Обсуждение: OBJECT_ATTRIBUTE is useless (or: ALTER TYPE vs ALTER TABLE for composites)

I found that the OBJECT_ATTRIBUTE symbol is useless.  I can just remove
it and replace it with OBJECT_COLUMN, and everything continues to work;
no test fails that I can find.

I thought we had a prohibition against ALTER TABLE when used on
composites, but it's not as severe as I thought.  The following commands
fail in master:

ALTER TABLE comptype RENAME TO comptype2;    -- HINT: Use ALTER TYPE
ALTER TABLE comptype SET SCHEMA sch;        -- HINT: Use ALTER TYPE

However, the following command works in master:

ALTER TABLE comptype RENAME COLUMN a TO b;
and has the same effect as this:
ALTER TYPE comptype RENAME ATTRIBUTE a TO b;

The RENAME ATTTRIBUTE case in RenameStmt is the only thing currently
using OBJECT_ATTRIBUTE; therefore, since in precisely that case we do
not prohibit using ALTER TABLE, we can just remove OBJECT_ATTRIBUTE
completely.  That leads to the attached patch, which changes no test
result at all.

This symbol was added in

commit e440e12c562432a2a695b8054964fb34e3bd823e
Author: Peter Eisentraut <peter_e@gmx.net>
Date:   Sun Sep 26 14:41:03 2010 +0300

    Add ALTER TYPE ... ADD/DROP/ALTER/RENAME ATTRIBUTE

    Like with tables, this also requires allowing the existence of
    composite types with zero attributes.

    reviewed by KaiGai Kohei

Thoughts?

--
Álvaro Herrera                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Please see check_object_ownership(). It checks relation's ownership
if OBJECT_COLUMN, however, type's ownership is the correct check if
OBJECT_ATTRIBUTE.

--
NEC OSS Promotion Center / PG-Strom Project
KaiGai Kohei <kaigai@ak.jp.nec.com>


> -----Original Message-----
> From: pgsql-hackers-owner@postgresql.org
> [mailto:pgsql-hackers-owner@postgresql.org] On Behalf Of Alvaro Herrera
> Sent: Tuesday, February 24, 2015 4:02 AM
> To: Pg Hackers; Peter Eisentraut
> Subject: [HACKERS] OBJECT_ATTRIBUTE is useless (or: ALTER TYPE vs ALTER TABLE
> for composites)
> 
> I found that the OBJECT_ATTRIBUTE symbol is useless.  I can just remove
> it and replace it with OBJECT_COLUMN, and everything continues to work;
> no test fails that I can find.
> 
> I thought we had a prohibition against ALTER TABLE when used on
> composites, but it's not as severe as I thought.  The following commands
> fail in master:
> 
> ALTER TABLE comptype RENAME TO comptype2;    -- HINT: Use ALTER TYPE
> ALTER TABLE comptype SET SCHEMA sch;        -- HINT: Use ALTER TYPE
> 
> However, the following command works in master:
> 
> ALTER TABLE comptype RENAME COLUMN a TO b;
> and has the same effect as this:
> ALTER TYPE comptype RENAME ATTRIBUTE a TO b;
> 
> The RENAME ATTTRIBUTE case in RenameStmt is the only thing currently
> using OBJECT_ATTRIBUTE; therefore, since in precisely that case we do
> not prohibit using ALTER TABLE, we can just remove OBJECT_ATTRIBUTE
> completely.  That leads to the attached patch, which changes no test
> result at all.
> 
> This symbol was added in
> 
> commit e440e12c562432a2a695b8054964fb34e3bd823e
> Author: Peter Eisentraut <peter_e@gmx.net>
> Date:   Sun Sep 26 14:41:03 2010 +0300
> 
>     Add ALTER TYPE ... ADD/DROP/ALTER/RENAME ATTRIBUTE
> 
>     Like with tables, this also requires allowing the existence of
>     composite types with zero attributes.
> 
>     reviewed by KaiGai Kohei
> 
> Thoughts?
> 
> --
> Álvaro Herrera                http://www.2ndQuadrant.com/
> PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Kouhei Kaigai wrote:
> Please see check_object_ownership(). It checks relation's ownership
> if OBJECT_COLUMN, however, type's ownership is the correct check if
> OBJECT_ATTRIBUTE.

Hmm.  Is there any case where the two are different?

-- 
Álvaro Herrera                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

> Kouhei Kaigai wrote:
> > Please see check_object_ownership(). It checks relation's ownership
> > if OBJECT_COLUMN, however, type's ownership is the correct check if
> > OBJECT_ATTRIBUTE.
> 
> Hmm.  Is there any case where the two are different?
>
AlterObjectTypeCommandTag()?

OBJECT_ATTRIBUTE makes "ALTER TYPE" tag, but "ALTER COLUMN" tag is
made with OBJECT_COLUMN.

Above two cases are all I could found.

Thanks,
--
NEC OSS Promotion Center / PG-Strom Project
KaiGai Kohei <kaigai@ak.jp.nec.com>

On 2/23/15 2:01 PM, Alvaro Herrera wrote:
> I found that the OBJECT_ATTRIBUTE symbol is useless.  I can just remove
> it and replace it with OBJECT_COLUMN, and everything continues to work;
> no test fails that I can find.

It appears that it would change the command tag from ALTER TYPE to ALTER
TABLE.

Peter Eisentraut wrote:
> On 2/23/15 2:01 PM, Alvaro Herrera wrote:
> > I found that the OBJECT_ATTRIBUTE symbol is useless.  I can just remove
> > it and replace it with OBJECT_COLUMN, and everything continues to work;
> > no test fails that I can find.
> 
> It appears that it would change the command tag from ALTER TYPE to ALTER
> TABLE.

Ah, that it does.  I guess I unconsciously believed that command tags
would be part of regression tests expected output, but clearly they are
not.

I don't think this change is all that terrible, but I've seen past
discussions about changing command type and they aren't pretty, so I
will drop this.

-- 
Álvaro Herrera                http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services