The package reads and writes data (like 'csv', 'tsv' and 'fwf') quickly. When reading it uses a quick initial indexing step, then reads the values lazily, so only the data you actually use needs to be read. The writer formats the data in parallel and writes to disk asynchronously from formatting.