Re:

2013/7/15 Luca Ferrari <fluca1978(at)infinito(dot)it>:
> On Sun, Jul 14, 2013 at 8:36 PM, Vincenzo Romano
> <vincenzo(dot)romano(at)notorand(dot)it> wrote:
>
>> I am only concerned about how late is done the binding between a table
>> name and the actual OID for other functions, views and triggers.
>
>
> Well, it should work like this: once the parser decides that a query
> looks good, it seaches for the catalogue to find out all the names of
> implied relations. Therefore, the binding you mention should happen as
> late as possible. However, I'm not sure if there a kind of caching
> mechanism that can invalidate such sentence. Someone with more
> experience can detail better.
>
> Luca

That's exactly my fear in doing that transactional DDL.
And it's not just the table names, but also the structure, the
indexes, the statistics and so on.
Is there anyone on this ML who "owns" the truth? :-)

The alternative is to do things the "good ol' way" by DELETING+INSERTING
(http://tapoueh.org/blog/2013/07/05-archiving-data-fast.html)
Where I'd fear for longer LOCKs.