ArrowRow

@contextmanager
def open(self):
    """Stream row contents from indexed file."""
    from pyarrow.csv import ParseOptions, ReadOptions
    from pyarrow.dataset import CsvFileFormat, dataset

    from datachain.lib.arrow import fix_pyarrow_format

    kwargs = dict(self.kwargs)
    if csv := kwargs.pop("csv", None):
        parse = ParseOptions(**csv["parse_options"])
        if csv.get("skip_invalid_rows"):
            parse.invalid_row_handler = lambda _row: "skip"
        kwargs["format"] = fix_pyarrow_format(
            CsvFileFormat(
                read_options=ReadOptions(**csv["read_options"]),
                parse_options=parse,
            ),
            parse,
        )

    if self.file._caching_enabled:
        self.file.ensure_cached()
        path = self.file.get_local_path()
        ds = dataset(path, **kwargs)
    else:
        path = self.file.get_fs_path()
        ds = dataset(path, filesystem=self.file.get_fs(), **kwargs)

    return ds.take([self.index]).to_reader()

def read(self):
    """Returns row contents as dict.

    Best effort, with no guarantee to match the stored row: it parses the
    file again from the start up to this row, and not every reader option is
    kept, e.g. column types.
    """
    with self.open() as record_batch:
        return record_batch.to_pylist()[0]