On Fri, Sep 14, 2012 at 5:58 AM, Igor Tandetnik <itandet...@mvps.org> wrote: > Pavel Ivanov <paiva...@gmail.com> wrote: >> There's another way how it could make sense - if you read that excerpt >> in context. ;-) The "statement-level read consistency" definition >> cited by Petite is the transaction isolation level. I.e. it defines >> what each statement executed in one transaction can see related to the >> changes done in another transaction. > > Well, in this case, if you only ever execute one statement per transaction in > SQLite, then it does support "statement-level read consistency." Though I > fail to see how it materially differs from "serializable". Am I missing > something obvious?
For one statement per transaction "statement-level consistency" and "serializable" do not differ, you are right. But IIRC Oracle doesn't have an auto-commit mode of operation and multi-statement transactions are more common, especially when we talk about stored procedures, triggers and things like that. And if you ever call "begin transaction" or execute multi-statement transactions then serializable will make sure that your transaction sees only changes committed before "begin", in "statement-level" you can see changes committed after "begin". The worst problem of this: with "statement-level consistency" two identical select queries executed inside the same transaction can return different results. >> This isolation level doesn't >> define visibility of changes done in the same transaction. These >> visibility rules are defined in different place in the documentation >> and don't depend on transaction isolation level. >> >> So returning to my example "statement-level read consistency" means >> that cursor can see myfield=2 if update is done in another transaction >> before cursor is opened but after begin is executed. > > Wait a minute. If "statement-level read consistency" only applies to two > queries in different transactions, what bearing, if any, does it have on your > example of "open, fetch, update, fetch"? There, everything happens in the > same transaction. I'm thoroughly confused. I'm sorry, I made that example before I carefully re-read Petite's email and understood what you are concerned about. Plus I was taking your example as a base for mine. :) But for statements inside one transaction Oracle uses the same rules as in statement-level consistency too. Although it's probably named a little bit differently and its definition doesn't include "committed data" (it would be "finished statements" or something like that). Pavel _______________________________________________ sqlite-users mailing list sqlite-users@sqlite.org http://sqlite.org:8080/cgi-bin/mailman/listinfo/sqlite-users
