The LP-rounding plus greed approach for partial optimization revisited