type: DT_FLOAT
    }
  }
}
node {
  name: "output_0"
  op: "Identity"
  input: "conv_net_2d_1/Relu"
  attr {
    key: "T"
    value {
      type: DT_FLOAT
    }
  }
}
library {
}

# CHECK: 'main' inputs:
# CHECK-NEXT: name: 'input'
# CHECK-NEXT: 'main' outputs:

            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));
  }

        outputs.fileProperties*.propertyFiles*.files.flatten() == [file("a")]
        outputs.fileProperties*.outputFile == [file("a")]
        outputs.fileProperties*.outputType == [FILE]
    }

    def "can register output dir"() {
        when:
        outputs.dir("a")
        then:
        outputs.files.files == files('a')
        outputs.fileProperties*.propertyName == ['$1']

     *     public void transform(TransformOutputs outputs) {
     *         outputs.dir(getInputArtifact().get().getAsFile());
     *         outputs.dir(new File(getInputArtifact().get().getAsFile(), "sub-dir"));
     *     }
     * }
     * </pre>
     *
     * <p>
     *     For a <strong>relative path</strong>, Gradle creates an output directory into which the {@link TransformAction} must place its output files.
     *     Example:
     * </p>

      )

      inputs = {'input_vocabs': input_vocabs_placeholder}
      outputs = {
          'lookup': lookup_tensor,
          'output': output_tensor,
      }

      self._save_tf1_model(
          sess,
          output_path,
          signature_def_key,
          tags,
          inputs=inputs,
          outputs=outputs,
          init_op=lookup_ops.tables_initializer(),

            output->DebugString(100));
}

REGISTER_OP("AllocateOutputOp0").Output("output1: float");

TEST_F(DeviceKernelOpTest, TestAllocateEmptyOutput) {
  auto my_compute_func = [](void* kernel, TF_OpKernelContext* ctx) {
    TF_Status* s = TF_NewStatus();
    // Allocate empty output
    int64_t dim = 0;
    TF_Tensor* output = TF_AllocateOutput(

        }

        and:
        outputs.stdOut.contains("BEFORE OUT")
        stripped(outputs.stdOut).contains(result1.output)
        stripped(outputs.stdOut).contains(result2.output)
        outputs.stdOut.contains("AFTER OUT")

        and:
        outputs.stdErr.contains("BEFORE ERR")
        stripped(outputs.stdOut).contains(result1.error)
        stripped(outputs.stdOut).contains(result2.error)

      [](const Scope& s, const std::vector<Output>& inputs, Output* output) {
        *output = ops::Less(s, inputs[0], 10);
        return s.status();
      },
      [](const Scope& s, const std::vector<Output>& inputs,
         std::vector<Output>* outputs) {
        outputs->push_back(ops::AddN(s, {inputs[0], inputs[1]}));
        outputs->push_back(ops::AddN(s, {inputs[1], 1}));
        outputs->push_back(inputs[2]);

        %0 = tf_executor.graph {
          %output_0, %control_0 = tf_executor.island wraps "tf.Const"() <{value = dense<1.0> : tensor<1x2xf32>}> : () -> tensor<1x2xf32>