std::vector<PartialTensorShape> shapes;
  shapes.reserve(num_shapes);
  for (int i = 0; i < num_shapes; ++i) {
    if (num_dims[i] < 0) {
      shapes.emplace_back();
    } else {
      shapes.emplace_back(absl::Span<const int64_t>(
          reinterpret_cast<const int64_t*>(dims[i]), num_dims[i]));
    }
  }
  desc->node_builder.Attr(attr_name, shapes);
}

    ShapeHandle shape_handle = c.output(i);
    TF_ShapeAndType& shape = output_shapes_result->items[i];
    shape.num_dims = c.Rank(shape_handle);
    if (shape.num_dims == InferenceContext::kUnknownRank) {
      shape.dims = nullptr;
      continue;
    }
    shape.dims = new int64_t[shape.num_dims];
    for (size_t j = 0; j < shape.num_dims; ++j) {
      shape.dims[j] = c.Value(c.Dim(shape_handle, j));
    }
  }

}

TEST(CAPI, ShapeInferenceError) {
  // TF_FinishOperation should fail if the shape of the added operation cannot
  // be inferred.
  TF_Status* status = TF_NewStatus();
  TF_Graph* graph = TF_NewGraph();

  // Create this failure by trying to add two nodes with incompatible shapes
  // (A tensor with shape [2] and a tensor with shape [3] cannot be added).
  const char data[] = {1, 2, 3};
  const int64_t vec2_dims[] = {2};

"$python" -m pip install *.whl $TFCI_PYTHON_VERIFY_PIP_INSTALL_ARGS
if [[ "$TFCI_WHL_IMPORT_TEST_ENABLE" == "1" ]]; then
  "$python" -c 'import tensorflow as tf; t1=tf.constant([1,2,3,4]); t2=tf.constant([5,6,7,8]); print(tf.add(t1,t2).shape)'
  "$python" -c 'import sys; import tensorflow as tf; sys.exit(0 if "keras" in tf.keras.__name__ else 1)'
fi
# Import tf nightly wheel built with numpy2 from PyPI in numpy1 env for testing.

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

    *   Start enforcing input shape assumptions when calling Functional API
        Keras models. This may potentially break some users, in case there is a
        mismatch between the shape used when creating `Input` objects in a
        Functional model, and the shape of the data passed to that model. You
        can fix this mismatch by either calling the model with correctly-shaped

{* ../../docs_src/security/tutorial005_an_py310.py hl[106,108:116] *}

## Verify the `username` and data shape { #verify-the-username-and-data-shape }

We verify that we get a `username`, and extract the scopes.

            EndpointPair.ordered(1, 1),
            EndpointPair.ordered(1, 3),
            EndpointPair.ordered(1, 2))
        .inOrder();
  }

  /**
   * Populates the graph with nodes and edges in a star shape with node `1` in the middle.
   *
   * <p>Note that the edges are added in a shuffled order to properly test the effect of the
   * insertion order.
   */
  private void populateStarShapedGraph() {
    putEdge(2, 1);

            EndpointPair.unordered(1, 1),
            EndpointPair.unordered(1, 3))
        .inOrder();
  }

  /**
   * Populates the graph with nodes and edges in a star shape with node `1` in the middle.
   *
   * <p>Note that the edges are added in a shuffled order to properly test the effect of the
   * insertion order.
   */
  private void populateTShapedGraph() {
    putEdge(2, 1);

You just pass it to `Depends` and **FastAPI** knows how to do the rest.

///

## Share `Annotated` dependencies { #share-annotated-dependencies }

In the examples above, you see that there's a tiny bit of **code duplication**.