Обсуждение: Set all variable-length fields of pg_attribute to null on column drop

I noticed that when a column is dropped, RemoveAttributeById() clears 
out certain fields in pg_attribute, but it leaves the variable-length 
fields at the end (attacl, attoptions, and attfdwoptions) unchanged. 
This is probably harmless, but it seems wasteful and unclean, and leaves 
potentially dangling data lying around (for example, attacl could 
contain references to users that are later also dropped).

I suggest the attached patch to set those fields to null when a column 
is marked as dropped.

On Thu, Nov 30, 2023 at 6:24 AM Peter Eisentraut <peter@eisentraut.org> wrote:
> I noticed that when a column is dropped, RemoveAttributeById() clears
> out certain fields in pg_attribute, but it leaves the variable-length
> fields at the end (attacl, attoptions, and attfdwoptions) unchanged.
> This is probably harmless, but it seems wasteful and unclean, and leaves
> potentially dangling data lying around (for example, attacl could
> contain references to users that are later also dropped).
>
> I suggest the attached patch to set those fields to null when a column
> is marked as dropped.

I haven't reviewed the patch, but +1 for the idea.

--
Robert Haas
EDB: http://www.enterprisedb.com

On 2023-Nov-30, Peter Eisentraut wrote:

> I noticed that when a column is dropped, RemoveAttributeById() clears out
> certain fields in pg_attribute, but it leaves the variable-length fields at
> the end (attacl, attoptions, and attfdwoptions) unchanged. This is probably
> harmless, but it seems wasteful and unclean, and leaves potentially dangling
> data lying around (for example, attacl could contain references to users
> that are later also dropped).

Yeah, this looks like an ancient oversight -- when DROP COLUMN was added
we didn't have any varlena fields in this catalog, and when the first
one was added (attacl in commit 3cb5d6580a33) resetting it on DROP
COLUMN was overlooked.

LGTM.

-- 
Álvaro Herrera         PostgreSQL Developer  —  https://www.EnterpriseDB.com/
"No hay ausente sin culpa ni presente sin disculpa" (Prov. francés)

On 22.12.23 10:05, Alvaro Herrera wrote:
> On 2023-Nov-30, Peter Eisentraut wrote:
> 
>> I noticed that when a column is dropped, RemoveAttributeById() clears out
>> certain fields in pg_attribute, but it leaves the variable-length fields at
>> the end (attacl, attoptions, and attfdwoptions) unchanged. This is probably
>> harmless, but it seems wasteful and unclean, and leaves potentially dangling
>> data lying around (for example, attacl could contain references to users
>> that are later also dropped).
> 
> Yeah, this looks like an ancient oversight -- when DROP COLUMN was added
> we didn't have any varlena fields in this catalog, and when the first
> one was added (attacl in commit 3cb5d6580a33) resetting it on DROP
> COLUMN was overlooked.
> 
> LGTM.

committed