@SafeVarargs
  public static <T extends @Nullable Object> Iterator<T> concat(Iterator<? extends T>... inputs) {
    return concatNoDefensiveCopy(Arrays.copyOf(inputs, inputs.length));
  }

  /**
   * Combines multiple iterators into a single iterator. The returned iterator iterates across the
   * elements of each iterator in {@code inputs}. The input iterators are not polled until
   * necessary.
   *

  TFE_Op* concatOp = TFE_NewOp(ctx, "Concat", status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_TensorHandle* inputs[] = {input1, input2};
  TFE_OpAddInput(concatOp, dim, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);
  TFE_OpAddInputList(concatOp, inputs, 2, status);
  CHECK_EQ(TF_OK, TF_GetCode(status)) << TF_Message(status);

                            ConversionPatternRewriter *rewriter) const {
    auto guard = OpBuilder::InsertionGuard(*rewriter);
    auto inputs = branch_func.getFunctionType().getInputs();
    Block *block = rewriter->createBlock(
        &branch_func.getBody(), branch_func.begin(), inputs,
        SmallVector<Location>(inputs.size(), branch_func.getLoc()));

    auto input_shape = block->getArgument(1);
    auto size_diff = block->getArgument(2);

        then: // no changes, should be up-to-date
        assertTransformationsExecuted()

        when:
        run ":app:resolveGreen"

        then: // have seen these inputs before
        assertTransformationsExecuted()
    }

    def "transform with changed project file dependencies content or path are re-executed"() {
        given:
        setupBuildWithSingleStep()

		// content ID. The attempt using the action ID makes sure that
		// if the link inputs don't change, we reuse the cached test
		// result without even rerunning the linker. The attempt using
		// the link output (test binary) content ID makes sure that if
		// we have different link inputs but the same final binary,
		// we still reuse the cached test result.

      return failure();
    }

    // TensorFlow convolution op only has two inputs, while the TFLite one has
    // three, with the bias vector marked as optional. However, TOCO has a
    // dedicated pass, EnsureBiasVectors, to create default bias vectors for all
    // those missing. So we model TFLite convolution op as requiring three
    // inputs to achieve the legalization task of EnsureBiasVector. this

These methods are now final to avoid subclasses overriding these methods and using state that is not initialized.

==== Inputs and outputs are not automatically registered
The Property instances that are returned by these methods are no longer automatically registered as inputs or outputs of the task.

You have to define the task inputs and outputs manually.

2. Using a task as a copy source, i.e., as an argument to `from()`, won't create an automatic task dependency between your task and that copy source.
As such, if you use the `copy()` method as part of a task action, you must explicitly declare all inputs and outputs to get the correct behavior.

        result.assertTasksNotSkipped(":b:producer", ":a:resolve")
        if (incremental || sensitivity == null || sensitivity == PathSensitivity.ABSOLUTE) {
            transformed("b-blue")
        } else {
            // Inputs have been normlaized and results are up-to-date
            transformed()
        }

        when:
        withProjectConfig("b") {
            projectBuildDir = "out"
            outputFileName = "new"

   *
   * <p>It's acceptable for the underlying multimap to contain null keys and null values provided
   * that the transformer is capable of accepting null inputs. The transformed multimap might
   * contain null values if the transformer sometimes gives a null result.
   *
   * <p>The returned multimap is not thread-safe or serializable, even if the underlying multimap