bool found = false;
  for (Node* node : graph->nodes()) {
    if (CanCreateXlaKernel(node->def())) {
      EXPECT_FALSE(found);
      found = true;
      std::vector<int32> hostmem_attr;
      EXPECT_TRUE(TryGetNodeAttr(node->def(), "_input_hostmem", &hostmem_attr));
      EXPECT_EQ(hostmem_attr.size(), 1);
      EXPECT_EQ(hostmem_attr[0], 1);
    }
  }

)

load("@rules_jvm_external//:defs.bzl", "maven_install")

maven_install(
    artifacts = [

    ],
    repositories = [
        "https://repo.maven.apache.org/maven2/"
    ],
)
"""
    }

    String generateBuildFile(Integer subProjectNumber, DependencyTree dependencyTree) {
        def isRoot = subProjectNumber == null

#include <vector>

#include "absl/memory/memory.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/str_format.h"
#include "tensorflow/compiler/jit/compilability_check_util.h"
#include "tensorflow/compiler/jit/defs.h"
#include "tensorflow/compiler/jit/flags.h"
#include "tensorflow/compiler/jit/kernels/xla_ops.h"
#include "tensorflow/compiler/tf2xla/const_analysis.h"
#include "tensorflow/compiler/tf2xla/mlir_bridge_pass.h"

// Registers the XLA_CPU device, which is an XlaDevice instantiation that runs
// operators using XLA via the XLA "Host" (CPU) backend.

#include <array>

#include "absl/memory/memory.h"
#include "tensorflow/compiler/jit/defs.h"
#include "tensorflow/compiler/jit/flags.h"
#include "tensorflow/compiler/jit/kernels/xla_ops.h"
#include "tensorflow/compiler/jit/xla_compile_on_demand_op.h"
#include "tensorflow/compiler/jit/xla_device.h"

   parameter of the `python_register_multi_toolchains` function.

3) In the `BUILD.bazel` file, add a load statement for the new version, e.g.

   ```
      load("@python//3.11:defs.bzl",
           compile_pip_requirements_3_11 = "compile_pip_requirements")
   ```

   Add a new entry for the loaded `compile_pip_requirements`, e.g.

   ```
      compile_pip_requirements_3_11(

				// If the cancel variable is defined outside function scope,
				// do not analyze it.
				if funcScope.Contains(v.Pos()) {
					cancelvars[v] = stmt
				}
			} else if v, ok := pass.TypesInfo.Defs[id].(*types.Var); ok {
				cancelvars[v] = stmt
			}
		}
		return true
	})

	if len(cancelvars) == 0 {
		return // no need to inspect CFG
	}

	// Obtain the CFG.

#include <array>
#include <set>

#include "absl/memory/memory.h"
#include "absl/strings/numbers.h"
#include "absl/strings/str_split.h"
#include "tensorflow/compiler/jit/defs.h"
#include "tensorflow/compiler/jit/flags.h"
#include "tensorflow/compiler/jit/kernels/xla_ops.h"
#include "tensorflow/compiler/jit/xla_device.h"
#include "tensorflow/compiler/jit/xla_device_ops.h"

  // names to their corresponding nodes at load time in order to look them up
  // in constant time.
  tensorflow::StatusOr<const tensorflow::NodeDef*> GetGraphDefNode(
      std::string name);

  // Returns a list of function defs in the SavedModel.
  const protobuf::RepeatedPtrField<FunctionDef>& GetFunctionDefs();

  // Returns a BundleReader for reading variable values.
  //
  // This TFPackage retains ownership of the underlying reader.

#include "tensorflow/compiler/jit/cluster_scoping_pass.h"

#include "absl/algorithm/container.h"
#include "absl/container/flat_hash_set.h"
#include "absl/strings/str_cat.h"
#include "tensorflow/compiler/jit/defs.h"
#include "tensorflow/compiler/jit/xla_cluster_util.h"
#include "tensorflow/core/framework/node_def_util.h"
#include "tensorflow/core/graph/algorithm.h"

namespace tensorflow {
namespace {

// flag to disable syntactic object resolution (which also saves CPU
// and memory), and instead use the type checker [go/types] if object
// resolution is desired. See the Defs, Uses, and Implicits fields of
// the [types.Info] struct for details.
type Object struct {
	Kind ObjKind
	Name string // declared name