On 2016/12/27 22:24, Ashutosh Bapat wrote:
> On Mon, Dec 26, 2016 at 3:36 PM, Amit Langote
> <Langote_Amit_f8@lab.ntt.co.jp> wrote:
>> Attached patches modifies MergeAttributesIntoExisting() such that we
>> increment attinhcount not only for user attributes, but also for the oid
>> system column if one exists.
>>
>> Thoughts?
>
> Yes, if a child inherits from a parent with OID, child gets OID column
> and its inhcount is set to 1.
[ ... ]
> So, I think your patch is on the right track.
>
> We allow creating user attribute with name "oid" so you do not want to
> search system attribute oid by name. Instead search by attribute id
> ObjectIdAttributeNumber.
Good point. Although, there can only be one of the two in a table at any
given time - either the "oid" system column or user-defined column named
"oid". I was afraid that with the patch, the attinhcount of a
user-defined oid column may get incremented twice, but before that could
happen, one would get the error: "child table without OIDs cannot
inherited from table with OIDs"
create table foo (a int) with oids;
create table bar (a int, oid int);
alter table bar inherit foo;
ERROR: table "bar" without OIDs cannot inherit from table "foo" with OIDs
And then:
alter table bar set with oids;
ERROR: column "oid" of relation "bar" already exists
Thanks,
Amit