The goal is to create a single version_repo object for each file loaded. In that way we may avoid loading the file for loading and writing.