std::vector<TFE_TensorHandle*> outputs(*num_retvals);
    TF_Status status;
    device_.execute(tensorflow::wrap(op), num_retvals, outputs.data(), &status,
                    info_);
    if (status.status.ok()) {
      for (int i = 0; i < *num_retvals; ++i) {
        retvals[i] = tensorflow::unwrap(outputs[i]);
        retvals[i]->Ref();
        TFE_DeleteTensorHandle(outputs[i]);
      }
    }

          path: |
            ~/.android/avd/*
            ~/.android/adb*
          key: avd-${{ matrix.api-level }}

      - name: Create AVD and generate snapshot for caching
        if: steps.avd-cache.outputs.cache-hit != 'true'
        uses: reactivecircus/android-emulator-runner@v2
        with:
          api-level: ${{ matrix.api-level }}
          arch: ${{ matrix.api-level == '34' && 'x86_64' || 'x86' }}

                                                    <td>${f:h(data.inputs)}</td>
                                                    <td>${f:h(data.outputs)}</td>
                                                </tr>
                                            </c:forEach>
                                            </tbody>
                                        </table>

            attributes.put("snippetsPath", "snippets");
            // Make sure the 'raw' location of the samples is available in all AsciidoctorTasks to access files with expected outputs in the 'tests' folder for inclusion in READMEs
            attributes.put("samplesPath", extension.getUserManual().getStagingRoot().dir("raw/samples").get().getAsFile());
            task.attributes(attributes);
        });

//
// Instead, we rely directly on the procfs.
// This rules out two possible methods:
// * use crictl to inspect the pod; this returns the bind-mounted network namespace file.
// * /var/lib/cni/results shows the outputs of CNI plugins; this containers the bind-mounted network namespace file.
//
// Instead, we traverse the procfs. Comments on this method are inline.
func getPodNetNs(pod *corev1.Pod) (string, error) {

				break
			}
		} else if p.arch.Family == sys.Loong64 {
			if arch.IsLoong64RDTIME(op) {
				// The Loong64 RDTIME family of instructions is a bit special,
				// in that both its register operands are outputs
				prog.To = a[0]
				if a[1].Type != obj.TYPE_REG {
					p.errorf("invalid addressing modes for 2nd operand to %s instruction, must be register", op)
					return
				}
				prog.RegTo2 = a[1].Reg
				break

        final EditBody body = new EditBody();
        body.id = entity.getId();
        body.dictId = dictId;
        body.inputs = entity.getInputsValue();
        body.outputs = entity.getOutputsValue();
        return body;
    }

        "$@"

# Stop when "xargs" is not available.
if ! command -v xargs >/dev/null 2>&1
then
    die "xargs is not available"
fi

# Use "xargs" to parse quoted args.
#
# With -n1 it outputs one arg per line, with the quotes and backslashes removed.
#
# In Bash we could simply go:
#
#   readarray ARGS < <( xargs -n1 <<<"$var" ) &&
#   set -- "${ARGS[@]}" "$@"
#

To regenerate the manifests, run:

```bash
$ make copy-templates update-golden
```

### Step 5. Create a PR using outputs from Steps 1 to 4

Your PR should pass all the checks if you followed these steps.

## Value deprecation

FastAPI uses a **unique ID** for each *path operation*, it is used for the **operation ID** and also for the names of any needed custom models, for requests or responses.

You can customize that function. It takes an `APIRoute` and outputs a string.

For example, here it is using the first tag (you will probably have only one tag) and the *path operation* name (the function name).