# bazel tests with a special flag "--define=no_tensorflow_py_deps=true", which
# drops all the bazel dependencies for each py_test; this makes all the tests
# use the wheel's TensorFlow installation instead of the one made available
# through bazel. This must be done in a different root directory, //bazel_pip/...,
# because "import tensorflow" run from the root directory would instead import
# the folder instead of the venv package.
# 

+    __attribute_copy__ (name);
 
 /* This comes between the return type and function name in
    a function definition to make that definition weak.  */
@@ -125,14 +131,16 @@
    If weak aliases are not available, this defines a strong alias.  */
 # define weak_alias(name, aliasname) _weak_alias (name, aliasname)
 # define _weak_alias(name, aliasname) \
-  extern __typeof (name) aliasname __attribute__ ((weak, alias (#name)));

// Destroy an options object.
TF_CAPI_EXPORT extern void TFE_DeleteContextOptions(TFE_ContextOptions*);

// "Context" under which operations/functions are executed. It encapsulates
// things like the available devices, resource manager etc.
// TFE_Context must outlive all tensor handles created using it. In other
// words, TFE_DeleteContext() must be called after all tensor handles have
// been deleted (with TFE_DeleteTensorHandle).

Note: Updating the
[rules-python](https://github.com/bazelbuild/rules_python/releases) version may
be required before going through the steps below. This is due to the new Python
versions becoming available through `rules-python`. \
See
[here](https://github.com/tensorflow/tensorflow/commit/f91457f258fdd78f693044a57efa63a38335d1de),
and

  _ll.load_pluggable_device_library(
      _os.getenv("TF_PLUGGABLE_DEVICE_LIBRARY_PATH")
  )

# Delete modules that should be hidden from dir().
# Don't fail if these modules are not available.
# For e.g. this file will be originally placed under tensorflow/_api/v1 which
# does not have "python", "core" directories. Then, it will be copied
# to tensorflow/ which does have these two directories.

try:

### 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

                         absl::Span<AbstractTensorHandle*> grad_inputs) = 0;
  virtual ~GradientFunction() {}
};

// Metadata from the forward operation that is made available to the
// gradient registerer to instantiate a GradientFunction.
struct ForwardOperation {
 public:
  string op_name;
  std::vector<AbstractTensorHandle*> inputs;
  std::vector<AbstractTensorHandle*> outputs;

        APIs will be removed as well.
    *   `tf.keras.callbacks.experimental.BackupAndRestore` is now available as
        `tf.keras.callbacks.BackupAndRestore`. The experimental endpoint is
        deprecated and will be removed in a future release.
    *   `tf.keras.experimental.SidecarEvaluator` is now available as
        `tf.keras.utils.SidecarEvaluator`. The experimental endpoint is

  // Runtime. This is necessary to decouple runtime-dependent
  // code that is layered on top of the runtime.
  virtual bool UsesTFRT() = 0;

  // List attributes of available devices
  virtual void ListDevices(std::vector<DeviceAttributes>* devices) = 0;

  // Add `devices` into context's device manager. Context's device manager
  // will take ownership and maintain devices' lifetime.

  ~GraphOperation() override {}

 private:
  friend class GraphContext;  // For access to op_.
  TF_Graph* g_;
  std::unique_ptr<TF_OperationDescription> op_;
  // Hold `op_type` and `op_name` till both are available since we need both
  // to build a graph operation.
  string op_type_;
  const char* op_name_ = nullptr;
  // TODO(srbs): Use this.
  string device_name_;
};