Re: [BUG]Update Toast data failure in logical replication

On Mon, Feb 7, 2022 at 1:27 PM Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
>
> On Mon, Feb 7, 2022 at 12:25 PM Amit Kapila <amit(dot)kapila16(at)gmail(dot)com> wrote:
> >
> > Attached please find the modified patches.
>
> I have looked into the latest modification and back branch patches and
> they look fine to me.
>

Today, while looking at this patch again, I think I see one problem
with the below change (referring pg10 patch):
+ if (attrnum < 0)
+ {
+ if (attrnum != ObjectIdAttributeNumber &&
+ attrnum != TableOidAttributeNumber)
+ {
+ modified = bms_add_member(modified,
+ attrnum -
+ FirstLowInvalidHeapAttributeNumber);
+ continue;
+ }
+ }
...
...
+ /* No need to check attributes that can't be stored externally. */
+ if (isnull1 || TupleDescAttr(tupdesc, attrnum - 1)->attlen != -1)
+ continue;

I think it is possible that we use TupleDescAttr for system attribute
(in this case ObjectIdAttributeNumber/TableOidAttributeNumber) which
will be wrong as it contains only user attributes, not system
attributes. See comments atop TupleDescData.

I think this check should be modified to if (attrnum < 0 || isnull1
|| TupleDescAttr(tupdesc, attrnum - 1)->attlen != -1). What do you
think?

* Another minor comment:
+ if (!heap_attr_equals(RelationGetDescr(relation), attrnum, value1,
+ value2, isnull1, isnull2))

I think here we can directly use tupdesc instead of RelationGetDescr(relation).

--
With Regards,
Amit Kapila.