Re: Declarative partitioning - Mailing list pgsql-hackers

On 2016/07/04 21:31, Ashutosh Bapat wrote:
> Hi Amit,
> I observed that the ChangeVarNodes call at line 1229 in
> get_relation_constraints() (code below) changes the varno in the cached
> copy of partition key expression, which is undesirable. The reason for this
> seems to be that the RelationGetPartitionCheckQual() returns the copy of
> partition key expression directly from the cache. This is mostly because
> get_check_qual_for_range() directly works on cached copy of partition key
> expressions, which it should never.

Yes, a copyObject() on key->partexprs items seems necessary. Will fix that.

> 1223     /* Append partition check quals, if any */
> 1224     pcqual = RelationGetPartitionCheckQual(relation);
> 1225     if (pcqual)
> 1226     {
> 1227         /* Fix Vars to have the desired varno */
> 1228         if (varno != 1)
> 1229             ChangeVarNodes((Node *) pcqual, 1, varno, 0);
> 1230
> 1231         result = list_concat(result, pcqual);
> 1232     }
>
> Because of this, the first time through the partition key expressions are
> valid, but then onwards they are restamped with the varno of the first
> partition.
>
> Please add testcases to your patch to catch such types of issues.

I will integrate tests into the patch(es) and add some more.

Thanks,
Amit