system-installed Python packages. \
Instead, an independent Python toolchain is registered, ensuring the right
dependencies are always used. \
See https://github.com/bazelbuild/rules_python/ for more details.

### Specifying the Python version

Note: Only a number of minor Python versions are supported at any given time.

By default, the lowest supported version is used.

// requires its type (e.g. "AddV2") to be set independently.
TF_AbstractOp* TF_NewAbstractOp(TF_ExecutionContext* ctx);
void TF_DeleteAbstractOp(TF_AbstractOp*);

// TODO(srbs): Add APIs for specifying attrs etc.
// `op_type` must outlive `op`.
void TF_AbstractOpSetOpType(TF_AbstractOp* op, const char* const op_type,
                            TF_Status* s);
// `op_name` must outlive `op`.

  Status SetAttrType(const char* const attr_name, DataType value) override {
    if (!op_) {
      return Status(
          absl::StatusCode::kFailedPrecondition,
          "op_type and op_name must be specified before specifying attrs.");
    }
    op_->node_builder.Attr(attr_name, value);
    return absl::OkStatus();
  }
  Status SetAttrShape(const char* attr_name, const int64_t* dims,
                      const int num_dims) override {

        *   Compared to `tf.lookup.experimental.DenseHashTable` this offers
            lower overall memory usage, and a cleaner API. It does not require
            specifying a `delete_key` and `empty_key` that cannot be inserted
            into the table.
    *   Added support for specifying number of subdivisions in all reduce host
        collective. This parallelizes work on CPU and speeds up the collective

    int noutputs, const TF_Output* outputs, const char* const* output_names,
    const TF_FunctionOptions* opts, const char* description, TF_Status* status);

// Similar to TF_GraphToFunction but allows specifying control outputs of the
// function.
//
//  The arguments of TF_GraphToFunction have the same meaning, but the new
//  arguments are as follows:
//