On Thu, 2011-08-04 at 20:55 -0400, Bruce Momjian wrote:
> It would act like COPY, meaning the table would have to be truncated or
> created in the same transaction.
Well, in that case it could work for any INSERT. No need for a SELECT to
be involved. For that matter, why not make it work for DELETE and
UPDATE, too?
However, I think this is all just a workaround for not having a faster
loading path. I don't object to applying this optimization to inserts,
but I think it might be more productive to figure out if we can support
loading data efficiently -- i.e. also set hint bits and frozenxid during
the load.
Regards,Jeff Davis