* #queue}, the Executor will complete its tasks, and then restore the interruption. This means
     * that once the Thread returns to the Executor that this Executor composes, the interruption
     * will still be present. If the composed Executor is an ExecutorService, it can respond to
     * shutdown() by returning tasks queued on that Thread after {@link #worker} drains the queue.
     */

   * values by applying the remapping function. The remapping function accepts the map's existing
   * value for that part of the range and the given value. It returns the value to be associated
   * with that part of the map, or it returns {@code null} to clear that part of the map.
   *
   * <p>Any parts of the range not already present in this map are mapped to the specified value,

                                                            TF_Status* status);
// Indicates that the caller will not be using `h` any more.
TF_CAPI_EXPORT extern void TFE_DeleteTensorHandle(TFE_TensorHandle* h);
TF_CAPI_EXPORT extern TF_DataType TFE_TensorHandleDataType(TFE_TensorHandle* h);
// This function will block till the operation that produces `h` has completed.
TF_CAPI_EXPORT extern int TFE_TensorHandleNumDims(TFE_TensorHandle* h,

 * take advantage of the {@linkplain Futures utilities built atop the class}. The way that you will
 * create {@code ListenableFuture} instances depends on how you currently create {@code Future}
 * instances:
 *
 * <ul>
 *   <li>If you receive them from an {@code java.util.concurrent.ExecutorService}, convert that

   * <p><b>Warning:</b> For any given key, every {@code loader} used with it should compute the same
   * value. Otherwise, a call that passes one {@code loader} may return the result of another call
   * with a differently behaving {@code loader}. For example, a call that requests a short timeout
   * for an RPC may wait for a similar call that requests a long timeout, or a call by an

     * @param scopesToCollect The dependency scopes that should be collected, may be {@code null}.
     * @param scopesToResolve The dependency scopes that should be collected and also resolved, may be {@code null}.
     * @param session         The current build session, must not be {@code null}.
     * @return The transitive dependencies of the specified project that match the requested scopes, never {@code null}.

   * for {@code asMapOfRanges().iterator().remove()}.
   *
   * <p>The returned range map will throw an {@link IllegalArgumentException} on an attempt to
   * insert a range not {@linkplain Range#encloses(Range) enclosed} by {@code range}.
   */
  // TODO(cpovirk): Consider documenting that IAE on the various methods that can throw it.

     * AtomicReferenceFieldUpdaterAtomicHelper, which is the default, which means that it's used
     * when we run AbstractFutureTest itself.
     */

    /*
     * We don't test UnsafeAtomicHelper here, since guava-android doesn't provide a way to use it
     * under the JVM. (We could arrange for one if we really wanted, but that will break once the

    return Platform.reserialize(object);
  }

  /**
   * Serializes and deserializes the specified object and verifies that the re-serialized object is
   * equal to the provided object, that the hashcodes are identical, and that the class of the
   * re-serialized object is identical to that of the original.
   *
   * <p><b>GWT warning:</b> Under GWT, this method simply returns its input, as proper GWT

new_app = UnicornMiddleware(app, some_config="rainbow")
```

But FastAPI (actually Starlette) provides a simpler way to do it that makes sure that the internal middlewares handle server errors and custom exception handlers work properly.

For that, you use `app.add_middleware()` (as in the example for CORS).

```Python
from fastapi import FastAPI
from unicorn import UnicornMiddleware

app = FastAPI()