On Fri, Jan 25, 2013 at 11:28:58PM -0500, Bruce Momjian wrote:
> On Fri, Jan 25, 2013 at 11:08:56PM -0500, Tom Lane wrote:
> > Bruce Momjian <bruce@momjian.us> writes:
> > > ! ereport(ERROR,
> > > ! (ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE,
> > > ! errmsg("cannot perform FREEZE because of previous table activity in the current
transaction")));
> >
> > [ itch... ] What is "table activity"? I always thought of tables as
> > being rather passive objects. And anyway, isn't this backwards? What
> > we're complaining of is *lack* of activity. I don't see why this isn't
> > using the same message as the other code path, namely
>
> Well, here is an example of this message:
>
> BEGIN;
> TRUNCATE vistest;
> SAVEPOINT s1;
> COPY vistest FROM stdin CSV FREEZE;
> ERROR: cannot perform FREEZE because of previous table activity in the current transaction
> COMMIT;
>
> Clearly it was truncated in the same transaction, but the savepoint
> somehow invalidates the freeze. There is a C comment about it:
The savepoint prevents the COPY FREEZE, because COPY FREEZE needs the table to
have been created or truncated in the current *sub*transaction. Issuing
"RELEASE s1" before the COPY makes it work again, for example.
>
> * BEGIN;
> * TRUNCATE t;
> * SAVEPOINT save;
> * TRUNCATE t;
> * ROLLBACK TO save;
> * COPY ...
This is different. The table was truncated in the current subtransaction, and
it's safe in principle to apply the optimization. Due to an implementation
artifact, we'll reject it anyway.