Re: Bug #578: pg_dumpall from 7.1.3 can not be imported in 7.2.0

"Rainer Tammer" <tammer(at)tammer(dot)net> writes:
> On Mon, 11 Feb 2002 08:20:39 -0800 (PST), Stephan Szabo wrote:
>> I see a check constraint on queue which says q_short<>''::text
>> and the same on category. Did you rename q_short as either
>> q_name or q_group? I know there's a bug that the reported
>> name for the check constraint doesn't follow a alter table
>> rename column properly.

> Yes I did alter the following tables: queue, category

> shd=# \d queue
> Table "queue"
> Attribute | Type | Modifier
> -----------+---------+--------------------------------------------------
> q_id | integer | not null default nextval('queue_q_id_seq'::text)
> q_name | text | not null
> q_group | text | not null
> Indices: queue_pkey,
> queue_q_name_key
> Constraints: (q_short <> ''::text)
> (q_name <> ''::text)

Ah so. The problem here is that pg_relcheck.rcsrc doesn't track column
renaming. One wonders why we have the column at all :-(

> Now the question:
> How can I correct this error.

Either edit the dump file, or manually update the pg_relcheck.rcsrc
field for that constraint.

As far as fixing the problem in the long term goes, I wonder if we
should change pg_dump to use pg_get_expr(rcbin) in place of rcsrc.

I seem to recall a discussion a long time back about whether it was
better to store constraints in source or expression-tree form. This
shows one case where the expression-tree is more robust...

regards, tom lane