Considering that INTERSECT is logically nothing but a special case of relational join (the exact opposite of cartesian product), where all columns are involved in the join condition, you should just be able to reuse any optimizations that exist for join, including primary/unique keys/etc. -- Darren Duncan
> Documentation says that INTERSECT implemented with temporary tables either > in memory or on disk. Is it always the case? The problem is that if I have > several selects (six for example) when each produces thousands of results > and the intersection is only hundreds the query takes about minute to > execute. But all these selects are properly ordered to do intersection > without extra files usage. For example I know that I can make it in my > code > by splitting these selects into separated prepared statements and making > synchronous stepping outputting only if all steps have equal value(s). If > there's no such optimization, is it possible to implement it in the sqlite > engine? I know that having very complex query with many selects it would > be > a hard task to recognize such a specific case but maybe it is easier than > it > seems. _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
