Method NextItem

tensorflow/core/grappler/inputs/file_input_yielder.cc:45–131

}

bool FileInputYielder::NextItem(GrapplerItem* item) {
  if (filenames_.size() == bad_inputs_) {
    // All the input files are bad, give up.
    return false;
  }

  if (current_file_ >= filenames_.size()) {
    if (current_iteration_ >= max_iterations_) {
      return false;
    } else {
      ++current_iteration_;
      current_file_ = 0;
      bad_inputs_ = 0;
    }
  }

  const string& filename = filenames_[current_file_];
  ++current_file_;

  if (!Env::Default()->FileExists(filename).ok()) {
    LOG(WARNING) << "Skipping non existent file " << filename;
    // Attempt to process the next item on the list
    bad_inputs_ += 1;
    return NextItem(item);
  }

  LOG(INFO) << "Loading model from " << filename;

  MetaGraphDef metagraph;
  Status s = ReadBinaryProto(Env::Default(), filename, &metagraph);
  if (!s.ok()) {
    s = ReadTextProto(Env::Default(), filename, &metagraph);
  }
  if (!s.ok()) {
    LOG(WARNING) << "Failed to read MetaGraphDef from " << filename << ": "
                 << s.ToString();
    // Attempt to process the next item on the list
    bad_inputs_ += 1;
    return NextItem(item);
  }

  if (metagraph.collection_def().count("train_op") == 0 ||
      !metagraph.collection_def().at("train_op").has_node_list() ||
      metagraph.collection_def().at("train_op").node_list().value_size() == 0) {
    LOG(ERROR) << "No train op specified";
    bad_inputs_ += 1;
    metagraph = MetaGraphDef();
    return NextItem(item);
  } else {
    std::unordered_set<string> train_ops;
    for (const string& val :
         metagraph.collection_def().at("train_op").node_list().value()) {
      train_ops.insert(NodeName(val));
    }
    std::unordered_set<string> train_ops_found;
    for (auto& node : metagraph.graph_def().node()) {
      if (train_ops.find(node.name()) != train_ops.end()) {
        train_ops_found.insert(node.name());