// Removes a function from the context. Once removed, you can no longer
// TFE_Execute it or TFE_Execute any TFE_Op which has it as an attribute or any
// other function which calls it as an attribute.
TF_CAPI_EXPORT extern void TFE_ContextRemoveFunction(TFE_Context* ctx,
                                                     const char* name,

// A TF_AbstractOp is the metadata we need to execute an operation. E.g. this
// could contain the op type and other attributes.
typedef struct TF_AbstractOp TF_AbstractOp;

// Stores a function representation that can be used for execution or for
// setting functional attributes of other composite ops e.g. control flow.
typedef struct TF_AbstractFunction TF_AbstractFunction;

  op->node.AddAttr(attr_name, attr_val);
  RecordMutation(graph, *op, "setting attribute");
}

void TF_ClearAttr(TF_Graph* graph, TF_Operation* op, const char* attr_name,
                  TF_Status* status) {
  using tensorflow::RecordMutation;
  mutex_lock l(graph->mu);
  op->node.ClearAttr(attr_name);
  RecordMutation(graph, *op, "clearing attribute");
}

void TF_SetFullType(TF_Graph* graph, TF_Operation* op,

    // TODO(srbs): Manage refcount of ForwardOperation's inputs/outputs.
    forward_op_->outputs.push_back(retvals[i]);
  }
  // TODO(b/166669239): This is needed to support AttrBuilder::Get for string
  // attributes. Number type attrs and DataType attrs work fine without this.
  // Consider getting rid of this and making the behavior between number types
  // and string consistent.
  forward_op_->attrs.BuildNodeDef();

  static bool classof(const AbstractTensorHandle* ptr) {
    return ptr->getKind() == kGraph || ptr->getKind() == kMlir;
  }
};

// An abstract operation describes an operation by its type, name, and
// attributes. It can be "executed" by the context with some input tensors.
// It is allowed to reusing the same abstract operation for multiple execution
// on a given context, with the same or different input tensors.

  // 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

  EXPECT_EQ(string("Invalid FunctionDef given to TF_FunctionImportFunctionDef"),
            string(TF_Message(s_)));
}

TEST_F(CApiFunctionTest, Attribute) {
  DefineFunction(func_name_, &func_);

  // Get non existent attribute
  TF_Buffer* attr_buf = TF_NewBuffer();
  TF_FunctionGetAttrValueProto(func_, "foo_attr", attr_buf, s_);
  EXPECT_EQ(TF_INVALID_ARGUMENT, TF_GetCode(s_));

# https://github.com/tensorflow/tensorflow/blob/cd67f4f3723f9165aabedd0171aaadc6290636e5/tensorflow/tensorflow.bzl#L396-L425
# And is usable in the "deps" attribute instead of the "srcs" attribute
# as a workaround for https://github.com/tensorflow/tensorflow/issues/34117
cc_import(
    name = "libtensorflow_framework_import_lib",
    shared_library = select({

    *   `Estimator.export_savedmodel` has been renamed to `export_saved_model`.
    *   When saving to SavedModel, Estimators will strip default op attributes.
        This is almost always the correct behavior, as it is more forwards
        compatible, but if you require that default attributes to be saved with
        the model, please use `tf.compat.v1.Estimator`.
    *   Feature Columns have been upgraded to be more Eager-friendly and to work

// Load the weight of a variable
TF_CAPI_EXPORT extern TF_Tensor* TF_CheckpointReaderGetTensor(
    TF_CheckpointReader* reader, const char* name, TF_Status* status);

// TF_NewAttrBuilder() returns an object that you can set attributes on as
// though it were an op. This allows querying properties of that op for
// type-checking purposes like if the op will run on a particular device type.
typedef struct TF_AttrBuilder TF_AttrBuilder;