Re: CheckAttributeType() forgot to recurse into multiranges

From: Chao Li <li(dot)evan(dot)chao(at)gmail(dot)com>
To: Heikki Linnakangas <hlinnaka(at)iki(dot)fi>
Cc: "pgsql-hackers(at)postgresql(dot)org" <pgsql-hackers(at)postgresql(dot)org>
Subject: Re: CheckAttributeType() forgot to recurse into multiranges
Date: 2026-04-23 03:24:26
Message-ID: 0B5FC478-013D-45D1-876A-AD3FE0C21E7C@gmail.com
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

> On Apr 23, 2026, at 04:56, Heikki Linnakangas <hlinnaka(at)iki(dot)fi> wrote:
>
> Happened to spot this little bug:
>
> create type two_ints as (a int, b int);
> create type two_ints_range as range (subtype = two_ints);
>
> -- CheckAttributeType() forbids this:
> alter type two_ints add attribute c two_ints_range;
> ERROR: composite type two_ints cannot be made a member of itself
>
> -- But the same with a multirange is allowed:
> alter type two_ints add attribute c two_ints_multirange;
> ALTER TYPE
>
> That looks like a straightforward oversight in CheckAttributeType(). When multiranges were introduced, it didn't get the memo.
>
> Fix attached. Assuming no objections, I'll commit and backpatch that.
>
> While working on the fix, I noticed that in case of dropped columns, CheckAttributeType() is called with InvalidOid. It tolerates that, but it seems accidental and it performs a bunch of futile syscache lookups with InvalidOid, so it would be better to not do that. The second patch fixes that.
>
> - Heikki
> <0001-Don-t-allow-composite-type-to-be-member-of-itself-vi.patch><0002-Don-t-call-CheckAttributeType-with-InvalidOid-on-dro.patch>

I traced this patch set, 0002 looks good, but I have a suspicion about 0001.

```
+ else if (att_typtype == TYPTYPE_MULTIRANGE)
+ {
+ /*
+ * If it's a multirange, recurse to check its plain range type.
+ */
+ CheckAttributeType(attname, get_multirange_range(atttypid),
+ get_range_collation(atttypid),
+ containing_rowtypes,
+ flags);
+ }
```

Looking at get_range_collation(), it only searches for RANGETYPE, so get_range_collation(atttypid) here will always return InvalidOid. This does not seem to cause a problem, because CheckAttributeType() will recurse into the TYPTYPE_RANGE path, and the collation will be evaluated there.

But to make the logic clearer, I think we could just pass InvalidOid as the collation OID in the TYPTYPE_MULTIRANGE case. If we really want to pass the actual collation OID here, I think it would need to be done more like this:

```
else if (att_typtype == TYPTYPE_MULTIRANGE)
{
Oid multirange_range_typid = get_multirange_range(atttypid);
Oid collation = get_range_collation(multirange_range_typid);
/*
* If it's a multirange, recurse to check its plain range type.
*/
CheckAttributeType(attname, multirange_range_typid,
collation,
containing_rowtypes,
flags);
}
```

Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/

In response to

Browse pgsql-hackers by date

  From Date Subject
Next Message John Naylor 2026-04-23 03:27:15 Re: GUC parameter ACLs and physical walsender
Previous Message David Rowley 2026-04-23 03:11:27 Re: The bogus calls in remove_self_join_rel()