How to return different type when using @overlaod?

def hpat_pandas_series_getitem(self, idx):

    if not isinstance(self, SeriesType):
        return None

    def hpat_pandas_series_getitem_index_impl(self, idx):
        index = self._index
        mask = numpy.empty(len(self._data), numpy.bool_)
        for i in numba.prange(len(index)):
            mask[i] = index[i] == idx
        if mask.sum() == 1:
            return self._data[numpy.argmax(mask == True)]
            return pandas.Series(data=self._data[mask], index=index[mask], name=self._name)
    return hpat_pandas_series_getitem_index_impl

Here, I use sdc_overload (which call overload method internally), but I want to return two types, int or series, according to different runtime value. How to do this? Right now, Numba complains about “Can’t unify return type from the following types: int64, series(int64, array(int64, 1d, C), StringArrayType(), True)”. And I could not differ into two implmentations just according to arg types.