"same location"        | null           | "file"     | "output.txt"               | "output.txt"
        "consuming ancestor"   | null           | "file"     | "build/dir/sub/output.txt" | "build/dir"
        "consuming descendant" | 'build/dir'    | "dir"      | "build/dir/sub/output.txt" | "build/dir/sub/output.txt"
    }

  const TF_TensorSpec* tensor_spec_out =
      TF_SignatureDefParamTensorSpec(param_out);
  const TF_Shape* shape_out = TF_TensorSpecShape(tensor_spec_out);

  // Output "output_0" is a scalar, float32 tensor
  EXPECT_EQ("output_0", std::string(TF_SignatureDefParamName(param_out)));
  EXPECT_EQ(TF_FLOAT, TF_TensorSpecDataType(tensor_spec_out));
  EXPECT_EQ(0, TF_ShapeDims(shape_out));

            file("build.gradle") << """
                task foo {
                    inputs.dir('src').ignoreEmptyDirectories(false).withPropertyName('src')
                    outputs.file('output.txt')
                    doLast {
                        file('output.txt') << 'do stuff'
                    }
                }
            """
            dir("src") {
                file("A.txt") << "fooA"

                            const std::vector<Output>& outputs,
                            const std::vector<Output>& inputs,
                            std::vector<Output>* grad_outputs) {
  std::vector<Output> grad_inputs;
  grad_inputs.reserve(outputs.size());
  for (const Output& output : outputs) {
    grad_inputs.emplace_back(ops::OnesLike(scope, output));
  }

  func.func @serving_default(%arg0: tensor<1024x1024xf32> {tf_saved_model.index_path = ["input_tensor"]}) -> (tensor<1024x3xf32> {tf_saved_model.index_path = ["output1"]}, tensor<1024x3xf32> {tf_saved_model.index_path = ["output2"]}) attributes {tf.entry_function = {control_outputs = "", inputs = "serving_default_input_tensor:0", outputs = "PartitionedCall:0"}, tf_saved_model.exported_names = ["serving_default"]} {
    %0 = stablehlo.constant dense<1.000000e+03> : tensor<1024x3xf32>

                @Override
                void transform(TransformOutputs outputs) {
                    // Simulate transform leaving file open
                    def output = outputs.file("output.txt")
                    output.createNewFile()
                    TestState.fileKeptOpen = output.newOutputStream()
                    TestState.fileKeptOpen << "output"
                    TestState.fileKeptOpen.flush()
                }

    }

    @Override
    public void transform(TransformOutputs outputs) {
        // We simulate fan-in behaviour:
        // We expect that a transform before this one outputs three artifacts: 1. analysis metadata, 2. the original file and 3. instrumentation marker file.
        // So if the input is analysis metadata we merge it and output it, otherwise it's original artifact, and we output that.
        File input = getInput().get().getAsFile();

      location, concat_dim_type, concat_dimension_attr);

  // Correctly set output shapes of concat op output if output shape is
  // statically known. Since the shape of TPUExecute op must be the same
  // across logical devices, we refer to the shape of 0th logical device
  // computation output.
  mlir::Type output_type;
  auto input_type = mlir::cast<mlir::TensorType>(inputs[0].getType());

[[sec:task_output_caching_details]]
== Cacheable tasks

Since a task describes all of its inputs and outputs, Gradle can compute a _build cache key_ that uniquely defines the task's outputs based on its inputs.
That build cache key is used to request previous outputs from a build cache or store new outputs in the build cache.