TapeVSpace vspace(ctx);
  std::vector<int64_t> target_tensor_ids = MakeTensorIDList(targets);
  std::vector<int64_t> source_tensor_ids = MakeTensorIDList(sources);
  tensorflow::gtl::FlatSet<int64_t> sources_set(source_tensor_ids.begin(),
                                                source_tensor_ids.end());
  std::unordered_map<int64_t, TapeTensor> sources_that_are_targets;

     * @see Model#isRoot()
     */
    boolean isRootProject();

    /**
     * Gets the root directory of the project, which is the parent directory
     * containing the {@code .mvn} directory or flagged with {@code root="true"}.
     *
     * @return the root directory of the project
     * @throws IllegalStateException if the root directory could not be found
     * @see Session#getRootDirectory()
     */

 * map from keys to <i>nonempty</i> collections of values:
 *
 * <ul>
 *   <li>a → 1, 2
 *   <li>b → 3
 * </ul>
 *
 * ... or as a single "flattened" collection of key-value pairs:
 *
 * <ul>
 *   <li>a → 1
 *   <li>a → 2
 *   <li>b → 3
 * </ul>
 *
 * <p><b>Important:</b> although the first interpretation resembles how most multimaps are

 * map from keys to <i>nonempty</i> collections of values:
 *
 * <ul>
 *   <li>a → 1, 2
 *   <li>b → 3
 * </ul>
 *
 * ... or as a single "flattened" collection of key-value pairs:
 *
 * <ul>
 *   <li>a → 1
 *   <li>a → 2
 *   <li>b → 3
 * </ul>
 *
 * <p><b>Important:</b> although the first interpretation resembles how most multimaps are

// a copy, returns the destination device of the copy. Note that the returned
// device name is not always the device holding the tensor handle's memory. If
// you want the latter, use TFE_TensorHandleBackingDeviceName. This function
// will block till the operation that produces `h` has completed.
TF_CAPI_EXPORT extern const char* TFE_TensorHandleDeviceName(
    TFE_TensorHandle* h, TF_Status* status);

 * and on the Java module path.</p>
 *
 * <h2>Relationship with Java compiler standard location</h2>
 * This enumeration is closely related to the {@link JavaFileManager.Location} enumerations.
 * A difference is that the latter enumerates input and output files, while {@code JavaPathType}
 * enumerates only input dependencies. Another difference is that {@code JavaPathType} contains

If your PR includes any changes to the Gradle Public API, it will cause the binary compatibility check to fail.
The binary compatibility check runs as a part of the broader sanity check.
The latter runs on every PR and is a prerequisite for merging.

If you run the sanity check locally with the `./gradlew sanityCheck`, you can see the binary compatibility error in the output.
It looks like the following:

```

   * 9.0], [9.0, 10.0] (and so on), no matter what the function is. This guarantees that we handle
   * correctly requests of varying weight (permits), /no matter/ what the actual function is - so we
   * can tweak the latter freely. (The only requirement, obviously, is that we can compute its
   * integrals).
   *
   * Note well that if, for this function, we chose a horizontal line, at height of exactly (1/QPS),

    final N node;

    /**
     * The successors left to be visited, or {@code null} if we just added this {@code
     * NodeAndRemainingSuccessors} instance to the stack. In the latter case, we'll compute the
     * successors if we determine that we need them after we've performed the initial processing of
     * the node.
     */
    @Nullable Queue<N> remainingSuccessors;

     * the current count of the element in the multiset, depending on the implementation. Note that
     * in the former case, this method can never return zero, while in the latter, it will return
     * zero if all occurrences of the element were since removed from the multiset.
     *
     * @return the count of the element; never negative
     */
    int getCount();

    /**