layers in order to work, where the Keras symbolic shape inference proves
        insufficient.
    *   Code that tries manually walking a `tf.keras.Model` layer by layer and
        assumes layers only ever have one positional argument. This assumption
        doesn't hold true before TF 2.4 either, but is more likely to cause
        issues now.
    *   Code that manually enters `keras.backend.get_graph()` before building a

  // 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, "");

  return position;
}

static void Flush(const TF_WritableFile* file, TF_Status* status) {