If a function is deterministic for then it can be executed once and its result can be memoized for the rest of the statement (or transaction even).
If a function is idempotent for a statement then it can be executed once per-statement, and its result(s) can be memoized and reused for the life of that statement. If a function is neither idempotent (even if it is deterministic in the sense of not being random, for example, a monotonically increasing counter) nor deterministic (e.g., a true random()) then it should be called exactly as many times as the statement seems to imply, no more and no fewer (e.g., once per-row of a correlated sub-query, ...). Can idempotence and determinism be considered synonymous for the purposes of an RDBMS engine? I think not quite: deterministic should imply that a memoization cache can be used always, across many statements and even transactions, whereas idempotence might imply only that memoization is permitted (but not required) on a per-statement basis. Another desirable attribute might be whether the function is fast or slow: for fast enough deterministic functions there will be no point in having a large memoization cache, or even any memoization cache. Nico -- _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
