csv::CSVReader generally assumes that most rows in a CSV are of the same length. If your CSV has important data stored in rows which may not be the same length as the others, then you may want to create your own subclass of CSVReader and override bad_row_handler.

Examples

csv::CSVReader::bad_row_handler
csv::internals::CSVGuesser::Guesser::bad_row_handler()