Re: [GENERAL] CREATE TABLE LIKE and SERIAL

Le jeudi 12 novembre 2009 à 10:12:00, Dave Page a écrit :
> On Thu, Nov 12, 2009 at 8:00 AM, Guillaume Lelarge
>
> <guillaume(at)lelarge(dot)info> wrote:
> > Le dimanche 1 novembre 2009 à 13:24:10, Dave Page a écrit :
> >> On Fri, Oct 30, 2009 at 11:12 PM, Guillaume Lelarge
> >>
> >> <guillaume(at)lelarge(dot)info> wrote:
> >> > Le vendredi 30 octobre 2009 à 23:48:45, Tom Lane a écrit :
> >> >> Thom Brown <thombrown(at)gmail(dot)com> writes:
> >> >> > How is pgAdmin determining the serial type in this case?
> >> >>
> >> >> Most likely it's looking for the pg_depend entry that shows the
> >> >> sequence as being "owned by" the column. However, that's an
> >> >> oversimplification of reality. I would imagine that pgAdmin will lie
> >> >> to you in exactly the same cases that used to break pg_dump (notably,
> >> >> where someone has manually adjusted either the default expression or
> >> >> the sequence...)
> >> >
> >> > I've just read this. I wasn't aware of this. Should we suppress this
> >> > behavior of pgAdmin?
> >>
> >> Do you have a proposed fix? I suspect people won't want us to stop
> >> showing columns as 'SERIAL' when they actually were created that way.
> >
> > Other ideas than doing what pg_dump does? I'm afraid not.
>
> pg_dump isn't designed to make the output look pretty, which is the
> major issue here. We're trying to do both.
>

Yes.

> It seems to me that correctness trumps aesthetics, so we probably will
> have to do what pg_dump does.

That's the main reason I started this thread :)

> Care to work up a patch?
>

Yep. On my TODO list right now.

--
Guillaume.
http://www.postgresqlfr.org
http://dalibo.com