The text stage currently implements the search "foo -bar" by querying the text index for "foo", fetching each document returned from the scan, using the text matcher to filter out documents containing "bar", and returning the resulting set.

Performance of searches using term negation would be improved greatly if the fetch/matcher steps are eliminated. For "foo -bar", the text stage should query the text index for "foo", then query the text index for "bar", compute the set difference, and then return the resulting set.