On 2016/06/22 18:14, Ashutosh Bapat wrote:
> I wonder whether such a whole-row-var would arise from the nullable side
>> of a join? I guess not. Not that I'm saying we shouldn't account for that
>> case at all since any and every whole-row-var in the targetlist currently
>> gets that treatment, even those that are known non-nullable. Couldn't we
>> have prevented the latter somehow? IOW, only generate the CASE WHEN when
>> a Var being deparsed is known nullable as the comment there says:
>>
>> deparse.c:
>>
>> 1639 /*
>> 1640 * In case the whole-row reference is under an outer join then it has
>> 1641 * to go NULL whenver the rest of the row goes NULL. Deparsing a join
>> 1642 * query would always involve multiple relations, thus qualify_col
>> 1643 * would be true.
>> 1644 */
>> 1645 if (qualify_col)
>> 1646 {
>> 1647 appendStringInfoString(buf, "CASE WHEN");
>> 1648 ADD_REL_QUALIFIER(buf, varno);
>> 1649 appendStringInfo(buf, "* IS NOT NULL THEN ");
>> 1650 }
>>
>> But I guess just fixing the expression as your patch does may be just fine.
>>
> I thought about that, but it means that we have compute the nullable relids
> (which isn't a big deal I guess). That's something more than necessary for
> fixing the bug, which is the focus in beta stage right now.
Agreed.
Thanks,
Amit