.set_maybe_type_name(type_index.name());

    Tensor* input = new Tensor(host_allocator_, DT_RESOURCE, TensorShape({}));
    input->scalar<ResourceHandle>()() = handle;
    tensors_.push_back(input);
    inputs_.push_back({nullptr, input});
  }

 protected:
  core::RefCountPtr<DeviceContext> device_context_;
  Allocator* host_allocator_;

    return absl::OkStatus();
  }
};

// Computes:
//
// @tf.custom_gradient
// def f(input):
//   def grad(grads):
//     return grads[0]
//   return tf.exp(input), grad
// outputs = [f(inputs[0])]
Status ExpWithPassThroughGrad(AbstractContext* ctx,
                              absl::Span<AbstractTensorHandle* const> inputs,
                              absl::Span<AbstractTensorHandle*> outputs) {

  if (!func || !func.isExternal()) return failure();
  // Get the inputs and attributes. The attributes include these from the
  // argument list and also these derived from the inputs.
  SmallVector<Value, 4> inputs;
  NamedAttrList argument_attrs;
  llvm::StringMap<Attribute> derived_attrs;
  if (failed(CollectInputsAndAttributes(rewriter, func, call_op, &inputs,
                                        &argument_attrs, &derived_attrs))) {

    // expected-error @+1 {{'tf.TPUReplicatedInput' op TF2XLA TPU bridge input check: number of inputs inconsistent. num_replicas=2 no. of inputs=3}}

  // The name assigned to the function which is the import result of the given
  // graph. If empty, a default one will be used.
  std::string graph_func_name;
  // Maps input node names to node data types and shapes.
  InputArrays inputs;
  // name:index strings for the data outputs.
  std::vector<string> outputs;
  // name strings for the control outputs.
  std::vector<string> control_outputs;

    @TaskAction fun action() { println(inputValue.get()) }
}

// tag::system-properties-task-inputs[]
tasks.register<PrintValue>("printProperty") {
    // Using the Gradle API, provides a lazy Provider<String> wired to a task input
    inputValue = providers.systemProperty("system")
}

  }

  private static void assertMessageDigestHashing(byte[] input, String algorithmName) {
    try {
      MessageDigest digest = MessageDigest.getInstance(algorithmName);
      assertEquals(
          HashCode.fromBytes(digest.digest(input)), ALGORITHMS.get(algorithmName).hashBytes(input));
      for (int bytes = 4; bytes <= digest.getDigestLength(); bytes++) {
        assertEquals(

                  const std::vector<Output>& inputs,
                  std::vector<Output>* outputs) {
  DCHECK(outputs != nullptr);
  DCHECK(outputs->empty());

  // The control dependency is analogous to that in CreateCond().
  Scope body_scope =
      scope.NewSubScope("body").WithControlDependencies(inputs[0]);
  TF_RETURN_IF_ERROR(body(body_scope, inputs, outputs));

  const size_t num_loop_vars = inputs.size();

  // The computation operand can either be:
  //   1) a TPUPartitionedInput Op if the input has a non-resource type;
  //   2) a ReadVariableOp else.
  //
  // Iterate through input arguments to the entry block of
  // tf_device.ClusterFunc. For input ops, look for XlaSharding ops.
  // XlaSharding ops can:
  //   1) Directly follow the input argument if input argument has non-resource
  //      types.