On Wed, May 1, 2024 at 12:39 AM Paul Jungwirth
<pj@illuminatedcomputing.com> wrote:
>
> On 4/30/24 09:24, Robert Haas wrote:
> > Peter, could you have a look at
> > http://postgr.es/m/47550967-260b-4180-9791-b224859fe63e@illuminatedcomputing.com
> > and express an opinion about whether each of those proposals are (a)
> > good or bad ideas and (b) whether they need to be fixed for the
> > current release?
>
> Here are the same patches but rebased. I've added a fourth which is my progress on adding the CHECK
> constraint. I don't really consider it finished though, because it has these problems:
>
> - The CHECK constraint should be marked as an internal dependency of the PK, so that you can't drop
> it, and it gets dropped when you drop the PK. I don't see a good way to tie the two together though,
> so I'd appreciate any advice there. They are separate AlterTableCmds, so how do I get the
> ObjectAddress of both constraints at the same time? I wanted to store the PK's ObjectAddress on the
> Constraint node, but since ObjectAddress isn't a Node it doesn't work.
>
hi.
I hope I understand the problem correctly.
my understanding is that we are trying to solve a corner case:
create table t(a int4range, b int4range, primary key(a, b WITHOUT OVERLAPS));
insert into t values ('[1,2]','empty'), ('[1,2]','empty');
I think the entry point is ATAddCheckNNConstraint and index_create.
in a chain of DDL commands, you cannot be sure which one
(primary key constraint or check constraint) is being created first,
you just want to make sure that after both constraints are created,
then add a dependency between primary key and check constraint.
so you need to validate at different functions
(ATAddCheckNNConstraint, index_create)
that these two constraints are indeed created,
only after that we have a dependency linking these two constraints.
I've attached a patch trying to solve this problem.
the patch is not totally polished, but works as expected, and also has
lots of comments.