> > Yes, the only difference seems to be, that the changes need not
> > be sync'd to disk, and you only need one level of nesting as long
> > as the user is not presented the ability to use nested tx.
> >
>
> Hmm,what do you want now ?
I basically just wanted to say yes, but stated some differences that are
minor
and can be ignored.
>
> Note that (f)sync is irrelevant at all.
> Partial rollback is the problem of only the backend to be rollbacked
> except locking.
>
> Vadim has already planned savepoints functionality instead of nested
> tx. I have never heard objections to the proposal.
I think this is the same as nested tx, at least that is my understanding.
> I could see little difference between the implementation of rollback
> to arbitrary savepoints and the implemention of rollback only to the
> savepoint implicitly placed immediately before current statement.
>
> Do you want another hack ?
No.
Andreas