// .backing_device of shape is CPU since the tensor is backed by CPU
    backing_device_name =
        TFE_TensorHandleBackingDeviceName(retvals[0], status.get());
    ASSERT_EQ(TF_OK, TF_GetCode(status.get())) << TF_Message(status.get());
    ASSERT_TRUE(absl::StrContains(backing_device_name, "CPU:0"))
        << backing_device_name;

    TFE_DeleteOp(shape_op);

#include "tsl/platform/errors.h"

// TODO(b/139060984): After all filesystems are converted, all calls to
// methods from `FileSystem` will have to be replaced to calls to private
// methods here, as part of making this class a singleton and the only way to
// register/use filesystems.

namespace tensorflow {

using UniquePtrTo_TF_Status =
    ::std::unique_ptr<TF_Status, decltype(&TF_DeleteStatus)>;

TFCI_BAZEL_TARGET_SELECTING_CONFIG_PREFIX=linux_arm64
# Note: this is not set to "--cpu", because that changes the package name
# to tensorflow_cpu. These ARM builds are supposed to have the name "tensorflow"
# despite lacking Nvidia CUDA support.
TFCI_BUILD_PIP_PACKAGE_ARGS="--repo_env=WHEEL_NAME=tensorflow"
TFCI_DOCKER_ENABLE=1
TFCI_DOCKER_IMAGE=gcr.io/tensorflow-sigs/build-arm64:tf-2-16-multi-python
TFCI_DOCKER_PULL_ENABLE=1

  awk '/Streaming build results to/ {print $NF}' "$TFCI_OUTPUT_DIR/script.log" | uniq
}

# Print out any ResultStore URLs for Bazel invocations' results.
# Each failed target there will have its own representation, making failures
# easier to find and read.
function resultstore_extract {
  local \
    XML_PATH="$TFCI_OUTPUT_DIR/Bazel_Test_and_Build_Results/sponge_log.xml"

  python3 \

#
# setup.packages.sh: Given a list of Ubuntu packages, install them and clean up.
# Usage: setup.packages.sh <package_list.txt>
set -e

# Prevent apt install tzinfo from asking our location (assumes UTC)
export DEBIAN_FRONTEND=noninteractive

apt-get update
# Remove commented lines and blank lines
apt-get install -y --no-install-recommends $(sed -e '/^\s*#.*$/d' -e '/^\s*$/d' "$1" | sort -u)

#
# setup.python.sh: Install a specific Python version and packages for it.
# Usage: setup.python.sh <pyversion> <requirements.txt>

# Sets up custom apt sources for our TF images.

# Prevent apt install tzinfo from asking our location (assumes UTC)
export DEBIAN_FRONTEND=noninteractive

# Set up shared custom sources
apt-get update
apt-get install -y gnupg ca-certificates

Readme.md would be the same file on Windows. In this test, you would get a
warning for "readme.md" because it makes everything lowercase. There are
repeats of these filename(s) with different casing:
EOF
    find . | tr '[A-Z]' '[a-z]' | sort | uniq -d | tee $BATS_FILE_TMPDIR/repeats
    [[ ! -s $BATS_FILE_TMPDIR/repeats ]]
}

# It's unclear why, but running this on //tensorflow/... is faster than running

# Download glibc's shared and development libraries based on the value of the
# `VERSION` parameter.
# Note: 'Templatizing' this and the other conditional branches would require
# defining several variables (version, os, path) making it difficult to maintain
# and extend for future modifications.
mkdir -p glibc-src
mkdir -p glibc-build
cd glibc-src

  }
  // TODO(b/166669239): This is needed to support AttrBuilder::Get for string
  // attributes. Number type attrs and DataType attrs work fine without this.
  // Consider getting rid of this and making the behavior between number types
  // and string consistent.
  forward_op_->attrs.BuildNodeDef();
  std::unique_ptr<GradientFunction> gradient_fn;
  TF_RETURN_IF_ERROR(registry.Lookup(*forward_op_, &gradient_fn));

### Saved graphs and checkpoints

When loading untrusted serialized computation graphs (in form of a `GraphDef`,
`SavedModel`, or equivalent on-disk format), the set of computation primitives
available to TensorFlow is powerful enough that you should assume that the
TensorFlow process effectively executes arbitrary code.

The risk of loading untrusted checkpoints depends on the code or graph that you