Jaime Casanova <systemguards@xxxxxxxxx> writes: > that is a mis-conception... a transaction *must* be atomic (all or nothing)... > the reason some databases act that bad is because they don't support > savepoints, and because postgres does it doesn't need that > awfulness... Well it's not as bad as all that. It's still "atomic" in that an interruption cannot leave half of the transaction committed and half undone. In other words "all" is just "all of the actions that didn't produce an error". It's up to the client whether to commit the transaction after an error has occurred. It's great that Postgres follows the standard here, but don't go overboard on the criticism of other databases either. Where is Postgres at with psql using savepoints implicitly to wrap every client command btw? My single biggest pet peeve with Postgres is that setting autocommit off in psql is basically unusable because any typo forces you to start your transaction all over again. -- greg
