On Thu, Feb 13, 2020 at 10:30 AM Kyotaro Horiguchi
<horikyota.ntt@gmail.com> wrote:
> At Mon, 10 Feb 2020 14:32:44 +0900, Amit Langote <amitlangote09@gmail.com> wrote in
> > Agree that ObjectIsUserObject(oid) is easier to read than oid >=
> > FirstNormalObject. I would have not bothered, for example, if it was
> > something like oid >= FirstUserObjectId to begin with.
>
> Aside from the naming, I'm not sure it's sensible to use
> FirstNormalObjectId since I don't see a clear definition or required
> characteristics for "user created objects" is. If we did CREATE
> TABLE, FUNCTION or maybe any objects during single-user mode before
> the first object is created during normal multiuser operation, the
> "user-created(or not?)" object has an OID less than
> FirstNormalObjectId. If such objects are the "user created object", we
> need FirstUserObjectId defferent from FirstNormalObjectId.
Interesting observation. Connecting to database in --single mode,
whether done using initdb or directly, is always considered
"bootstrapping", so the OIDs from the bootstrapping range are
consumed.
$ postgres --single -D pgdata postgres
PostgreSQL stand-alone backend 13devel
backend> create table a (a int);
backend> select 'a'::regclass::oid;
1: oid (typeid = 26, len = 4, typmod = -1, byval = t)
----
1: oid = "14168" (typeid = 26, len = 4, typmod = -1, byval = t)
Here, FirstBootstrapObjectId < 14168 < FirstNormalObjectId
Maybe we could document that pg_is_user_object() and its internal
counterpart returns true only for objects that are created during
"normal" multi-user database operation.
Thanks,
Amit
