if (input_arg.number_attr().empty()) {
    status->status = tensorflow::errors::NotFound(
        op_name, " does not have number_attr() defined.");
    return nullptr;
  }

  // The returned string is owned by OpRegistry, so liveness is not a concern.
  return input_arg.number_attr().c_str();
}

int TF_OpIsStateful(const char* op_type, TF_Status* status) {

                                        std::move(mixed_handles), status.get());
  ASSERT_TRUE(TF_GetCode(status.get()) == TF_OK);
  // Can't take the shape of a parallel tensor with varying numbers of axes, but
  // running operations on them is OK.
  TF_ASSERT_OK(unknown_length_vector->Shape(&shape));
  EXPECT_THAT(*shape, ElementsAre(-1));
  std::unique_ptr<TFE_Op, decltype(&TFE_DeleteOp)> size_op(

    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();
  std::unique_ptr<GradientFunction> gradient_fn;

// reference counts drop to zero.
void ParallelTensorDeallocator(void* data) {
  delete reinterpret_cast<ParallelTensor*>(data);
}

// Used as an argument to TFE_NewCustomDeviceTensorHandle, for computing the
// number of dimensions of a parallel tensor.
int ParallelTensorNumDims(void* data, TF_Status* status) {
  const std::vector<int64_t>* shape;
  Status s = reinterpret_cast<ParallelTensor*>(data)->Shape(&shape);
  if (!s.ok()) {

        first_bad_status.reset(TF_NewStatus());
        TF_SetStatus(first_bad_status.get(), TF_INTERNAL,
                     "Parallel ops produced different numbers of tensors.");
      }
    }
  }
  if (first_bad_status != nullptr) {
    TF_SetStatus(status, TF_GetCode(first_bad_status.get()),
                 TF_Message(first_bad_status.get()));
    return result;

        context.get(), components, device_name, status.get());
    ASSERT_TRUE(TF_GetCode(status.get()) == TF_INVALID_ARGUMENT)
        << TF_Message(status.get());
  }

  {
    // Try to extract the wrong number of components from a parallel tensor
    std::array<TFE_TensorHandle*, 1> correct_components{value_one.get()};
    TensorHandlePtr combined_value = CreatePerDeviceValues(

  if (scheme == nullptr)
    return errors::InvalidArgument(
        "Attempted to register filesystem with `nullptr` URI scheme");
  return OkStatus();
}

// Checks if the plugin and core ABI numbers match.
//
// If the numbers don't match, plugin cannot be loaded.
static Status CheckABI(int pluginABI, int coreABI, StringPiece where) {
  if (pluginABI != coreABI)
    return errors::FailedPrecondition(

#include "tensorflow/c/experimental/filesystem/plugins/gcs/gcs_filesystem.h"

#include <stdlib.h>
#include <string.h>

#include <variant>

#include "absl/strings/numbers.h"
#include "absl/strings/str_cat.h"
#include "absl/types/variant.h"
#include "google/cloud/storage/client.h"
#include "tensorflow/c/env.h"
#include "tensorflow/c/experimental/filesystem/plugins/gcs/gcs_helper.h"

    return tensorflow::ops::Identity(ctx, input, output, name);
  } else if (!DataTypeIsComplex(BaseType(dtype)) &&
             BaseType(dtype) != DT_VARIANT) {
    return errors::InvalidArgument(
        "Expected numeric or variant tensor, got dtype ", dtype);
  }
  return tensorflow::ops::Conj(ctx, input, output, name);
}

class AddGradientFunction : public GradientFunction {
 public:
  Status Compute(AbstractContext* ctx,

    return errors::Unimplemented(
        "Filesystem does not support SetConfiguration()");
  }
  if (values.empty()) {
    return errors::InvalidArgument(
        "SetConfiguration() needs number of values > 0");
  }

  TF_Filesystem_Option option;
  memset(&option, 0, sizeof(option));
  option.name = const_cast<char*>(name.c_str());
  TF_Filesystem_Option_Value option_value;