@app.get("/items/{item_id}")
def get_item(item_id: str, username: str = Depends(get_username)):
    if item_id == "portal-gun":
        raise InternalError(
            f"The portal gun is too dangerous to be owned by {username}"
        )
    if item_id != "plumbus":
        raise HTTPException(
            status_code=404, detail="Item not found, there's only a plumbus here"
        )

@app.get("/items/{item_id}")
def get_item(item_id: str, username: Annotated[str, Depends(get_username)]):
    if item_id == "portal-gun":
        raise InternalError(
            f"The portal gun is too dangerous to be owned by {username}"
        )
    if item_id != "plumbus":
        raise HTTPException(
            status_code=404, detail="Item not found, there's only a plumbus here"
        )

@app.get("/items/{item_id}")
def get_item(item_id: str, username: str = Depends(get_username)):
    if item_id == "portal-gun":
        raise InternalError(
            f"The portal gun is too dangerous to be owned by {username}"
        )
    if item_id != "plumbus":
        raise HTTPException(
            status_code=404, detail="Item not found, there's only a plumbus here"
        )

@app.get("/items/{item_id}")
def get_item(item_id: str, username: Annotated[str, Depends(get_username)]):
    if item_id == "portal-gun":
        raise InternalError(
            f"The portal gun is too dangerous to be owned by {username}"
        )
    if item_id != "plumbus":
        raise HTTPException(
            status_code=404, detail="Item not found, there's only a plumbus here"
        )

                 std::unique_ptr<tensorflow::Tensor>* out_tensor,
                 TF_Status* out_status) const;

 private:
  // Uses "v2_reader_" to build "var name -> shape" and "var name -> data type"
  // maps; both owned by caller.
  // REQUIRES: "v2_reader_ != nullptr && v2_reader_.status().ok()".
  std::pair<std::unique_ptr<TensorSliceReader::VarToShapeMap>,
            std::unique_ptr<TensorSliceReader::VarToDataTypeMap> >
  BuildV2VarMaps();

};

struct TF_ImportGraphDefOptions {
  tensorflow::ImportGraphDefOptions opts;

  // Backing memory for TensorId fields in opts.
  // TODO(skyewm): it'd be better if ImportGraphDefOptions owned this.
  std::vector<std::string> tensor_id_data;
};

struct TF_ImportGraphDefResults {
  std::vector<TF_Output> return_tensors;
  std::vector<TF_Operation*> return_nodes;

 - ConnectionAcquired
 - ... Standard Events...
 - ResponseBodyEnd _(0 bytes)_
 - **CacheHit**
 - ConnectionReleased
 - CallEnd

## Cache directory

The cache directory must be exclusively owned by a single instance.

Deleting the cache when it is no longer needed can be done.  However this may delete the purpose of the cache
which is designed to persist between app restarts.

```kotlin
cache.delete()
```

    string key(v2_reader_->key());
    (*var_to_shape_map)[key] = TensorShape(entry.shape());
    (*var_to_data_type_map)[key] = DataType(entry.dtype());
  }
  // The returned pointers are owned by the caller.
  return std::make_pair(std::move(var_to_shape_map),
                        std::move(var_to_data_type_map));
}

}  // namespace checkpoint

An architecture module is responsible for providing a coherent set of features and:

- Provides a set of APIs and services for use from outside the module.
- Has a private implementation.
- Is owned by a single team. A team may own multiple architecture modules.

The modules are arranged into several different "Gradle platforms".
A Gradle platform is a logical distribution that provides support for a specific kind of automation.

   *     invocation conversion.
   * @throws InvocationTargetException if the underlying method or constructor throws an exception.
   */
  // All subclasses are owned by us and we'll make sure to get the R type right, including nullness.
  @SuppressWarnings({"unchecked", "nullness"})
  @CanIgnoreReturnValue
  public final @Nullable R invoke(@Nullable T receiver, @Nullable Object... args)