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

namespace tensorflow {

// Builds and returns a `TracingContext` using the default tracing impl.
AbstractContext* BuildFunction(const char* fn_name);

// Creates parameters (placeholders) in the tracing `ctx` using the shape and
// dtype of `inputs`.
Status CreateParamsForInputs(AbstractContext* ctx,
                             absl::Span<AbstractTensorHandle* const> inputs,
                             std::vector<AbstractTensorHandle*>* params);

    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());

    TFE_Op* shape_op = ShapeOp(ctx, hgpu);
    TFE_OpSetDevice(shape_op, gpu_device_name.c_str(), status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
    TFE_TensorHandle* retvals[1];
    int num_retvals = 1;
    TFE_Execute(shape_op, &retvals[0], &num_retvals, status.get());

"$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
# VERY basic check to ensure the [and-cuda] package variant is installable.

*   C API gives access to inferred shapes with `TF_GraphGetTensorNumDims` and
    `TF_GraphGetTensorShape`.
*   Shape functions for core ops have moved to C++ via
    `REGISTER_OP(...).SetShapeFn(...)`. Python shape inference RegisterShape
    calls use the C++ shape functions with `common_shapes.call_cpp_shape_fn`. A
    future release will remove `RegisterShape` from python.

    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());

    TFE_Op* shape_op = ShapeOp(ctx, hgpu);
    TFE_OpSetDevice(shape_op, gpu_device_name.c_str(), status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
    TFE_TensorHandle* retvals[1];
    int num_retvals = 1;
    TFE_Execute(shape_op, &retvals[0], &num_retvals, status.get());

            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);

            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);

                                                           TF_DeleteStatus);

  TF_Tensor* tensor_shape = Int32Tensor({37, 1});
  TF_Operation* shape = Const(tensor_shape, func_graph.get(), s.get(), "shape");
  TF_Operation* random =
      RandomUniform(shape, TF_FLOAT, func_graph.get(), s.get());

  TF_Output outputs[] = {{random, 0}};
  *func = TF_GraphToFunction(func_graph.get(), name,

  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK) << TF_Message(status.get());
  TFE_OpSetAttrType(op.get(), "dtype", TF_FLOAT);
  TFE_OpSetAttrShape(op.get(), "shape", {}, 0, status.get());
  TFE_OpSetAttrString(op.get(), "container", "", 0);
  TFE_OpSetAttrString(op.get(), "shared_name", "", 0);
  TFE_OpSetDevice(op.get(), name, status.get());