//   If file exists, truncate its size to 0.
  int dst_fd = open(dst, O_WRONLY | O_CREAT | O_TRUNC, open_mode);
  if (dst_fd < 0) {
    close(src_fd);
    return -1;
  }

  // Both files have been opened, do the transfer.
  // Since errno would be overridden by `close` below, save it here.
  int error_code = 0;
  if (CopyFileContents(dst_fd, src_fd, size) < 0) error_code = errno;

  close(src_fd);
  close(dst_fd);

  std::vector<const Node*> control_output_nodes;
  control_output_nodes.reserve(ncontrol_outputs);
  for (int i = 0; i < ncontrol_outputs; ++i) {
    control_output_nodes.push_back(&control_outputs[i]->node);
  }

  // Do the actual function creation.
  DCHECK(append_hash_to_fn_name <= 1);
  tensorflow::FunctionDef fdef;
  status->status = tensorflow::GraphToFunctionDef(
      fn_body->graph, fn_name, append_hash_to_fn_name != 0,

    TF_SetStatus(status, TF_OK, "");
}

static int64_t Tell(const TF_WritableFile* file, TF_Status* status) {
  auto posix_file = static_cast<PosixFile*>(file->plugin_file);

  // POSIX's `ftell` returns `long`, do a manual cast.
  int64_t position = int64_t{ftell(posix_file->handle)};
  if (position < 0)
    TF_SetStatusFromIOError(status, errno, posix_file->filename);
  else
    TF_SetStatus(status, TF_OK, "");

        // but need special handling for gradients (gradient of copy-on is not
        // just copy-off but includes a sum) and consideration of performance.
        //
        // TODO(allenl): There may be smarter ways to do this copy in some
        // cases, i.e. with a collective broadcast. We'll need to be careful
        // about things that are taken as inputs on the host or on their
        // existing device (for multi-device functions).

        TF_RETURN_IF_ERROR(unwrap(tensors_[0].get())->Shape(&first_shape));
        return errors::Unimplemented(absl::StrCat(
            "Computing the shape of a ParallelTensor when the components do "
            "not all have the same rank is not supported. One tensor had "
            "shape ",
            first_shape.DebugString(), " and another had shape ",
            component_shape.DebugString()));

  EXPECT_EQ(TF_GetCode(status.get()), TF_INVALID_ARGUMENT);
  EXPECT_THAT(TF_Message(status.get()), HasSubstr("assertion failed"));

  // Note that future collectives with the same context do not work at the
  // moment; once canceled, the collective executor requires the program to be
  // restarted / context to be reset.
}

TEST(PARALLEL_DEVICE_LIB, TestDifferentShapes) {

  }
  return status.status;
}

TEST(RamFileBlockCacheTest, IsCacheEnabled) {
  auto fetcher = [](const string& filename, size_t offset, size_t n,
                    char* buffer, TF_Status* status) -> int64_t {
    // Do nothing.
    TF_SetStatus(status, TF_OK, "");
    return 0;
  };
  tf_gcs_filesystem::RamFileBlockCache cache1(0, 0, 0, fetcher);
  tf_gcs_filesystem::RamFileBlockCache cache2(16, 0, 0, fetcher);

  new_entry->lra_iterator = lra_list_.begin();
  new_entry->timestamp = timer_seconds_();
  block_map_.emplace(std::make_pair(key, new_entry));
  return new_entry;
}

// Remove blocks from the cache until we do not exceed our maximum size.
void RamFileBlockCache::Trim() {
  while (!lru_list_.empty() && cache_size_ > max_bytes_) {
    RemoveBlock(block_map_.find(lru_list_.back()));
  }
}

  TF_FilesystemPluginInfo info;
  memset(&info, 0, sizeof(info));
  auto TF_InitPlugin =
      reinterpret_cast<int (*)(TF_FilesystemPluginInfo*)>(dso_symbol);
  TF_InitPlugin(&info);

  // Step 4: Do the actual registration
  return filesystem_registration::RegisterFilesystemPluginImpl(&info);
}

#include "tensorflow/core/protobuf/rewriter_config.pb.h"
#include "tsl/platform/env.h"
#include "tsl/platform/errors.h"

namespace {

#define VALIDATE_STRUCT_SIZE(STRUCT_NAME, STRUCT_OBJ, SIZE_VALUE_NAME) \
  do {                                                                 \
    if (STRUCT_OBJ.struct_size == 0) {                                 \
      return absl::Status(absl::StatusCode::kFailedPrecondition,       \